Survey on Terahertz Nanocommunication and Networking: A Top-Down Perspective

1 Surv e y on T erahertz Nanocommunication and Networking: A T op-Do wn Perspecti v e Filip Lemic ∗ , Sergi Abadal † , W outer T a vernier ‡ , Pieter Stroobant ‡ , Didier Colle ‡ , Eduard Alarc ´ on † , Johann Marquez-Barja ∗ , Jeroen Famae y ∗ ∗ Internet T echnology and Data Science Lab (IDLab), Univ ersiteit Antwerpen - imec, Belgium † NaNoNetworking Center in Catalun ya (N3Cat), Univ ersitat Polit ` ecnica de Catalunya, Spain ‡ Internet T echnology and Data Science Lab (IDLab), Ghent Univ ersity - imec, Belgium Email: ﬁlip.lemic@uantwerpen.be Abstract —Recent developments in nanotechnology herald nanometer -sized devices expected to bring light to a number of groundbr eaking applications. Communication with and among nanodevices will be needed for unlocking the full potential of such applications. As the traditional communication approaches cannot be directly applied in nanocommunication, several alter- native paradigms have emerged. Among them, electromagnetic nanocommunication in the terahertz (THz) frequency band is particularly promising, mainly due to the br eakthrough of no vel materials such as graphene. For this reason, numerous research efforts are nowadays targeting THz band nanocommunication and consequently nanonetworking . As it is expected that these trends will continue in the future, we see it beneﬁcial to summarize the current status in these research domains. In this survey , we therefor e aim to provide an over view of the current THz nanocommunication and nanonetworking resear ch. Specif- ically , we discuss the applications en visioned to be supported by nanonetworks operating in the THz band, together with the requir ements that such applications pose on the underlying nanonetworks. Subsequently , we provide an overview of the current contrib utions on the different layers of the protocol stack, as well as the av ailable channel models and experimentation tools. As the ﬁnal contrib ution, we identify a number of open resear ch challenges and outline several potential future resear ch directions. Index T erms —Nanotechnology , electr omagnetic, terahertz, nanocommunication, nanonetworking, protocols, channel models, experimentation tools. I . I N T RO D U C T I O N “There’ s Plenty of Room at the Bottom: An Invitation to Enter a New Field of Physics” [1] was the title of a visionary lecture gi ven by the Nobel Prize recipient Prof. Richard Feynman at the annual American Physical Society meeting at the California Institute of T echnology (Caltech) in December 1959. Prof. Feynman discussed the possibility of directly manipulating materials on an atomic scale - and he surely wasn’t joking. The concepts originally outlined by Prof. Feynman later became circumscribed under the term “nanotechnology”, ﬁrst introduced by Prof. Norio T aniguchi from T okyo Uni versity of Science in 1974. Due to substantial research efforts in recent years, nanotechnology is today paving the way toward sub- µ m scale de vices (i.e., from one to a few hundred nanometers). Controlling materials on such a scale is expected to give rise to integrated nanodevices with simple sensing, actuation, data processing and storage, and communication capabilities, opening the horizon to a large variety of novel, ev en groundbreaking applications. Communication and coordination among the nanodevices, as well as between them and the macro-scale world, will be required to achie v e the full promise of such applications. Thus, se veral alternative nanocommunication paradigms have emerged in the recent years, the most promising ones being electromagnetic, acoustic, mechanical, and molecular commu- nication [2]. In molecular nanocommunication, a transmitting device releases molecules into a propagation medium, with the molecules being used as the information carriers [3]. Acoustic nanocommunication utilizes pressure variations in the (ﬂuid or solid) medium to carry information between the transmit- ter and recei ver . In mechanical (i.e., touch-based) nanocom- munication, nanorobots are used as carriers for information exchange [4]. Finally , electromagnetic nanocommunication uses the properties of electromagnetic waves (e.g., amplitude, phase, delay) as the information carriers [5]. Among the abov e paradigms, molecular and electromagnetic nanocommunica- tion sho w the greatest promise for enabling communication be- tween nanodevices [6]. This work focuses on electr omagnetic nanocommunication . This is due to the fact that it is among the two most promising nanocommunication paradigms, as well as due to its suitability to dif ferent propagation mediums (e.g., in-body , free space, on-chip) and potential for meeting the applications’ requirements (more details in Section III). Classical electromagnetic communication and networking paradigms are not directly applicable to the majority of nanoscale communication and networking scenarios, predom- inantly due to the small sizes and limited capabilities of nan- odevices [7], [8]. That it to say , microwa ve or even millimeter- wa ve (mmW a ve) frequencies impose relativ ely large antennas (i.e., mm-scale and larger) that do not ﬁt into nanodevices. T o meet the size requirements of nanodevices, a classical metallic antenna would be required to use very high radiation frequencies. For example, a one-micrometer -long dipole an- tenna w ould resonate at approximately 150 terahertz (THz) [9]. Although the communication bandwidth increases with the in- crease in the antenna’ s resonance frequency , the same happens with the propagation loss. Due to the highly constrained power of nanodevices [10], the feasibility of nanonetworks would be compromised if this approach would be adopted. In addition, nanoscale material properties of common metals are unknown, hence the common assumptions from antenna theory might not be correct [11]. Finally , it is currently also technologically 2 infeasible to develop miniature transceiver that could operate at such high frequencies [9]. As an alternativ e approach, graphene has attracted sub- stantial research attention, mainly due to its unique electrical and optical properties [12]. The interaction of electromagnetic radiation with graphene and its deri v ativ es (i.e., carbon nan- otubes (CNT) [13] - rolled graphene and graphene nanoribbons (GNRs) - thin strips of graphene), differs from that of the con ventional metals. Speciﬁcally , it has been demonstrated that graphene supports the propagation of Surface Plasmon Polari- ton (SPP) wa ves in the THz frequenc y band [14]–[16]. Due to that, graphene-enabled electromagnetic nanocommunication in the THz frequencies is able to deli ver miniaturization [17], [18], i.e., a prerequisite for nanocommunication. In addition, the possibility to operate at lower frequencies relax es the energy and power requirements for the nanodevices [9], [17]. Moreov er , graphene shows unique tunability properties, which allow to steer the beam or tune the resonance frequenc y by just changing a bias voltage [18], [19]. For these reasons and gi ven its experimentally demonstrated compatibility with CMOS manufacturing processes [20]–[22], graphene allows versatile antennas to be directly integrated in nanocommuni- cation devices as envisaged in multiple works [5], [19], [23]. Consequently , graphene-enabled THz nanocommunication at- tracted substantial attention from a broad scientiﬁc community . As the efforts targeting THz band nanocommunication yielded highly encouraging results, the research focus soon spread from the communication to the networking community , giving birth to nanonetworking in the THz band. Sufﬁce to say , the results of these efforts are equally encouraging and promising, at this point arguably also abundant. Hence, we believ e a summary of the research efforts and current State- of-the-Art (SotA) on THz nanocommunication and nanonet- working w ould be beneﬁcial to the community , which provides the main motiv ation for this surve y . In this surve y , we ﬁrst discuss several application domains that could be enabled by nanonetworks operating in the THz band. These include software-deﬁned metamaterials, wireless robotic materials, body-centric communication, and Wireless Networks-on-Chip (WNoCs). Moreover , we deriv e a set rule- of-thumb requirements that each of the application domains posits on the supporting nanonetworks. Then, we utilize a top-down approach in discussing the SotA of different layers of the nanonetworks’ protocol stack. W e belie ve a top-do wn approach to be more natural than the bottom-up alternativ e, as it provides the reader with a straightforward mapping between the application requirements on the one side, and the underlying protocols and their design goals on the other . In addition, we provide an ov erview of existing channel models and simulation and experimentation tools a vailable for THz nanocommunication and nanonetworking research. At the end of each section, we discuss the corresponding research “gaps” and open challenges. Moreov er , we summarize several additional open challenges related to THz nanocommunication and nanonetworking in general, but not directly pertaining to speciﬁc layers of the protocol stack, channel models, or experimentation tools. W e aim to provide a straightforward introduction to THz nanocommunication and nanonetworking research, as well at as to identify the “missing pieces” in the current results and suggest potential future research directions. A. Methodology This surve y has been de veloped follo wing the Systematic Literature Re vie w (SLR) methodology [24], [25]. W e aimed to select all the relev ant works in the context of THz nanocom- munication and nanonetworking, in turn resulting in the de- tection of the open challenges and missing “pieces” in the existing research literature. W e hav e considered the following inclusion criteria: i) an article proposes a full novel solution rather than conceptual, informativ e, or ongoing efforts (i.e., position, poster , demo, and W ork-in-Progress (W iP) articles were not considered), ii) an article went through a peer -re view process and is either publicly av ailable (e.g., in ArXiv) or included in the W eb of Science (W oS), Scopus, IEEEXplore, A CM, and/or Springer databases, and iii) an article is written in English. In terms of the step-by-step methodology , we hav e used the Google Scholar database in the initial revie w phase, during which the articles were included or excluded based on their titles and abstracts. In the consequent step, the full texts were assessed along the above-mentioned inclusion criteria. As a commonly accepted practice, we hav e placed extra emphasis on articles published since 2015, highly cited ones, and the ones published in the most important venues, e.g., IEEE T ransactions on Nanotechnology , ELSEVIER’ s Nano Communication Netw orks, IEEE Nanotechnology Magazine, and ACM International Conference on Nanoscale Computing and Communication (A CM N ANOCOM). B. Structur e The rest of this paper is structured as follows. In Section II, we provide an overvie w of the related surveys in the existing literature. The application domains that could potentially be supported by nanonetworks operating in the THz band are, together with the requirements such applications pose on the underlying nanonetworks, discussed in Section III. The current research efforts on the network, link, and physical layers of the protocol stack are summarized in Sections IV, V, and VI, respectively . Moreov er , in Section VII we discuss the av ailable channel models for THz nanocommunication, while the existing experimentation and simulation tools are outlined in Section VIII. In Section IX, we discuss several more general open challenges, i.e., the ones not directly related to the content of the previous sections, yet still relev ant to THz nanocommunication and nanonetw orking research. Finally , we conclude the survey in Section X. I I . R E L A T E D W O R K There are several contributions generically targeting THz band communication [26]–[28]. They provide useful insights into the paradigm from the device’ s perspectiv e, as well as from the communication point of view . In terms of communi- cation, adv ances and challenges pertain to channel modeling, dev elopment of communication protocols, and establishment of supporting experimentation tools and facilities. 3 There are also sev eral specialized surve y papers that discuss different aspects of THz band nanocommunication. The pio- neering work tar geting this topic was published by Akyildiz and Jornet [5], providing an in-depth view on nanotechnology and discussing se veral options for enabling communication among nanonodes. In [5], the authors take the nanodevice standpoint and ﬁrst outline the en visioned architecture of the nanode vice, consisting of units for sensing, actuating, powering, data processing and storage, and communication. The authors follow by outlining several application groups for W ireless Nano-Sensor Networks (WNSNs): biomedical, en vironmental, industrial and consumer goods, and military and defense. Moreo ver , the y brieﬂy discuss the architecture of a WNSN consisting of nanonodes, nanorouters, nano-micro interfaces, and gate ways. In addition, sev eral open research questions and promising directions for future research are outlined. It is worth mentioning that the survey is dated to 2010. It is also worth noting the work in [8], which provides an ov erview of molecular and electromagnetic nanocommunica- tion. In addition, [36] brieﬂy discusses electromagnetic com- munication from a stand-point of future healthcare systems, with the main focus of the work being big data analytics for healthcare data. In terms of electromagnetic nanocommunica- tion utilizing THz frequencies, [8] provides similar insights as [5]. Moreo ver , in [35], the authors provide a comparati ve surve y of different Media Access Control (MA C) protocols for WNSN. In [37], the authors provide a survey of MA C protocols for THz band communication in general. In addition, performance analyses of these protocols have been carried out in terms of consumed energy , transmission distance, and probability of collisions. Sev eral contrib utions tar get dif ferent aspects of the IoNT, most notably [29]–[34]. In these works, the authors tend to agree on the IoNT architecture, consisting of a nanonetw ork connected to the macro-world through a nano-macro gatew ay . In addition, sev eral application domains hav e been outlined in these works, which can roughly be grouped into health-care, en vironmental and agricultural monitoring, multimedia, and military and defense. Moreov er , the works conv erge tow ard the idea that there are two encouraging nanocommunication options, namely molecular and electromagnetic utilizing THz frequencies. The former is predominantly feasible for in-body nanocommunication. Among the abov e-mentioned contribu- tions targeting IoNT , arguably the most relev ant ones for this surve y are [34] and [30]. The w ork in [34] is interesting as it summarizes se veral requirements for nanocommunication for enabling in-body health-care applications. These pertain to legal (e.g., le gislation on duration of the body contact, in v a- siv eness or implantability of nanodevices), functional (i.e., the purpose of communication between nanonetwork and macro- world), and technical (e.g., reliability , safety , priv acy , real-time capabilities) requirements. The w ork is also rele v ant because it discusses several challenges for in-body nanocommunication, as well as useful simulation tools for the problem at hand. Jornet et al. [30] provide a relev ant read due to the fact that it explicitly outlines nanocommunication challenges in the multimedia-focused IoNT. These challenges pertain to data compression and signal processing, THz channel modeling, and challenges on different layers of the protocol stack. In order to position our work in the context of the above- outlined contributions (summarized in T able I), it is ﬁrst relev ant to note that these contributions, apart from [36], [37], date from 2016 or before. Sev eral nov el proposals hav e been made on v arious aspects of THz nanocommunication and nanonetworking in the recent years, ranging from no vel protocols in different layers, to THz channel modeling and experimentation tools. W e belie ve that at this point in time a critical summary of these contributions and their position- ing in regard to the older ones would beneﬁt the scientiﬁc community . Second, some of the current surveys focus either on THz communication in general or on nanocommunication, resulting in neglecting numerous aspects relev ant to THz band electromagnetic nanocommunication. Adversely , other contributions are substantially more focused, targeting for example only the IoNT or MA C layer protocols for THz band (nano)communication. Third, due to its very recent roll- out with the most important works being only a few years old, THz band nanonetworking has not received suf ﬁcient treatment in the existing surveys. W e aim at ﬁlling the abov e- stated gaps by providing a full view on the problem, ranging from applications and their requirements, dif ferent layers of the protocols stack, channel models, and experimentation tools, to challenges and open research questions, but pertain- ing exclusi vely to electromagnetic nanocommunication and nanonetworking in the THz band. Finally , it is worth emphasizing that this work does not cov er THz macro-scale applications, protocols, nor channel models. Interested readers are directed to se v eral recent papers surve ying these topics, e.g., [38]–[42]. In addition, this surv ey deals with electromagnetic nanocommunication only . Hence, molecular and other potential paradigms for nanocommunica- tion and networking are out of scope of this paper . Interested readers can ﬁnd more details in recent surv eys from the litera- ture [43]–[47]. Similarly , THz nanocommunication hardware- related aspects are out of scope of this work. Although arguably not abundant, existing survey papers discuss aspects such as THz sources [48], [49], graphene antennas [19], [50], and circuits/devices [51], [52]. I I I . A P P L I C A T I O N D O M A I N S E NA B L E D B Y T H Z N A N O N E T W O R K S In this section, we provide an overvie w of the most promi- nent application domains with the potential of being enabled by THz-operating nanonetworks. For each of the domains, we discuss its requirements, which are at the end of the section summarized in T able II. Note that in the table we include the most stringent requirements for each application domain, although these can potentially be more lenient for some applications in the domain. Also note that we differ - entiate the application domains based on their requirements. For example, we distinguish wireless robotic materials and body-centric communication, although both domains contain a v ariety of sensing-only applications. Due to that, some works from the literature (e.g., [32], [53]) specify WNSNs or IoNT 4 T ABLE I: Surve ys on THz band communication and networking Name Y ear T ype Summary of cover ed topics Akyildiz et al. [29] 2010 Internet of Nano-Things SotA in THz nanocommunication for the IoNT; research challenges (channel modeling, information encoding, and protocols for the IoNT). Akyildiz et al. [5] 2010 THz nanocommunication SotA in nanodevice technology; summaries of WNSN applications and architec- tures; overvie w of nanocommunication and networking challenges. Jornet et al. [30] 2012 Internet of Nano-Things SotA, open challenges, and research directions in the THz nanocommunication and Internet of Multimedia Nano-Things (IoMNT). Rikhtegar et al. [8] 2013 THz nanocommunication Molecular and electromagnetic (THz) communication for nanoscale applications; summary of nanoscale communication paradigms and potential applications. Balasubramaniam et al. [31] 2013 Internet of Nano-Things Challenges in realizing the IoNT (data collection and routing, bridging the outside world and nanonetworks, etc.); possible IoNT applications. Akyildiz et al. [27] 2014 THz communication THz band applications at macro and nanoscale; SotA in THz band transceivers and antennas; open challenges from communication and networking perspecti ves; simulation and experimentation tools for THz band communication. Akyildiz et al. [26] 2014 THz communication SotA in THz band transceivers and antennas; open challenges from communi- cation and networking perspectiv es. Akyildiz et al. [32] 2015 Internet of Nano-Things Introduction to the Internet of Bio-Nano Things (IoBNT); bridging the outside world and the IoBNT ; open challenges in the IoBNT . Miraz et al. [33] 2015 Internet of Nano-Things Short overvie w and future research directions pertaining to the Internet of Things (IoT), Internet of Everything (IoE), and Internet of Nano-Things (IoNT). Dressler et al. [34] 2015 Internet of Nano-Things Bridging the outside world and in-body nanonodes for health-care applications; IoNT network architectures; simulation tools for in-body nanonetworks. Petrov et al. [28] 2016 THz communication SotA in THz band communication; engineering trade-offs in typical applications; open challenges and research directions in THz band communication. Alsheikh et al. [35] 2016 THz nanocommunication Existing MAC protocols for WNSN; performance analysis and design guidelines for WNSN MAC protocols. Rizwan et al. [36] 2018 Nanocommunication in healthcare Nanosensors and nanonetworks for healthcare systems; big data analytics (data sources, preprocessing, feature extraction, visualization, predictive modelling) Ghafoor et al. [37] 2019 THz communication Features of the THz band; THz macro and nanoscale applications; design requirements for THz MAC protocols; classiﬁcation of existing MA C protocols; open challenges for MA C protocols. This survey 2021 THz nanocommunication THz nanoscale applications; THz nanocommunication and nanonetworking pro- tocols; THz nanoscale channel models; simulation and experimentation tools as applications that could potentially be enabled by THz band nanonetworks, which is in our case “embedded” in some of the speciﬁed application domains. A. Softwar e-Deﬁned Metamaterials Metamaterials (and metasurfaces, their two-dimensional counterparts [54]) are manufactured structures that enable powerful control of electromagnetic wav es. As such, meta- materials can be used to realize de vices with engineered and ev en unnatural properties related to the reﬂection, absorption, or transmission of electromagnetic radiation. For instance, metamaterials have been proposed for the electromagnetic cloaking of objects [55], noise [56] cancellation, holography [57], anomalous reﬂection [58], and focusing of energy with unprecedented accuracy [59]. Such unprecedented control is achiev ed through the careful design of a periodic array of subwa velength elements typically called unit cells . The main issue with current metamaterials, howe ver , is that the unit cells are “hard-coded” for a single application and operational condition (e.g., to work for a single angle of incidence) and cannot be reused across applications nor reprogrammed for different operations. T o alleviate this issue, Liaskos et al. [60] proposed Software-Deﬁned Metamaterials (SDMs), a new paradigm of programmable metamaterials where the unit cells can be reconﬁgured through a software interface with a set of well-deﬁned instructions. By virtue of the unprecedented and real-time control of the absorption, reﬂection, and transmission characteristics of metasurfaces, the SDM approach is e xpected to enable a plethora of applications in areas such as sensing, communi- cations, or imaging [61]. In wireless communications, SDMs hav e been proposed as a basic building block of intelligent r eﬂecting surfaces enabling the revolutionary concept of pr o- grammable wir eless envir onments [62], [63]. By controlling the reﬂection and scattering proﬁles at selected spots, both propagation loss and multipath effects can be mitigated in any wireless channel. This has proven to be a true paradigm shift in wireless communications, as the recent explosion of works can attest [64]–[67], because the channel has been traditionally an inescapable limiting factor . Another application related to wireless communications is the de velopment of simpliﬁed architectures for wireless communication transmitters [68]. There, the metasurf aces are utilized to directly modulate the carrier wa ve in multiple channels using the baseband signal without the need of core components of the RF chain such as mixers, ﬁlters, ampliﬁers, which can become hard to realize at high frequencies. Finally , we also highlight the applicability of SDMs in the area of holographic imaging, where real- time programmability opens the door to dynamic holograms with outstanding spatial resolution given by the subwa velength granularity of the unit cells [69]. Fundamentally , the behavior of SDMs hinges on the collec- tiv e response of each of its subwav elength unit cells. T o enable the reprogramming without compromising the autonomy of the metamaterial, the SDM paradigm proposes to embed a communication network of controllers within the metamate- rial, as shown in Figure 1. In such a scenario, each controller interacts locally with its associated unit cells to adjust its properties and communicates with other unit cells to obtain or 5 A S A S A S A C C C C C C C R R R R R R R Metamate rial plane Sensing and actuation plane Shielding plane Computing plane Communication plane Interf ace to the outside world (e.g., nano - micro, nano - IoT) Gatewa y Rout er Controller Actuat o r Sensor Recon figurab le E M behavior Figure 1: En visioned high-level architecture for enabling software-deﬁned metamaterials [72] distribute the desired behavior . Howe v er , the subwa velength scale the SDM unit cells poses a frequency-dependent form- factor limit on the internal network of controllers. In this context, to enable the SDM applications in a wide range of frequencies, nanocommunications in the THz band becomes a desired paradigm to implement the communication between the unit cell controllers within an SDM [70]. Depending on the actual application, the number of unit cells can range from thousands to millions [70], [71], translating to a similar number of controllers in the THz nanonetwork. Their exact number will depend on the physical sizes of the SDMs and the ﬁnal application, which in some predictions could co ver the walls of an entire ofﬁce for the programmable wireless en vironment case [64]. Moreov er , the controllers will have to be interconnected for better adaptation and operative range purposes. Due to the small form-factor and a huge number of en visioned controllable metamaterials comprising an intra- SDM network, the energy consumption of each metamaterial and consequently of the intra-SDM network will hav e to be low . Owing to practical reasons, these metamaterials will only have energy harvesting capabilities, with capacitor-based storage instead of batteries. Currently , a ﬁrst wa ve of SDM designs is under dev elop- ment to showcase the capabilities of this new technology [61]. The latenc y requirements are expected to be relaxed in this ﬁrst stage, i.e., between a few milliseconds and a few seconds [70]. Moreov er , very simple controllers and intra-SDM netw ork infrastructure will be deployed in ﬁrst prototypes. The network trafﬁc is expected to be downlink mostly , predominantly used for controlling the beha vior of metamaterials. Moreov er , the initial phase en visions no mobility , i.e., the metamaterials and network nodes are expected to be static. Lo w reliability of data deliv ery will presumably sufﬁce and the security is not expected to play an important role [70]. This is mainly because, in the current stage, the aim is to deliv er a proof of concept of the SDM paradigm. T o the best of our knowledge, deriv ations in regard to the performance requirements for SDMs in their roll-out phase are scarce in the current literature [73], [74] and we make a brief analysis here based on the discussions within [61]. In essence, the SDM is a programmable metasurface with a control sub- system that modiﬁes the response of the metasurface at a giv en maximum rate that depends on the application. In holographic displays, SDMs should be programmable at the typical refresh rates of displays, this is, around 60 times per second. In wireless programmable environments, SDMs need to be f ast enough to adapt to changes in the channel due to mobility or other phenomena, which typically occur at the 1-100 milliseconds scale. Finally , in SDMs used as wireless communication transmitters, the required refresh rate of the metasurface is thus the modulation rate of the transmitter . The ﬁgures above allo w us to make an estimation of the performance requirements of intra-SDM networks. On the one hand, the intra-SDM communication latency needs to be a fraction of the required refresh time: in the millisecond range for holographic displays and programmable wireless en viron- ments and below for metasurface-based wireless transmitters. W ith respect to the throughput, we can assume that an SDM module will consist of 10000 unit cells and will cover an area of 100 cm 2 based on recent estimates [64], [68]. If the range of an indi vidual intra-SDM controller operating in THz fre- quencies is 1 cm, roughly 10x10 interconnected controllers are needed for controlling the whole metamaterial. Say that ev ery 100 ms the metamaterial elements have to be updated and that this update is performed in a ﬂood-like multi-hop f ashion using a byte-long command, starting from a controller positioned in one corner of the metamaterial surface. Note that a byte-long command has been selected as most EM functionalities can be achieved with 8-16 states [75]. Under these assumptions, the number of hops until all controllers are reached equals 2 × (10 − 1) × 10 = 180 . Note that each transmission between controllers can potentially also be utilized by instrumenting the metamaterials controlled by the controller , hence we arrive to the required minimal data throughput of 1800 transmissions per sec = 14.4 kbps. Arguably , the signaling ov erhead in such scenario is relatively low and, to further simplify the analysis, we can assume that the netw ork throughput equals the data throughput. Note that, due to a variety of simpliﬁcations made in our deri vations, the derived value should be used only as a very rough indication of the minimum required network throughput. W ith that in mind, we can then provide a rule of thumb estimate of the required network throughput in the order of 1 to 50 kbps (T able II). As the SDM technology ev olves, new prototypes will arise that e xplore its full potential. Demonstrations of mission- critical applications with stronger timing requirements on the order of microseconds are expected [61], [70]. Moreover , SDMs are also envisioned to become wearable, thus ha ving the ability to bend, stretch, and roll [54]. F or the supporting network, this will represent an additional requirement in terms of shape resiliency and operation in high mobility scenarios. For enabling mission-critical applications and although SDMs show certain resilience to faults [76], the reliability of com- munication will have to be high, while some guarantees for security will also hav e to be in place. For supporting a v ariety of potential applications in this domain, addressing will be required on a lev el of an individual controller, or even on the lev el of an indi vidual metamaterial element. For similar reasons, the communication links will have to be bidirectional enabling communication from the user to the SDM unit cells and vice versa, as well as among SDM unit cells to implement distributed sensing and intelligence within the device [61]. In terms of throughput and using the same approach for analysis 6 as before and using 10 ms and 10 µ s as the metamaterial unit update period, we arriv e to a minimum required network throughputs of 144 kbps and 144 Mbps, respectively . Hence, roughly speaking the required network throughput for SDMs of the second generation will be in the range of 50 kbps to 500 Mbps. Note that in the deriv ation we assumed that each transmission between controllers is simultaneously used as a command for controlling the corresponding metamaterials. For more detailed trafﬁc analyses, we refer the reader to [73]. B. W ir eless Robotic Materials In contrast to SDMs that are en visioned to control electro- magnetic wav es, wireless robotic materials are expected to en- able smart composites that autonomously change their shape, stiffness, or physical appearance in a fully programmable way [77], [78]. The term wireless robotic material has been coined in [77], [79]. They deﬁne the robotic materials as multi- functional composites that tightly inte grate sensing, actuation, computation, and communication to cr eate smart composites that can sense their en vir onment and change their physical pr operties in an arbitrary pr ogrammable manner . The ap- plications that the authors in [77] suggest are airfoils that change their aer odynamic pr oﬁle, vehicles with camouﬂage abilities, bridges that detect and repair damage, or r obotic skins and pr osthetics with a realistic sense of touch. Similarly , the authors in [79] envision applications such as tactile sensing skin, robots (i.e., nanodevices) that can reproduce patterns projected onto them for camouﬂage, and a dress that can localize the direction of incoming sound and display it to its wearer . Several promising applications en visioned to be enabled by wireless robotic materials are depicted in Figure 2. As argued in [79], the en visioned applications will require very lar ge swarms of nanode vices tightly inte grated into fabric, on skin, etc. This poses limitations in terms of the size of the elements of robotic materials, yielding the THz band as one of the most promising communication paradigms for controlling these elements. Moreov er , the network size and density will be largely inﬂuenced by the application that the network is envisioned to support. W e belie ve that the application of enabling camouﬂage abilities will require the lar gest (i.e., cov ering an entire vehicle or human body) and most dense networks. Nonetheless, the requirements for network size and node density are expected to be less pronounced than for the SDMs, primarily due to the expected difference in sizes of the robotic materials (a fe w millimeters) and metamaterials (potentially much smaller than 1 mm). In terms of netw ork traf ﬁc, the authors in [79] ar gue that the en visioned sensors and actuators embedded in wireless robotic materials could generate information ranging from binary (e.g., for enabling distrib uted gesture recognition [80]) to a fe w-hundreds-of-hertz-bandwidth signals (e.g., localized texture recognition by robotic skin [81]). Let us provide a simple calculation for deri ving for supported trafﬁc load by the network of wireless robotic materials. Similar to the previous calculations, we make a v ague assumption that a robotic material patch will consist of 100 elements that have to be updated ev ery 20 ms, which is the assumption taken Figure 2: Example applications of wireless robotic materials from the T actile Internet use-cases [82] in which the network latency has to be comparable to the human observ ational abilities. Furthermore, we assume that this update is per- formed in a ﬂood-like multi-hop fashion, starting/ending at the source/sink node positioned in the corner of the wireless robotic material patch. Under these assumptions and following the same approach as before, the network traf ﬁc needed for controlling the wireless robotic material patch then equals 450 kbps and 3.6 Mbps for control signals carrying 1 and 8 bits of information, respectiv ely . Utilizing the numbers, we estimate that the network throughput of roughly between 100 kbps and 10 Mbps will be needed for enabling the wireless robotic materials-related applications. Note that, intuiti vely , the network will have to support bidirectional traf ﬁc, primarily for enabling the vision of sensing and actuating networks [83]. As summarized in T able II, the reliability of data deliv ery in the networks enabling wireless robotic materials will have to be high for some applications, e.g., for detecting and repairing damage on bridges, ceilings, and other “critical” structures. T aking the same application as an example, the security of data transmission will have to be high for the abov e-mentioned applications. The applications also in volv e wearable electronics, “smart” dresses, and artiﬁcial skin, all being carried by a person. Hence, the mobility is expected to be very high for some of the applications. In some cases, there will be a need to localize the nodes of the network under such mobility conditions, e.g., for localized texture recognition by robotic skin [81]. Furthermore, some applications will tolerate cluster -based addressing of nodes (e.g., bridge repairs), while some others may require individual addressing (e.g., camouﬂage). Finally , the energy consumption of the de vices and consequently in the netw orks will in some cases ha ve to be low , e.g., when the devices are expected to have long lifetime such as in construction monitoring scenarios. Howe ver , since these devices are expected to be larger than the metamaterials discussed above, their energy efﬁcienc y and power proﬁle requirements are not expected to be as stringent as for the metamaterials. Although the larger robotic materials could be powered by smaller batteries, for the smaller robotic materials and dense networks the authors in [79] suggest to use energy scav enging and harvesting [84]. 7 Note that in the literature, researchers often make a distinc- tion between wireless robotic materials and wireless nanosen- sor networks. W e ﬁnd such a distinction unnecessary , given that we separate dif ferent applications based on the require- ments they pose on the supporting nanonetwork. Wireless nanosensor networks have been proposed in [53] and envision applications such as high-resolution environmental monitor- ing [85], wearables [86], nanocameras-based extreme spatial resolution recordings [87], nanoscale imaging [88], and the Internet of Multimedia Nano-Things [30]. Nonetheless, these applications can be viewed as a sensing-only subset of appli- cations enabled by the wireless robotic materials, hence we do not group them into a separate category . C. In-body Communication Mobile medical nanodevices are a promising technology for in-situ and in-vivo applications [89], [90]. These nanode- vices will be access small regions of the human body (e.g., gastrointestinal, brain, spinal cord, blood capillaries, inside the eye), while essentially being non-in vasi v e. The authors in [89] argue that mobile medical nanode vices could even enable access to unprecedented sub-millimeter size r e gions inside the human body , which have not been possible to access curr ently with any medical device technology . Similarly , the authors in [91] en vision: “ ther apeutic nanomachines able to operate either in inter- and intra-cellular ar eas of the human body “. By doing so, pioneering applications, such as immune system support, bio-hybrid implant, drug delivering system, health monitoring, and genetic engineering, will be enabled, argue the authors. The authors in [2] en vision applications such as human physical movement monitoring, early diagnosis and treatment of malicious agents (e.g., viruses, bacteria, cancer cells), bone-growth monitoring in diabetes patients, and organ, nervous track, or tissue replacements (i.e., bio-hybrid implants). Finally , the authors in [92] outline sev eral example applications that could be supported by THz nanonetworks, such as smart drug administration, nanoscale surgeries, and epidemic spread detection and management. As indicated in Figure 3, the nanodevices would be able to perform sensing (e.g., blood composition or functioning of speciﬁc organs) and actuation (e.g., targeted drug deli very), all while reporting or being controlled from the outside world. Obviously , the form-factor of these nanodevices will be of prime importance, again yielding THz band communication as one of the most suitable communication paradigms. The number of mobile medical nanode vices is expected to be very large (up to a billion according to some estimations [93]) for some applications (e.g., for tissue engineering or detecting bacteria via swarms of sub-millimeter-scale nanodevices). The amount of trafﬁc that the network of mobile medical nanodevices will hav e to support will lar gely depend on the application. Giv en that the aim of the nanodevice is to enter and sense/inﬂuence something in sub-millimeter regions inside the human body , the main goals of the network will be to deliv er this information to the outside world and to support the control of the nanodevice. The network per-se will be formed by only a set of devices relaying the information to/from the Gatew ay (e. g., to the Internet, centr al server) Actuati on (e. g., dru g delivery , nano - surgery , targete d ce ll removal) Static sensin g (e.g., blood o r bone composition) Ta r g e t e d s e n s i n g (e.g., functio ning of organs, muscle strai ning ) Interf ace to the outside world ( e.g., nano - micro, na no - IoHT) Figure 3: En visioned high-level architecture for enabling body-centric applications outside world. Howe ver , given that the aim of a swarm of medical nanodevices is to sense a variety of ev ents inside of a body (e.g., the presence of a bacteria) or to form a tissue, the supporting network will arguably be mesh-like. The nanodevices will in this case also need to create control loops with the outside world, hence the network will have to operate under real-time constraints. Let us assume one such scenario, i.e., a swarm of mobile nanodevices is sensing a human brain and potentially creating an artiﬁcial tissue if there is damage in the brain. Roughly speaking, the network is distributed across the area of 1000 cm 3 and consists of 10 6 mobile nanodevices (a relatively conservati ve assumption), each one sending or receiving 8 bits of information ev ery 20 msec. W e assume ﬂood-based distribution of traf ﬁc from or toward the outside world. Utilizing similar calculation approach as before, we come to the staggering number of transmissions which equals roughly 3 × 10 6 transmissions/s, resulting in the network throughput of 24 Mbps. Along the abov e deriv ation, the required network throughput for enabling body-centric applications will - as a rule of thumb - be in the range between 1 and 50 Mbps. The body-centric communication related network require- ments are arguably the most challenging among the appli- cations potentially supported by THz band communication, as summarized in T able II. In addition to large network sizes, high throughput requirements, and in-body propagation of signals, the energy consumption of the nanodevices and consequently the networks supporting their operation hav e to be very low , primarily due to the required small form-factor of the devices. Hence, these nanode vices are presumably going to use energy harvesters are their only energy source. Moreov er , the reliability of communication will hav e to be very high (e.g., for controlling medical nanode vices in a brain), while the end-to-end latencies will hav e to be very lo w for enabling real-time control of the nanodevices. Obviously , the security of communication will have to be very high, especially for the ones en visioned to stay in a person’ s body for a longer period (e.g., for monitoring purposes). This is to avoid the nanodevices being “hijacked” by the attackers, while the patient is not being in a controlled en vironment shielded from the potential attackers. In cases when the nanonetworks are not extremely large, the addressing will have to be individual 8 in order to e.g., control the mo vements of a particular nan- odevice [89]. Moreover , the nanodevices are envisioned to be localizable and traceable [89] for enabling localized sensing and movement control of the nanode vices. Finally , due to both blood streams in a person’ s body and potential movement of the person (primarily the relativ e movement of person’ s body parts respectiv e to one another), the nodes are expected to be highly mobile, which poses an additional challenge for the supporting nanonetworks [89]. Note that the abo ve example, as well as the requirements that the applications are e xpected to pose on the supporting nanonetworks, have been derived for the most challenging applications. In contrast to the other promising application do- mains discussed in this w ork, for in-body communication there are se v eral recent efforts targeting the deriv ation of network requirements for only certain applications in the domain. In this direction, the interested reader is referred to [94], [95], where the authors deriv e the achiev able throughput as a func- tion of various network parameters such as transmission rate and nanonodes’ av ailable energy . The work was carried out assuming ﬂow-guided nanonetworks (i.e., the ones operating in the blood circulatory system). Moreover , the work was extended in [96] to demonstrate the feasibility of ﬂo w-guided nanonetworks in enabling lo w throughput applications such as viral load monitoring, restenosis, sepsis and bacterial blood infections, and heart attacks. Moreover , the authors in [92] qualitativ ely distinguish nanonetw ork requirements based on the high-le vel function of the applications. They en vision the monitoring, detection, and therapy-related functionalities, howe v er without providing quantitative characterizations. D. On-chip Communication V irtually all processors nowadays are based on multi-core architectures where a single chip contains multiple indepen- dent processor cores and a given amount of on-chip memory . The different processors compute in parallel and use the memory to share data and synchronize their executions. In this context, the current trend to increase performance is to integrate more cores within the same chip [97]. This, ho wev er , places an increased burden to the on-chip interconnect, which is used to send control signals and mov e shared data across the chip, to the point of turning intra-chip communication into the ke y determinant of the processor’ s computational performance and energy ef ﬁciency [98]. Hence, substantial research efforts focused on the on-chip interconnects, with initial bus-based interconnects soon giving way to more efﬁ- cient and resilient Networks-on-Chip (NoCs). Initially , NoCs were wired solutions, which posed challenges in terms of delay , po wer requirements, and area o verhead as more cores were integrated within a single chip [99]–[101]. This has prompted the proposal of multiple alternative interconnect technologies [101], [102], among which we ﬁnd wireless on- chip communications. The advantages of employing wireless communication for intra-chip networks include reduced propagation delay , re- conﬁgurability , and improved scalability in terms of latency , throughput, and energy consumption [103], [104]. Ne verthe- less, as shown in Figure 4, current WNoCs are predominantly Figure 4: En visioned high-level architecture for enabling on-chip communication [105] utilized for long-range point-to-point links for decreasing the av erage hop count of traditional NoC solutions. In other words, WNoC are currently deployed for enhancing the wired NoC, mostly due to the relati vely large sizes of the metal- lic antennas required for enabling wireless communication in the mmW av e band, which has generally been assumed in this context. Recently , Abadal et al. [105] proposed the employment of nanoscale WNoCs by means of graphene nanoantennas. Graphene-based nanoantennas with sizes of only a few micrometers, i.e., two orders of magnitude below the dimensions of metallic antennas, could provide inter-core communication utilizing the THz frequencies. Moreover , the antennas are inherently tunable, providing new ways to recon- ﬁgure the network. This novel approach is expected to fulﬁll the stringent requirements of the area-constrained, latency- bound, and throughput-intensi ve on-chip communication. This concept has recently been further dev eloped [106], [107]. In the en visioned multicore on-chip communication, on- chip traf ﬁc typically consists of a mixture of short control messages employed for cache coherence, data consistency , and synchronization purposes, together with larger data trans- fers. The communication is clearly bidirectional, while the addressing in the network has to be individual, as reﬂected in T able II. Depending on the memory access patterns exhibited by the application, communication will hav e varying degrees of unicast, multicast, and broadcast transmissions. Note, for instance, that some applications hav e strong all-to-all commu- nication patterns. Although latency is the primary concern in on-chip communications, as delays in the serving of pack ets essentially delay the whole computation, throughput is an important metric to av oid throttling of the computing cores. The recent literature reports the on-chip network throughput requirement in the range of 10-100 Gbps [98], [108], which can be further pushed to the Tbps barrier in communication- intensiv e architectures such as accelerators [109]. As mentioned above, the WNoC scenario has very chal- lenging end-to-end latency requirements in the range of sub- µ s [98] with high reliability (typical Bit-Error-Ratios (BERs) lower than 10 − 15 [98], [110] to compete with that of on- chip wires). As the chips are generally shielded, security issues related to on-chip communication are not of paramount importance. In the unlikely ev ent of hardware trojans being present within the chip, several lightweight measures can be taken to av oid spooﬁng, eav esdropping, or Denial of Service (DoS) attacks [111]. Moreov er , although the chips are e xpected to be mobile, the relativ e locations of the nodes on a chip are 9 T ABLE II: Summary of requirements in different application domains Requirements Software-deﬁned metamaterials Wir eless robotic materials In-body communication On-chip communication Gen. 1 Gen. 2 Network size 10 3 to 10 6 10 9 10 to 10 6 10 3 to 10 9 Up to 10 3 Node density 100 to 10000 nodes per cm 2 1 to 100 nodes per cm 2 > 10 3 nodes per cm 3 10-100 per mm 2 Latency ms to s µ s ms ms to s 10-100 ns Throughput 1-50 kbps 50 kbps to 500 Mbps 100 kbps-10 Mbps 1-50 Mbps 10-100 Gbps T rafﬁc type do wnlink bidirectional bidirectional bidirectional bidirectional Reliability low medium high very high very high Energy consumption very low very low low very low low Mobility none medium to high high high none Addressing none to cluster individual cluster to individual individual individual Security none low to medium high v ery high medium Additional features localization in-body communication localization & tracking not going to change. In that sense, there is no mobility of network nodes that should be anticipated. Due to very small sizes of the chips, the energy ef ﬁciency of the network should be high. This is to avoid overheating of the chip due to high energy dissipation [112].  A summary of the requirements that different application domains pose on the supporting communication networks is giv en in T able II. Compared to other application domains, software-deﬁned metamaterials posit the least stringent con- strains for the majority of requirements. This pertains to the throughput and security requirements, as well as the trafﬁc type and mobility support. On-chip communication can be considered as a relativ ely unique application domain as the topology of a network and propagation characteristics can be considered as static and kno wn upfront [113]. Nonetheless, compared to other domains, this domain poses the most stringent constraints in terms of node density , deliv ery latency , achiev able throughput, and reliability of communication. W e hav e mentioned before that electromagnetic nanocommuni- cation can be utilized in different propagation mediums, in contrast to molecular nanocommunication. In addition, we’ ve stated that electromagnetic communication in THz frequen- cies, primarily due to graphene, can support device min- imization and antenna tunability , in contrast to mmW av e, microw av e, and other electromagnetic frequency bands. Due to that and in contrast to other paradigms, electromagnetic nanocommunication in THz frequencies is a promising candi- date for supporting all of the discussed application domains. In theory , the THz band can support very large bit-rates, up to se veral Tbps. Ho we ver , it is clear from the discussion abov e and T able II that such throughputs will not be required for enabling the en visioned applications domains. Nonetheless, a very large bandwidth enables new simple communication mechanisms suited for the expected limited capabilities of nanodevices, primarily in terms of their energy levels [114]. In addition, a large bandwidth enables the de velopment of efﬁcient medium sharing schemes [115], which can both enable low energy performance and scale to the required numbers and densities of the nanonodes. In addition, it is worth emphasizing again that the THz frequency band is “the last piece of RF spectrum puzzle” [116]. Hence, the interferences from other communication sources in the same frequencies are currently virtually non-existent. In the future, they are expected to be very lo w , due to the low utilization, high attenuation, and large bandwidth of the THz frequency band [116]. This suggests that highly reliable nanocommunication, which is required by se veral of the outlined application domains, can be enabled by utilizing the THz band. The above considerations suggesting the potential for lo w power , highly scalable, and reliable communication make a strong argument in fav or of using electromagnetic nanocom- munication in the THz frequency band for supporting the desired application domains. Certainly , there are unresolved challenges to be addressed, and one aim of this survey is to detect, summarize, and discuss them. I V . N E T W O R K L A Y E R The network layer functionality is responsible for enabling data communication between connected THz nanonodes at arbitrary distance from each other . T o enable such commu- nication, nanonodes might rely on intermediate nanonodes (hops) to forward information. Forwarding functionality might be av ailable for all or a subset of nanonodes. Routing function- ality ensures that (collective) forwarding beha viour results into successful paths between data sources and destinations. The nanonodes therefore might rely on mechanisms to be identiﬁed and/or addressed, either individually or as a group (e.g., based on their physical location). Due to hardware limitations in transceiv ers, traditional routing, forwarding, and addressing schemes are often not applicable. Traditional routing proto- cols, as used in e.g., the Internet, rely on the exchange of control or meta messages to learn and distribute informa- tion about the network topology or reachability . Howe ver , memory , channel, and energy restrictions in nanonetworks impose stringent requirements on the relay , forw arding, and routing functionalities. These constraints are similar to those encountered in the domain of W ireless Sensor Networks (WSNs). Some THz routing protocols expand upon con ven- tional mechanisms by incorporating THz-speciﬁc link models when deciding between various paths (e.g., by incorporating molecular absorption loss). Ho wev er , an additional challenge of routing in THz nanonetworks is the scale, since the ma- jority of WSN routing protocols are optimized for up to 200 motes [117], while some of the application domains enabled by THz nanonetworks require signiﬁcantly larger number of nanonodes, as summarized in T able II. In addition, in many applications nanonodes are more prone to failures than sensor motes, as their only powering option is to harvest energy from 10 THz Network Layer Protocols Relaying & Forwarding (Sect. IV -A) Xia et al. [118] Rong et al. [119] Y u et al. [120] PESA WNSN [121] Pierobon et al. [122] C ´ anovas-Carrasco et al. [123] C ´ anovas-Carrasco et al. [124] Routing Flooding- based routing (Sect. IV -B) Liaskos et al. [125] Tsioliaridou et al. [126] Afsana et al. [127] LENWB [128] Buther et al. [129] E 3 A [130] Pathﬁnding Regular topologies (Sect. IV -C) Radetzki et al. [131] Maze routing [132] Fukushima et al. [133] Jovano vic et al. [134] Ebrahimi et al. [135] Saeed et al. [136] Irregular topologies (Sect. IV -D) Tsioliaridou et al. [137] Tsioliaridou et al. [138] Figure 5: Classiﬁcation of network layer protocol for THz nanocommunication the av ailable en vironmental sources. Therefore, other research efforts take application-speciﬁc information into account for settings which are outside the scope of WSNs. Below we provide an overvie w on existing research cat- egorized according to the core mechanism they rely on. Figure 5 depicts an ov ervie w of our approach. Subsection IV -A discusses forwarding and relaying methods. These works focus on whether to transmit a packet directly to a hop which is further away , or whether to relay the information via one or more intermediary nanonodes. The other subsections focus on algorithms aiming at ﬁnding the nanonodes that should send and/or forward a packet in order to reach one or more destinations (i.e., routing). Subsection IV -B focuses on ﬂooding-based mechanisms, which are used for one-to-many routing. Subsections IV -C and IV -D discuss one-to-one routing in regular and irregular topologies, respectiv ely . A. Relaying and F orwar ding for THz Nanonetworks Enabling multi-hop communication in THz nanonetworks introduces design restrictions at the lower layers. Xia and Jornet [118] mathematically characterize relaying strategies maximizing network throughput with respect to transmission distance, transmission po wer ener gy and pack et generation rate in a THz band network consisting of nanonodes with direc- tional antennas. The unique characteristics of THz communi- cation are considered by modeling the distance-dependent and frequency-selecti ve characteristics of molecular absorption. T o reduce the BER of cooperative relaying strategies in WNSNs, amplify-and-forward and decode-and-forward relaying nanon- odes are ev aluated by Rong et al. [119], whilst taking into account spreading and molecular absorption losses. Using the former forwarding technique, relaying nanonodes amplify the receiv ed signal as it is, whilst using the latter technique, relay- ing nanonodes demodulate the signal before it is forwarded. The bit error rate of both schemes is compared to that of direct transmission, and a relaying gain of approximately 2.2 dB was observed for amplify-and-forward and 5 dB for decode- and-forward. As molecular absorption and other frequency- selectiv e features seriously hamper multi-hop throughput in the THz band, Y u et al. [120] propose channel-aw are forwarding schemes in WNSNs, ensuring that data will not be forwarded to: i) a relaying nanonode in a region which is adversely affected by molecular absorption, or ii) to a short-distant nanonode, which will result in unnecessarily large hop count. Compared to direct transmission, the authors report an increase of end-to-end capacity by a factor 12-13, whilst limiting the delay increase to less than 0.02%. Similarly accounting for molecular absorption, Y en et al. [121] combine signal quality-aware forwarding with data aggregation in WNSNs, while also taking free path space loss into account. Data aggregation (e.g., taking the minimum of receiv ed values) in intermediary nanonodes is often possible, depending on the particular application (e.g., if only the lo west value is needed). The resulting design problem is formulated as an optimization of routing decisions, for which an efﬁcient heuristic has been proposed for calculating minimum cost spanning trees considering transmission power and signal quality . Energy efﬁciency is crucial in THz band nanonetworks. The throughput attainable in a these en vironments is therefore strongly related to the associated nanoscale energy harvesting processes. Pierobon et al. [122] present a hierarchical cluster- based architecture for WNSNs, extending their work on MAC protocols [115]. Each cluster is assumed to have a nanocon- troller , which is a nanonode with more advanced capabilities and which can transmit directly to all nanonodes (when giv en sufﬁcient transmission po wer). The nanocontroller is in charge of optimizing the trade-of f between multi-hop forward- ing among individual nanonodes vs. communication via the nanocontroller from the perspective of throughput and lifetime of the network and its associated connectivity . Based on the probability of energy saving through multi-hop transmis- sion, the nanocontroller instructs nanonodes what transmission power to use for optimal hop distance and throughput, as well as selects the next hop on the basis of their energy and current load. Both total path loss and molecular absorption loss are considered in modeling the channel. For nanonodes at a distance of 1 to 10 mm, the authors ﬁnd that the total energy spent per bit halves using forwarding as compared to direct transmission, whilst the delay improv es by a factor 3-4. Another similar hierarchical architecture with a focus on energy management is proposed by C ´ anov as-Carrasco et al. ([123] and [124]). Concretely , a body area network is pro- posed, compromising a large number of nanonodes injected into the bloodstream and thus always moving. These nanon- odes communicate with one or more nanorouters, which are more powerful devices implanted under the skin. The nanorouters in turn connect to an Internet gateway . The ﬁrst of these works [123] focuses on a scenario in which a nanorouter is implanted in a human hand. The work discusses a molecular absorption loss model which takes the different layers of tissues inside the hand into account. Also, it proposes concrete power supply components which harvest energy from the bloodstream itself and from an external ultrasound source (the ultrasound-based po wer source also allo ws nanonodes to detect whether the y are in vicinity of the nanorouter). The authors estimate that each nanonode is able to send a 40 bit packet ev ery 52 minutes. [124] considers a scenario in which the nanonodes report events with different levels of importance. A Marko v Decision Process (MDP) for the transmission polic y of a nanonode is proposed, the states of which take the type of ev ent detected by the nanonodee into account, as well as the battery lev el and the distance between the nanonode and the last detected nanorouter . The authors compare their approach 11 to policies in which all ev ents are transmitted, only high priority e vents are transmitted or a decision whether to report an e vent is made based on an estimate of the nanonode’ s po wer lev el when reaching the ne xt nanorouter . The throughput of the proposed approach always exceeds the throughput of any other policy by at least a couple of percents. Moreo ver , as ev ents become more frequent, the advantage of the MDP over the other approaches becomes more pronounced. B. Flooding-based Routing Baseline THz nanonetworking mechanisms are in essence direct extensions of MAC protocols which rely on ﬂooding to deliv er data to their intended destinations. Flooding inv olves unconditional packet retransmission by all in v olved nanon- odes. Advantages of ﬂooding include its simplicity , its reliabil- ity through redundant transmissions and a lack of (topology- dependent) initialization, making it a good choice for ap- plications with mobile nanonodes. Unfortunately , unmodiﬁed ﬂooding inv olves a high number of redundant transmissions. Therefore, sev eral efforts hav e been undertaken to miti- gate the number of redundant transmissions by limiting the in volv ement of particular nanonodes or through selectively forwarding in a probabilistic manner . Liaskos et al. [125] adopt an adaptiv e ﬂooding scheme where wireless nanonodes in a static environment can deactiv ate themselves based on their percei ved signal to interference ratio and resource le vels and thus create a dynamic ﬂooding infrastructure. Compared to ﬁne-tuned probabilistic ﬂooding, the proposed scheme achiev es an improv ement of approximately 10% in packet rate to achieve full network coverage and a similar latency . It also manages to fully remove interference effects. Tsioliaridou et al. [126] follow a similar mechanism relying on the Misra- Gries algorithm to determine if each nanonode should act as user or retransmitter . Concretely , each incoming packet is classiﬁed as (i) suffering from a parity error, (ii) a duplicate of an earlier receiv ed packet, (iii) a correctly decoded packet which is recei ved for the ﬁrst time. This results in a stream of ev ents, with the Misra-Gries algorithm is employed to estimate the frequency of each e v ent, thus resulting in statistics which are used to decide if a nanonode should act as a retransmitter . Molecular absorption, fading effects and elec- tromagnetic scattering (simulated using a ray tracing engine) are taken into account in simulating the physical layer . In the context of body area networks, Afsana et al. [127] propose a cluster-based forwarding scheme in which cluster controllers are elected as a function of their residual energy . Network communication is then optimized for intra- and inter-cluster communication considering energy consumption at the link lev el. The resulting performance has been ev aluated in terms of Signal to Interference Plus Noise Ratio (SINR) (for an SNR up to 20 dB) and outage probability up to 0.5, considering the impact of shadowing, molecular absorption, and spreading loss. In all cases, the proposed scheme results in a data-rate improv ement of 3-15% over the baseline. V arious ﬂooding mechanisms are discussed by Stelzner et al. [128] in the context of in-body nanonetworks. The baseline forwarding mechanism ﬂoods messages based on a ﬁxed, predetermined probability . A more advanced scheme, ‘probA ’ in volv es adaptively changing that probability according to the estimated density of transmitters close to a given nanonode. Both mechanisms are ev aluated within an en vironment mod- eling an aorta and artery (a cylindrical environment with a wide or narrow radius). Probabilistic ﬂooding requires ap- proximately 60% of nanonodes to participate in the routing to achiev e full network coverage in both cases. ProbA performs well in the aorta and requires participation of only 40% of nanonodes, but f ails to achiev e full network coverage in the more restricted artery . As an alternativ e to probabilistic ﬂooding, Stelzner et al. [128] also propose a scheme (i.e., LENWB ) requiring nanonodes to store information up to 2-hop distant neighbors. In the aorta, LENWB achiev es full network coverage when 50% of nanonodes participate, and in the artery LENWB performs even better , achieving full network coverage with only 20% of nanonodes forwarding the messages. Ho we ver , we do note that this high delivery reliability and coverage comes at a signiﬁcant cost in terms of memory requirements. Buther et al. [129] go a step further , and propose individual nanonodes to learn and store their hop distance to a micro- scale gateway . The hop count is then used as a direction indicator to determine if the packet needs to be ﬂooded or not. They consider a naive version of the algorithm, in which nanonodes do not remember if a message has been forwarded, and propose an optimized approach by inv alidating the hop count of nanonodes once they hav e acted as a forwarding nanonode. This results in decreasing the total massage count from exponential to quadratic beha viour . A similar approach is taken by Al-T urjman et al. [130] in the context of IoNT , where the next hop to wards a gatew ay node is not only determined by its hop count, but is also restricted to the ﬁrst next hop which satisﬁes some energy-related constraints (e.g., at least 50% battery remaining). The path loss model takes both molecular absorption and spread path loss into account. When compared to a shortest path data collection strategy , the proposed approach avoids ov erusing nanonodes which are close to a gateway , thus resulting in a larger network lifetime (6 times or more). This howe ver comes at the cost of a higher latency due to the selection of longer paths (2-3 times higher). Each of the discussed algorithms is (to a v arying degree) successful in reducing the number of redundant transmissions, and may be considered for implementing one-to-many com- munication. Howe ver , for when one-to-one communication is required, ﬂooding is inefﬁcient by nature, since all nanonodes in the network receive each message. ([129] is an exception, but even there all nanonodes will receive a message if the target is far enough from the gateway .) Furthermore, the reduction in the number of transmitted messages is typically at the cost of some advantage of ﬂooding: probabilistic ﬂooding may fail to deliver messages if not tuned carefully , and other mechanisms collect topology-dependent information, making them less useful if the nanonodes are mobile. C. Routing in (Semi-)Regular T opologies Some applications require to transmit messages to a spe- ciﬁc nanonode, in which case ﬂooding (which transmits a 12 T ABLE III: Summary of network layer protocols Protocol Distinct Featur es Potential Applications Evaluation T opology Evaluation Metrics Xia et al. [118] - directional antennas - directivity-sensiti ve buf fer and queuing - wireless robotic materials - 2D Poisson distributed nodes - grid of relay nodes - throughput vs. packet generation rate and transmission distance Rong et al. [119] - amplify-and-forward vs. decode-and-forward - wireless robotic materials - body-centric communication - 3 node network - bit-error rate Y u et al. [120] - channel-aware forwarding - wireless robotic materials - body-centric communication - 1D (string) network - uniform or random nodes - 5 to 50 nodes - end-to-end capacity - average latency PESA WNSN [121] - power and signal quality -aware arc weight - data aggregation - software-deﬁned met. (gen. I) - wireless robotic materials - body-centric communication - 2D random nodes - 10000 nodes - power consumption vs. transmission range Pierobon et al. [122] - hierarchical cluster-based - distributed probabilistic energy-based forwarding - wireless robotic materials - body-centric communication - 2D Poisson distributed nodes - 100 nodes - average latency - capacity per node - energy efﬁciency C ´ anov as-Carrasco et al. [123] - hierarchical cluster-based - circulating nanonodes - energy harvesting - body-centric communication - 1 nanorouter implanted in the skin and 1 nanonode in the center of a vein - energy efﬁciency C ´ anov as-Carrasco et al. [124] - hierarchical cluster-based - circulating nanonodes - MDP-guided event reporting - body-centric communication - cylindrical volume - random nodes - 6000 nodes - energy efﬁciency - effect of nanorouter position on throughput Liaskos et al. [125] - adaptive ﬂooding - nodes may deactiv ate - software-deﬁned met. (gen. I) - wireless robotic materials - body-centric communication - on-chip communication - 2D uniform grid - 625 to 4000 nodes - achieved cov erage - mean service time - energy efﬁciency Tsioliaridou et al. [126] - adaptive ﬂooding - based on reception statistics - Misra-Gries algorithm - software-deﬁned metamaterials - on-chip communication - 2D and 3D uniform grids - 1000 to 8000 nodes - achieved cov erage - mean service time - energy efﬁciency Afsana et al. [127] - cluster-based forwarding - centers selected based on residual energy - body-centric communication - unspeciﬁed - throughput vs. SNR LENWB [128] - probabilistic ﬂooding vs. (locally) adaptive ﬂooding - body-centric communication - cylindrical volumes of various sizes - random nodes - 10 to 2000 nodes - coverage - memory usage Buther et al. [129] - ﬂooding based comm. - using hop-distance - destructive mode to av oid broadcast storms - body-centric communication - cylindrical volume - random and uniform nodes - up to 60 nodes - power consumption vs. network size E 3 A [130] - sensor-to-gate way comm. - forwarding based on hop- distance and energy level - software-deﬁned met. (gen. I) - wireless robotic materials - 2D uniform grid - 1500 nodes - average latency - failure rate - energy efﬁciency Tsioliaridou et al. [137] - geographic routing using hop-distance coordinates - integer calculations only - software-deﬁned met. (gen I.) - 2D uniform grid and random nodes - 10000 nodes - failure rate - energy efﬁciency Tsioliaridou et al. [138] - geographic routing using hop-distance coordinates - integer calculations only - coordinates selection alg. - software-deﬁned metamaterials - wireless robotic materials - 3D uniform grid and random nodes - 5000 nodes - failure rate - energy efﬁciency message to all nanonodes in the network) is an inefﬁcient solution. T o achieve such unicast communication, some strong form of addressing is needed (i.e., nanonodes needing an identiﬁer , physical or logical location). When network topolo- gies follow regular patterns, relativ ely simple addressing and routing mechanisms become possible. This is the case for NoC applications, where nanonodes are often static and laid out according to a regular grid pattern, enabling hard-coded coordinates for routing nanonodes among the two main axes. Researchers designing wireless NoC protocols often en vision cores communicating using THz frequencies because carbon nanotube-based antennas hav e the potential for higher trans- mission rates and lo wer po wer and area o verhead than e.g., ultra-wide band communication [139]. A simple XY routing mechanism lets routers forward mes- sages along the X-axis (horizontal) ﬁrst, until a router is found with the same X coordinate, and subsequently the routers forward along the Y -axis until the destination is reached. Greedy forwarding mechanisms rely on a distance metric which can be calculated between coordinates and aim to choose a neighbor which minimizes the distance to the desti- nation. In the presence of faults or in non-planar topologies, greedy forwarding might result in a message being routed to a non-destination nanonode with no distance decreasing neighbors. Thus, f ault tolerant routing techniques have been dev eloped, an overvie w of which is provided by Radetzki et al. [131]. Since some of the core ideas for fault-tolerance were originally proposed for (wired or wireless) NoC settings, we provide an overvie w of these papers. Since these efforts do not speciﬁcally target THz nanocommunication, they are omitted in T ables III and VII. One important technique in fault tolerant routing is face routing, which deﬁnes rules to route around spots with faulty nanonodes. Maze routing [132] adds e xtra ﬁelds to a message, allowing to route around fault regions without requiring nanonodes to store any information other than their coordinates. Maze routing is howe ver limited to 13 planar topologies, and the route may be far from optimal. Fukushima et al. [133] propose to collect local information to group faulty nanonodes in rectangular non-overlapping fault regions. The restriction that the fault regions must be non- ov erlapping may require to expand a fault re gion to include some operational nanonodes. Such healthy nanonodes inside a fault region cannot receiv e packets and thus must be switched off. More complex shaped fault regions can be realized by expanding the information collection region [134]. Using fault-tolerant routing to reroute packets around faulty regions will increase the packet latency and create congestion around the faulty region. Ebrahimi et al. [135] augments the XY algorithm with local nanonode information to route along failures reducing congestion through the incorporation of local queue and buf fer information. Similarly , Saeed et al. [136] pro- pose two fault-adaptiv e XY routing mechanisms (one av oiding loops, and another maximizing success deli very probability) enabling communication between a network of controllers in the context of hyper-surfaces. These fault tolerant algorithms typically trade off the information required to take a routing decision (and the corresponding algorithmic) complexity , with ﬂexibility to deal with more complex fault scenarios. An exception to this is maze routing, but that approach is restricted to planar topologies and may result in suboptimal paths. D. Irr e gular T opologies When network topologies get even more irregular , nanon- odes need a mechanism to determine their own coordinates. Tsioliaridou et al. [137] was one of the ﬁrst approaches to use a number of ﬁxed anchor points in a 2D or 3D space (as in software-deﬁned meta-materials) to ﬂood announcements of their existence across the network. Based on triangulation, nanonodes could determine their (non-unique) coordinate. Once the initialization phase is over , nanonodes can participate in a stateless manner in the packet forwarding process which consists of selectiv e ﬂooding tow ards the destination. The authors compare their approach with probabilistic ﬂooding and a dynamic infrastructure ﬂooding approach. Since the ﬂooding approaches deliv er each packet to all nanonodes in the network, they both require many more nanonodes to unnecessarily spend energy on decoding the massage (5 times more in the paper , but this depends on the network size). This approach was further reﬁned in Tsioliaridou et al. [138] by proposing a routing approach which further minimizes required retransmits. A mechanism is proposed relying only on integer calculations and nanonode-local information enabling each nanonode to deduce whether it is located on the linear segment connecting the sender to the recipient nanonode. The energy efﬁcienc y of the scheme can be further optimized by tuning the width of the linear path. This path width allows to trade-off reliability and energy efﬁcienc y: a larger path width allows the algorithm to deal with more irregular scenarios, but it also results in an increased amount of transmitted messages. THz-band speciﬁc features such as molecular absorption and shadow fading are also taken into account when e valuating the algorithm. Both of these schemes of fer ef ﬁcient one-to-one routing solutions, but they require a static en vironment, as nanonode mobility would result in frequent in v alidation of the coordinate systems. Another open challenge is that these techniques may enable nanonodes to determine their own coordinates, but these works do not address ho w they might obtain the coordinates or addresses of other nnanonodes with whom communication is needed. The particular addressing needs of medical application scenarios are conceptually inv estigated in Stelzner et al. [140]. The authors distinguish addressing from guidance concepts. The latter refer to alternate solutions to reach a target, without requiring an explicit address in communication, e.g., through kinds of wiring, electromagnetic ﬁelds, or bio-circuits. Function-centric nanonetworking refers to a scheme were location and functional capability of groups of nanonodes are addressed rather than the communication (individual) end- point(s). The location can refer to an area deﬁned in the human body , the function refers to a type or category of nanonodes rather than an individual one (e.g., blood pressure sensor).  T able III lists publications focusing on one or more network layer protocols. For each publication, the main novel features are gi ven, as well some applications proposed by the authors themselves. W e also mention the speciﬁc topology in which the routing schemes were ev aluated and the ev aluation metrics considered by the respectiv e authors. Across different applications and topologies, there is a large focus on minimizing power consumption and ener gy ef ﬁciency . Howe v er , these metrics are often exclusi vely e valuated based on the number of sent (and recei ved) messages. This provides an incomplete picture, since the power per message also depends on the transmission distance and the message length. Resiliency has also been studied in various settings: papers focusing on ﬂooding typically discuss network coverage (the percentage of nanonodes which recei ve a one-to-all broadcast). Another example is Tsioliaridou et al. [138], where a path redundancy parameter is introduced, which allows to tune the number of nanonodes which participate in the transmission of a point-to-point message. Most applications require large numbers of participating nanonodes. Nonetheless, the scalability of routing protocols is often not ev aluated. T able III shows that almost all works focus on networks with a set number of nanonodes, typically corresponding to a small fraction of the larger network. Howe v er , papers which do look into the effect of network size, such as Liaskos et al. [125] and Stelzner et al. [128] ﬁnd that parameters such as network cov erage and power consumption depend non-linearly on both the area/volume covered by the network, and the density of the nanonodes. Future work should ev aluate the sensitivity in function of this metric, as limiting the e valuation to a set network size may result in conclusions which do not generalize well to larger or smaller networks. Whilst collecting these results,the ef fect of nanonode mo- bility remains often overlooked. This is permitted in some applications, where the topology is indeed static, or net- work changes occur very infrequently . Such scenarios include network-on-chip, the monitoring of mission critical materials and some software-deﬁned metamaterials and IoNT/WSN set- tings. Howe ver , in-body network applications, the network is intrinsically mobile: when nanonodes operate within a blood- 14 stream they move due to the blood ﬂow , if they are attached to tissue, the body in which they are situated may move and change their relative positioning. From the e xisting research which targets in-body network applications, only Buther et al. [129] and C ´ anov as-Carrasco et al. [123] [124] mention nanonode mobility . The work of [123] and [124] focuses on nanonodes reporting events rather than nanonodes communi- cating directly with each other . Since an ev ent message can be transmitted via any nanorouter, the main issue in this setting is a) knowing if, and b) estimating how long it will be until, a particular nanonode is in reach of a nanorouter . The ﬁrst issue is addressed in [123] via an ultrasound energy source. The second issue is approached in [124] using MDPs. In [129], the mobility issue is addressed by periodically in v alidating and reinitializing all routing information. Whilst such an approach is (in theory) always possible, we note that this may place a considerable burden on the network load and is costly in terms of energy consumption. Also, even in a slow moving en vironment, the frequency at which an individual connection change between an y pair of nanonodes may be high (due to there being a large number of nanonodes), and thus, routing protocols which depend on global network information may require frequent updates. In future work, the effect of mobility on a network protocol should be carefully ev aluated and simulated in an en vironment in which nanonodes are continuously moving. Ideas from research regarding mobile ad-hoc networks could provide a valuable starting point. Finally , due to the large variety of application scenarios, ev aluation topologies and metrics, it is difﬁcult to compare the results of various authors. There are various approaches to remedy this problem: open-sourcing the network topologies may allow researchers to easily ev aluate their algorithms in a standardized variety of settings (without needing the in-depth knowledge to generate the speciﬁc topologies). Howe ver , there are many other modeling and ev aluation parameters which may inﬂuence the results and complicate any comparison. In vie w of these observ ations, contributions focusing on the numerical comparison of various existing algorithms in a variety of settings, such as Stelzner et al. [128] are highly valuable tools to provide a comparison of existing work. Such work is unfortunately rare, and should be further encouraged. V . L I N K L A Y E R In THz band nanonetworks, link layer protocols are used for enabling direct communication between a pair of nanonodes or between a nanonode and a more powerful de vice (e.g, nanorouter , nanocontroller, gatew ay , nano-macro interface). The primary functions of link layer protocols are channel access coordination, which is traditionally performed on the MA C sub-layer , and recovery from bit transmission errors, usually performed on the Logical Link Control (LLC) sub- layer . Classical MAC and LLC protocols cannot be directly applied in THz band nanocommunication for the following reasons [35], [114]. First, the existing link layer protocols hav e been predominantly designed for band-limited channels. Howe v er , the THz frequency band provides almost a 10 THz wide bandwidth windo w , which constitutes the main dif ference THz Link Layer Pr otocols Distributed (Sect. V -A) Akkari et al. [144] Alsheikh et al. [145] PHLAME [114] DRIH-MA C [146] TCN [147] Xia et al. [148] Smart-MA C [149] APIS [150] Hierarchical (Sect. V -B) W ang et al. [115] EEWNSN-MA C [151] EESR-MA C [152] CSMA-MA C [153] On-Chip Communication (Sect. V -C) Mansoor et al. [154] BRS-MA C [108] Dynamic MA C [155] ; Figure 6: Classiﬁcation of link layer protocols for THz nanocommunication Nanonode Nanonode Nanocontroller Communication link Distributed protocol Hierarchical protocol Figure 7: Distributed vs. hierarchical MA C protocols between graphene-enabled THz nanocommunication and clas- sical link layer protocols. Second, existing link layer protocols are mostly carrier-sensing techniques and are therefore too complex for THz nanocommunication in a number of ap- plication domains and scenarios. Finally , nanode vices feature highly limited energy , in various scenarios requiring the usage of energy-harvesting systems [10], [141], [142]. This changes the av ailability of the nanode vices’ communication systems ov er time, which is not a constraint posited to the classical link layer protocols [72]. This has been recognized in the research community and se veral link layer (often MA C sub-layer only) protocols for THz nanocommunication have been proposed. As shown in Figure 6, they can be grouped into hierar- chical protocols that assume the a vailability of more power - ful nanocontrollers, and distributed, in which all nanonodes are considered as equal (Figure 7). In addition, protocols speciﬁcally designed for on-chip nanocommunication can be separately grouped, due to the uniqueness of communication features, as in more details discussed below . Compared to the distributed protocols, the main advantages of the hierarchi- cal link-layer protocols include reduced energy consumption, increased network scalability , and increased reliability due to interference and collision probability [143]. Their main disadvantages are their comparably higher complexity and deliv ery latency . A. Distributed Pr otocols Akkari et al. [144] reason that there is a need for novel link layer protocols for nanonetworks due to the fact that nanodevices hav e strict power , memory , energy , and com- putation constraints. Thus, the authors argue, the nanonodes may only be able to store one packet, requiring packets to be delivered before certain hard deadlines. Motiv ated by this claim, they propose a distributed and computation-light 15 MA C protocol for nanonetworks. The protocol determines the optimal transmission times for nanonodes so that the lar gest set of trafﬁc rates can be supported, while ensuring deli very within a hard deadline. Optimal transmission decisions are made locally using Carrier-Sense Multiple Access (CSMA) Markovian chain and L yapunov optimization, which are based on nanonode’ s incoming trafﬁc rate and queue length. The pro- tocol achie ves continuous communication by simultaneously considering its energy consumption and harvesting rate. Motiv ated by the fact that nanonodes communicating in the THz band are capable of achie ving very high transmission bit- rates at v ery short distances, the authors in [145] also argue that classical MA C protocols cannot be directly applied in THz band nanocommunication. Therefore, they propose an energy- aware MA C protocol for synchronizing the communication between nanodevices. The proposed protocol is based on a T ime Spread ON-OFF k eying (TS-OOK) scheme (more details in Section VI), which is a pulse-based communication scheme for THz band nanonetworks, and assumes grid-like distribution of the nanonodes. In the protocol, active nanonodes transmit their packets in an interleaved way to the receivers based on the calculated Critical Transmission Ratio (CTR), i.e., the maximum ratio between the transmission time and the energy harvesting time needed for continuous operation. Jornet et al. [114] propose the PHysical Layer A ware MAC protocol for Electromagnetic nanonetworks in the T erahertz Band (PHLAME), which is built on top of modiﬁed TS- OOK. In PHLAME, the transmitting and receiving nanonodes jointly select the transmission symbol rate (the ratio between time between symbols and pulse duration) and channel coding scheme. They do that by performing a handshaking process initiated by a nanonode that has information to transmit and enough energy to complete the transmission. Using the common coding scheme, the transmitter generates a packet containing the synchronization trailer , transmitter ID, receiver ID, packet ID, randomly selected transmitting data symbol rate, and error detecting code. The handshaking acknowledg- ment is issued by the recei ver upon the reception of the packet. If the handshake is accepted, the receiv er issues a packet containing the synchronization trailer , transmitter ID, receiv er ID, packet ID, transmitting data coding scheme, and error detecting code. The transmitter then transmits a data packet using the symbol rate speciﬁed by the transmitter , and encoded with the weight and repetition order speciﬁed by the receiv er during the handshake process. The authors show that the proposed protocol is able to support densely populated nanonetworks in terms of energy consumption per useful bit of information, average packet delay , and achiev able throughput. Another distributed MAC protocol is Distributed Recei ver - Initiated Harvesting-A ware MAC (DRIH-MA C) [146]. Simply stated, in DRIH-MA C (RIH-MA C in its preliminary ver - sion [156]) the communication is initiated by the receiv er by transmitting a Ready-to-Receiv e (R TR) packet to one or multiple transmitters. The recipients of the R TR packet transmit a data packet to the receiv er . Scheduling in DRIH- MA C is based on a probabilistic scheme-based on the edge- coloring problem. The idea of edge-coloring is to color the edges of a graph such that two edges incident to the same node are of different colors. Obviously , in DRIH-MA C different edge colors translate to transmission sequences. The authors claim DRIH-MA C to be scalable and light-weight, with the minimized probability of collisions and maximized utilization of harvested energy . The authors in [147] propose T iming Channel for Nanonet- works (TCN), a link layer THz band nanocommunication protocol that exploits “timing channels”. They deﬁne the timing channels as logical channels in which the informa- tion is encoded in the silence duration between two conse- quent transmissions. The authors argue that, by using tim- ing channel-based communications, TCN enables ener gy ef- ﬁcient lo w data-rate communication. Moreover , by introduc- ing ackno wledgment-based collision detection, TCN enables recov ery from transmission errors. Retransmissions are then en visioned for improving the reliability of communication. A link-layer synchronization and MA C protocol for wireless communication in the THz band is presented in [148]. The link-layer synchronization capability is achieved through a receiv er-initiated (i.e., one-way) handshake procedure. The core idea of the handshake is to pre vent data transmissions when the receiv er does not hav e enough energy for reception. Additionally , the protocol aims at maximizing the channel utilization and minimizing the packet discard probability . It does that by making use of a sliding ﬂow-control window at the link-layer , i.e., the receiv er speciﬁes the amount of data that can be receiv ed for its current energy level. The authors in [149] propose Smart-MA C, a MA C protocol deliv ered as a part of the NanoSim simulator and often used as a baseline for benchmarking of other MA C protocols. Smart- MA C uses a handshake procedure for discovering nanonodes within its transmission range. If at least one nanonode is discov ered, the packet is transmitted. In case the nanonode does not ha ve an y neighbors, Smart-MA C uses a random back- off delay prior to restarting the handshak e procedure. In case of multiple packets in the queue, a new transmission is scheduled after a speciﬁc time computed through the same back-off mechanism, which reduces the probability of collisions. A distributed scheduling mechanism named Adaptiv e Pulse Interval Scheduling (APIS) is proposed [150] for the intercon- nection between nanonetworks and an IoT gateway , targeting software-deﬁned metamaterial and wireless robotic material applications. In APIS, the sink distribute a transmission sched- ule, followed by the nanonodes transmitting data based on the channel sensing results and trafﬁc patterns. The proposed protocol was ev aluated in terms of reliability of data delivery , achiev able throughput, and energy consumption. Due to the scheduled transmissions, high time precision and continuous operation of the nanonodes are required for scheduling, which could be infeasible under limited computational capacity and stored energy . Moreover , the communication between the nanonodes is abstracted and not considered in this work. B. Hierar chical Pr otocols The authors in [115] propose an energy and spectrum- aware MA C protocol for THz band nanocommunication. First, they propose to utilize the hierarchical network architecture 16 characteristic to WNSNs for shifting the protocol complexity to more resourceful nanocontrollers. Therefore, the nanocon- troller regulates the channel access on behalf of the nanon- odes of its cluster . The nanocontroller does that by utilizing T ime Division Multiple Access (TDMA) and based on the nanonodes’ data requirements and energy constraints. The proposed MA C protocol utilizes CTR, i.e., the maximum allow able ratio between the transmission and energy harvest- ing times below which the nanonode consumes less energy than harvested. Based on this and assuming TS-OOK as the physical layer communication scheme, a symbol-compression scheduling protocol is proposed for assigning each nanonode with different sets of transmission slots in such a way that the overall nanonetwork achiev es optimal throughput, while maintaining transmission ratios below the CTR for achieving energy balancing. Note that the protocol utilizes the TS- OOK’ s elasticity in the inter-symbol spacing, allowing mul- tiple nanonodes to transmit their packets in parallel without inducing collisions. Rikhtegar et al. [151] present the Energy Efﬁcient Wire- less Nano Sensor Network MAC (EEWNSN-MAC), a MAC protocol for mobile multi-hop THz band nanonetworks. They assume a network comprised of nanonodes moving randomly at a constant speed, as well as static nanorouters and a nano-micro interface. EEWNSN-MA C is di vided into three steps: i) handshaking-based selection of a cluster head (i.e., nanorouter); ii) TDMA-based scheduling phase in which a nanorouter schedules the transmission times for the nanonodes in its cluster; iii) the data transmission phase in which the nanonodes send their packets to the nanorouters, followed by their aggregation and forwarding to the nano-micro interface. The authors in [152] propose EESR-MA C, an energy efﬁ- cient, scalable, and reliable MA C protocol for THz nanonet- works. In the protocol, the cluster head is ﬁrst deriv ed as the one that is equidistant from other nanonodes in the cluster . Then, the classic TDMA approach is used for inter and intra- cluster communication, where the master node is assumed to transmit the schedule to the other nanonodes. The approach is presented here for completeness purposes, as it has not been ev aluated nor benchmarked against other approaches (and is therefore omitted in T able VII). A slotted CSMA/CA based MA C protocol (CSMA-MA C) is proposed for energy harvesting nanonodes in [153]. The protocol assumes a coordinator node periodically transmitting beacon packets containing the super-frame structure. A nanon- ode that wants to transmit receiv es the beacon, synchronizes itself to the super-frame structure, followed by transmitting its data using slotted CSMA/CA channel access mechanism. The protocol has been benchmarked in terms of achiev able data-rate against a simple round robin approach. C. Pr otocols for On-Chip Communication The authors in [98] provide a context analysis of MAC protocols for on-chip communication. The y ar gue that, from the link layer perspective, on-chip communication represents a unique scenario with respect to traditional wireless communi- cation. This is predominantly due to the fact that the topology of a network, chip layout, and characteristics of the building materials are static and known in adv ance [113]. Therefore, the on-chip wireless channel can be accurately characterized as quasi-deterministic from the link layer perspectiv e. As discussed before, the on-chip applications require v ery low and deterministic latenc y , high reliability , and very high throughput. As the secondary requirement, the energy con- sumption should be constrained to limit the heat dissipation on the chip. This has to an extent been recognized in the existing literature. Note that these works do not consider the usage of THz frequencies for communication, but instead focus on the sub-THz (i.e., 30-300 GHz) band. Nonetheless, we will outline them here due to the intrinsic similarities between the THz and mmW av e bands in the context on on-chip nanocommunication. T wo ﬂav ors of a token-passing (i.e., dynamic TDMA-based) MA C protocol for on-chip wireless communication are pro- posed in [154]. In token-passing, the channel access is based on the possession of a token which circulates between nodes in a round-robin fashion to ensure fairness. The duration of token possession is in [154] determined on predicted estimates of communication demands for dif ferent nanonodes. The protocol is e v aluated in terms of average data-rate, ener gy ef ﬁciency , and latency of packet delivery . Mestres et al. [108] propose the { Broadcast, Reliability , Sensing } protocol (BRS-MAC). BRS-MAC combines the CS- MA/CA and CSMA/CD mechanisms by utilizing preamble- based collision detection. Speciﬁcally , as the ﬁrst en visioned step and only if the channel is sensed idle, the sender trans- mits a preamble (otherwise the node backs off). Nanonodes that correctly receiv ed it remain silent for the rest of the transmission, while the ones that detected a collision respond with a Negativ e A CKno wledgment (N A CK). If the N A CK was received, the original sender cancels the transmission and backs off, while the other nodes discard the preamble. Adversely , the sender transmits the rest of the packet. The protocol has been ev aluated in terms of achiev able throughput and delivery latency . The authors in [155] argue that there is a need for recon- ﬁgurable wireless links for optimizing the utilization of the on-chip channel bandwidth. Grounded on that observation, they propose a dynamic MA C protocol by integrating the CSMA and token-based mechanisms. The proposed protocol utilizes the token-passing mechanism in case of high trafﬁc loads. When the trafﬁc loads are low , token passing becomes energy inefﬁcient. Therefore, in such cases the dynamic MAC unit of the protocol switches the operation to a CSMA-based mechanism in which the consumed energy is due to valid transmission only . The protocol is e v aluated in terms of achie v- able throughput, energy efﬁciency , and protocol overhead (i.e., power , area, and delay characteristics).  T able IV pro vides a summary of the above-discussed link layer protocols for THz band nanonetworks. As visible in the table, there are se veral open challenges pertaining to the link layer protocols for nanonetworks. First, the majority of the current protocols deal with the MAC sub-layer , with only the PHLAME and TCN protocols additionally proposing a LLC sub-layer mechanism. This implies that the majority of current protocols are not tailor -made for applications that 17 T ABLE IV: Summary of link layer protocols Protocol Sub-layer Distinct Featur es Potential A pplications Evaluation Metrics Akkari et al. [144] MA C - CSMA-based with hard deadline - continuous communication - software-deﬁned metamaterials - wireless robotic materials - timely-delivery ratio 1 Alsheikh et al. [145] MAC - blind transmission - continuous communication - grid constellation - software-deﬁned metamaterials (gen. I) - wireless robotic materials - consumed energy - collision probability - transmission delay - network throughput PHLAME [114] MA C, LLC - handshake-based - interference minimization - retransmissions possible - software-deﬁned metamaterials (gen. I) - wireless robotic materials - body-centric communication - consumed energy - collision probability - network throughput DRIH-MA C [146] MA C - receiver -initiated communication - schedule-based transmissions - wireless robotic materials - transmission delay - capacity utilization - energy utilization TCN [147] MA C, LLC - scheduled transmissions (TDMA) - acknowledgement-based - retransmissions possible - body-centric communication - energy per bit - collision probability - energy consumption Xia et al. [148] MA C - handshake-based - receiver -initiated synchronization - sliding window ﬂow control - wireless robotic materials - packet discard probability - network throughput Smart-MA C [149] MA C - default MAC in NanoSim - handshake and backof f-based - wireless robotic materials - packet-loss ratio - scalability APIS [150] MA C - designed for gatew ay to sink links - based on trafﬁc and channel sensing - software-deﬁned metamaterials - wireless robotic materials - achiev able throughput - energy efﬁciency - delivery reliability W ang et al. [115] MA C - TDMA-based - continuous communication - fairness-oriented - software-deﬁned metamaterials (gen. I) - wireless robotic materials - single-user throughput - achiev able data-rate EEWNSN-MA C [151] MAC - TDMA with clustering - mobility and multi-hopping - software-deﬁned metamaterials (gen. I) - wireless robotic materials - energy consumption - scalability - packet-loss ratio EESR-MA C [152] MA C - TDMA with clustering - software-deﬁned metamaterials (gen. I) - wireless robotic materials - none CSMA-MA C [153] MA C - beaconing-based -CSMA/CA channel access - body-centric communication - wireless robotic materials - achiev able throughput Mansoor et al. [154] MA C, LLC - token-passing-based - based on trafﬁc estimates - on-chip communication - average data-rate - energy efﬁciency - transmission delay BRS-MA C [108] MA C, LLC - CSMA and NA CK-based - preamble-based collision detection - on-chip communication - achiev able throughput - transmission delay Dynamic MAC [155] MA C, LLC - combines CSMA and token-passing - based on expected trafﬁc loads - on-chip communication - achiev able throughput - energy efﬁciency - protocol overhead require high communication reliability (e.g., body-centric and on-chip communication). One of the open research questions is to improve the reliability of THz band nanocommunication on the link layer . As an example, PHLAME supports packet repetitions as means for improving the nanonetwork reliability (i.e., Pack et Reception Rate (PRR)). Howe ver , when and how many repetitions should be utilized is still unclear . Such a decision could potentially be based on the current ener gy le v els of the nanonodes, their distances, and/or the amounts of trafﬁc, as in more detail discussed in [157]. Second, the majority of the existing protocols do not opti- mize for the latency of data delivery . The only protocol that goes in this direction is Akkari et al. [144], where a hard deadline on data deliv ery is imposed, howe ver the minimiza- tion of deliv ery latency is not attempted. Optimization of link layer protocols in terms of latency is required for se veral of the envisioned applications (e.g., the second generation of software-deﬁned metamaterials, on-chip communication). In addition and to the best of our knowledge, protocols aiming at jointly optimizing throughput and latency , which is required for on-chip nanocommunication, are currently missing. Third, apart for EEWNSN-MA C, none of the protocols explicitly accounts for the fact that nanonodes can be mobile. Even the EEWNSN-MA C protocol, giv en that it hierarchi- cal, requires the selection of cluster heads, which is known to yields unsatisfactory performance in scenarios with high mobility . In mobility scenarios, are handshake and clustering- based hierarchical protocols are generally expected to yield poor performance. This is further accentuated by the fact that, due to harvesting, a certain amount of time will usually have to pass between the handshake and data transmission. In high mobility scenarios, the optimal strategy for data transmission could be to just send data when there is data to send and enough energy for transmission. Nonetheless, such strategies hav e yet to be inv estigated. In T able IV, we have listed the application domains that could potentially be supported by a given protocol. There are seemingly no link layer protocols explicitly targeting on- chip THz band communication. W e have outlined the most promising candidates from the mmW ave band, which is indeed similar to the signal propagation in THz frequencies. Y et, propagation is not the same for these two bands, mostly due to the fact that THz signals attenuate faster and resonate with water , in contrast to mmW av e signals resonating with oxygen molecules. Hence, the applicability of the listed protocols for on-chip THz nanocommunication is yet to be ev aluated. 1 Percentage of packets successfully delivered before the deadline. 18 Finally , all of the existing protocols have been ev aluated ei- ther analytically or by means of simulation. Their performance results are potentially not accurate, as they have not been de- riv ed with a v ery high le vel of realism. For example, the energy consumption of a nanonode’ s communication system is in the ev aluations of all protocols attributed to either transmission of reception, while idling energy has been fully neglected. Evaluation results with higher levels of realism are certainly needed. In addition, the metrics used in the ev aluations are non-standardized and non-exhaustiv e, as sho wn in the table. Hence, various performance insights are currently lacking. For example, one of the primary requirements for many of the en- visioned applications is scalability . Nonetheless, the scalability of link layer protocols has been ev aluated only in EEWNSN- MA C and e v en there the conclusion is that “ EEWNSN-MA C is potentially a scalable pr otocol ”. Due to the fact that the ev aluation metrics are currently non-standardized and non- exhausti ve, comprehensive comparison of protocols for differ- ent application scenarios is at the moment infeasible. Giv en that THz nanocommunication is in many aspects challenging, ev en minor improv ements in the protocol design could yield high beneﬁts. Thus, comprehensive protocol benchmarking is certainly a promising research direction of high priority . V I . P H Y S I C A L L A Y E R The physical layer deﬁnes the means of transmitting raw bits over a physical link interconnecting two nodes. In the speciﬁc case of wireless communication, the physical layer is concerned with the modulation, the coding, error control, and other methods that determine the data rate and error rate of the solution, as well as its area and power . The scenario of THz nanocommunication has a unique blend of constraints and requirements that greatly impacts the physical layer and prevents the use of well-established techniques. The nanoscale dimension imposes v ery stringent restrictions on the av ailable resources (i.e., area, energy , mem- ory) that, despite being dependent on the particular application context, suggest the use of simple and ultra-efﬁcient modula- tions and coding. This is especially limiting in intermittent computing applications where devices are powered via energy harvesting. This is evidently impacting the devices’ av ailable energy , b ut also causing intermittency in de vices’ operation, posing an additional challenge in regard to their reliability . The THz dimension of the scenario affects the ph ysical layer of the design as well. The main reason is technological, as mature THz circuits and systems for communication are yet to come, although the community is making signiﬁcant leaps forward [19], [158]–[161]. Moreov er , the THz channel intro- duces the effect of molecular absorption which, for increasing distances, becomes another impairment for communication. Overall, the existing proposals for THz nanocommunication hav e embraced simplicity as one of the main design drivers. As we discuss next in Section VI-A, on-chip communication works mostly advocate for fast Continuous-W ave (CW) On- Off Ke ying (OOK) to avoid power-hungry circuits and mini- mize signal processing delay , whereas other applications need to simplify the physical layer further via Impulse Radio (IR)- like techniques. In the latter case, modulations rely on the THz Physical Layer Pr otocols Pulse-based Modulations (Sect. VI-B) TS-OOK [165] PHLAME [114] Shrestha et al. [166] SRH-TSOOK [167] ASRH-TSOOK [168] V avouris et al. [169] Zarepour et al. [170] Han et al. [171] Pulse-based Coding (Sect. VI-C) Jornet et al. [172] MEC [173] Chi et al. [174] SBN [175] DS-OOK [176] Beaming and Detection (Sect. VI-D) Hosseininejad et al. [177] Lin et al. [178] Cid-Fuentes et al. [179] Singh et al. [180] Gupta et al. [181] Iqbal et al. [182] ; Figure 8: Classiﬁcation of physical layer protocol for THz nanocommunication transmission of femtosecond-long pulses and, not surprisingly , adopt cross-layer strategies in an attempt to further simplify the protocol stack. As shown in Figure 8, we outline the main alternativ es of such pulse-based modulations and coding techniques in Sections VI-B and VI-C, respecti v ely . Finally , we analyze recent proposals for simpliﬁed beaming, detection, and synchronization in Section VI-D. A. Continuous-W ave (CW) vs. Impulse Radio (IR) W ireless communication networks hav e been established and hav e grown dominated by CW technologies. This means that the modulation is based on the manipulation of the amplitude, phase, or other characteristics of a continuous carrier wa ve at the desired frequency . T echnology scaling has allowed to increase the carrier frequenc y in the quest for higher bandwidths and device miniaturization. Higher bandwidths are naturally attained as the carrier frequency increases. Minia- turization is achiev ed because the antenna and other passiv e elements, which are typically the largest components within RF transceiv ers, scale with the inv erse of the carrier frequency . As a simple but illustrativ e example, the resonance frequency f R of a dipole antenna of length L surrounded by air is given by f R = c 0 2 L where c 0 is the speed of light. The energy efﬁciency of CW transceiv ers in the THz band depends on several factors such as the carrier frequency , the choice of modulation, or the transmission range. Based on an analysis of recent transceivers from 0.06 to 0.43 THz [159], [162]–[164], Figure 9 shows how CW transceivers are reaching 10+ Gb/s speeds with around 1 pJ/bit for high- rate applications at the on-chip, off-chip, and indoor scales. Such trend is e xpected to continue in the THz band, where signiﬁcant ef forts are de voted to ﬁlling the so-called THz gap [19], [158]–[161]. The existing scaling tendencies in CW transceivers are good news for on-chip communication applications as multiproces- sors demand very high transmission speeds. Since the energy supply in computer systems is typically sustained, classical CW modulation techniques can be used. Nev ertheless, as described in Section III-D, the scenario demands ultra-low latency and low power consumption. Due to these factors and to the relativ ely immature state of THz technology , high- order modulations or techniques requiring signiﬁcant signal processing are discouraged. Instead, most proposals advocate for simple modulations such as OOK and non-coherent (i.e., amplitude) detection [162], [183]. The OOK modulation con- sists in transmitting silence when the symbol is ’0’ and the 19 1 10 100 Data Rate (Gb/s) 10 100 1000 Power Consumption (mW) WPAN On-chip Predictions 1 pJ/bit 2 pJ/bit 10 pJ/bit Figure 9: Energy efﬁcienc y of sub-THz and THz transceivers (from 0.06 to 0.43 THz) for short-range high-rate wireless applications. Each data point indicates the power and data rate of a single transceiv er prototype from the literature for W ireless Personal Area Networks (WP AN, blue squares) and on-chip communication applications (red circles). Purple diamonds correspond to theoretical predictions made in the literature for future transceivers. Finally , straight lines represent the frontiers of energy ef ﬁciency of 1, 2, and 10 pJ per transmitted bit, so that transcei vers located abov e (below) each line are less (more) efﬁcient than indicated by the frontier . Data extracted from [164]. carrier wa ve when the symbol is ’1’. Such modulation can be achiev ed by simply connecting the stream of bits to the circuit that generates the carrier wav e. This av oids the use of power -hungry circuits such as those needed for coherent (i.e., phase) modulation and detection. It has been thus shown that OOK can be 1.5X and 2.5X more energy efﬁcient than BPSK and QPSK in on-chip en vironments [162]. The do wnside of using low-order modulations is that, to scale the transmission rates, one may need to resort to multiple carriers to combat dispersion (i.e., a single carrier modulated at ultra-high speeds leads to an ultra-high bandwidth signal very sensiti ve to multipath and delay spread). Notably , Han et al. proposes a multi-carrier modulation that adapts to the frequency-selecti v e molecular absorption effects of the THz band [184]. In applications where energy av ailability is intermittent and not guaranteed, CW techniques cannot be used due to the cost of generating and using a carrier signal. Therefore, IR- like modulations where the information is encoded in short pulses instead of a carrier wa ve have been proposed instead. In the work by Zarepour et al. [185], carrier-less pulse-based OOK, Binary Phase Shift K eying (BPSK), Pulse Amplitude and Position Modulations (P AM and PPM, respecti vely) were compared in order to assess their ﬁtness for IoNT applications. Using analytical models, it was concluded that although BPSK is relatively more complex in terms of decoding logic, it is the most efﬁcient and reliable among all the contenders. OOK and PPM are simpler , but less reliable and efﬁcient than BPSK. The analysis discouraged the use of P AM due to its low performance and efﬁcienc y . Zarepour et al. re visited a widely re garded trade-of f between complexity and performance. BPSK and, by extension, other signaling schemes such as T ransmitted Reference (TR) [186], are preferable ov er OOK but may not be af fordable in extreme scenarios. In such cases, in fact, even con ventional OOK may be prohibitiv e. As a result and as we see next, the physical layer research has continued to push the ef ﬁciency and simplicity boundaries of pulse-based modulations. User 1 Pu ls e Si le nc e OOK 1 0 1 0 0 1 1 1 1 0 1 1 1 0 0 0 1 1 TS - OOK 1 1 0 1 1 1 0 PHLA ME or Sh retsa et al . 1 1 0 1 1 1 0 1 User 2 Pulse Sil ence AS RH - TSOO K 1 1 0 1 1 0 1 V av ouri s et al . 00 0 01 11 00 00 10 Figure 10: Comparison between the different variants of pulse-based OOK found in the literature [114], [165], [166], [168], [169]. B. Pulse-based Modulations One of the ﬁrst works to discuss modulations suitable for nanoscale wireless communication w as [165]. The proposed scheme is named T ime-Spread On-Off K eying (TS-OOK) and is a pulse-based modulation. The main characteristics are that (i) pulses are around 100-fs long, thus leading to bandwidths in the THz range, and that (ii) the separation between pulses is much larger than the duration of the pulse. This scheme retains the simplicity of con ventional OOK and, by having such a large separation between pulses, it is compatible with applications where energy is v ery limited or needs to be harvested. Moreover , by knowing the time between pulses, synchronization is only needed at the preamble and can be kept throughout the communication. The work in [165], besides proposing the modulation, conﬁrms that the achiev able capacity is in the order of Tbps and also opens the door to simple multi-user approaches that exploit the long time between pulses to interlea ve other communication. The authors provide an interference model that, in subsequent works, hav e been validated experimentally [187]. The seminal work by Jornet et al. has been followed by sev eral variants that optimize or particularize TS-OOK for different scenarios as graphically summarized in Figure 10. For instance, in [114], [166] the authors tackle one of the weaknesses of TS-OOK: if multiple users transmit with the same rate and collide in one pulse, they are bound to collide in all pulses. In the multi-user scheme proposed in [114], referred to as Rate Division Multiple Access (RDMA), users are assigned co-prime transmission rates during handshake to minimize interference at a reduced cost. In [166], the RDMA scheme is generalized for both ad hoc and infrastructure- based networks and the choice of prime numbers is further justiﬁed. Later , Mabed et al. argued that RDMA leads to rate imbalance as users are assigned different effecti ve rates. T o ov ercome this issue, they proposed to employ pseudo-random time-hopping sequences to determine the time between pulses that, on average, would yield similar rate for all users [167] or capable of adjusting the rate to the user needs [168]. Finally , Singh et al. propose a completely dif ferent approach to accommodate multiple users, which is to combine OOK with 20 sender-distinguishing direct sequence codes [176]. In this case, energy ef ﬁciency is sacriﬁced to achiev e higher performance. Another TS-OOK variant in the literature is [169]. In this case, the authors aim to maximize ener gy ef ﬁciency and, to that end, propose to combine TS-OOK with PPM. The approach consists of the modulation of a symbol as the time between pulses, which is at all times much larger than the pulse duration. It is demonstrated that when increasing the symbol order , multiple bits can be encoded as a silence between pulses, therefore improving the energy efﬁciency at the cost of a de gradation of the data rate. The PPM v ariant has also been combined with time-hopping in [188], where a thorough ev aluation is carried out assuming non-coherent detection and multiple modulation orders. Finally , it is worth mentioning proposals that also adapt to the particularities and ne w features of wireless communication in the THz band. On the one hand, Zarepour et al. propose the use of frequency-hopping as a means of overcoming the problem of dynamic molecular absorption in composition- varying channels [170]. The blind use of frequency-hopping eliminates the need for channel state observation, simplifying the modulation, while still ensuring that the transmission will succeed with a giv en probability . This is opposed to [171], which assumes static channel composition and proposes to estimate distance between transmitter and receiver to select the most appropriate w av eforms or frequencies for transmission. On the other hand, we also highlight the work of Lin et al. which hinges on the use of graphene-based directional agile antennas in the THz band. More speciﬁcally , they propose to use beam hopping to switch among spatial channels during transmission [178], a technique that can further reduce inter- ference among users in TS-OOK scenarios. C. Pulse-based Coding Coding to reduce power consumption and interference with- out increasing the transceiv er complexity has been another hot topic in nanocommunication research. Jornet et al. ﬁrst proposed the use of low-weight coding together with TS- OOK [172], [189]. Rather than utilizing channel codes to detect and correct transmission errors, this simple mechanism exploits silences to sav e power and mitigate interference without reducing the transmission rate of each individual user . It was later observed in [190] that minimizing only the av- erage weight does not ensure minimum energy . Follo wing this argument, the authors deriv e optimal codebooks that minimize the energy of transmissions for arbitrary codeword lengths. Further , Kocaoglu et al. extend the discussion to account for arbitrary input probability distributions and keeping the codew ord length unconstrained, arguing that minimum energy coding with high reliability is achiev ed in all cases [173]. Later , the authors in [174] add the property of preﬁx freedom and the constraint of maximum a verage codeword length to the problem of minimum energy coding. Preﬁx-free codes ensure that no codeword is contained within any other code- word, allowing instantaneous decoding of information. The concept of simple block nanocodes is applied in [175] to add reliability with very small cost in nanonetworks. The same authors provide a comprehensi ve comparison between the different proposed coding schemes in [191], e valuating energy efﬁcienc y , bandwidth expansion, robustness, and interference. Finally , we highlight the recent w ork by Y ao et al. [192], which goes beyond existing forward error correction strategies and adopts a hybrid mechanism suitable for energy harvesting. Their proposed error control strategy is compatible with lo w- error codes, but incorporates probing packets. Before starting data transmission, the source sends probing packets. The receiv ers then acknowledge the probe only if they anticipate to hav e enough ener gy to recei ve the data packets; otherwise, they remain silent. This way , data packets are not sent if receivers are in a state of low energy , which is PHY -layer dependent. D. Beaming and Detection Beaming at the transmitter side (both beam switching and beam forming) and detection at the receiver side are two func- tions that deserve attention due to the highly limited resources and peculiar requirements of THz nanocommunication. While THz band communication is moving towards direc- tiv e antennas and thus will likely require ef ﬁcient beaming methods, the nanocommunication scenario discourages its use unless simple and eff ectiv e methods are conceiv ed. In this direction, recent works have discussed the role of graphene antennas. Graphene not only allo ws to miniaturize antennas, but also confers them with ultraf ast reconﬁgurability achiev- able by simply changing the electrostatic voltage applied to the antenna. This has led to proposals where both the beam direction and frequency of resonance can be controlled with very simple approaches [19]. Leveraging these features, Hosseininejad et al. [177] propose a programmable PHY interface to graphene antennas to expose such beam-switching and frequency tunability to upper layers. As an example, such a controller could easily implement the bit-le vel beam- switching [178] to implement beam multiplexing methods compatible with TS-OOK and the interference mitigation techniques discussed above. At the receiv er side, simple means of detection are crucial to ensure the viability of nanocommunication. Cid-Fuentes et al. implement a low-complexity Continuous-T ime Moving A verage (CTMA) with a single low-pass ﬁlter and a peak detector [179]. The ev aluations contained therein demonstrate the potential for Tbps detection with relaxed synchronization requirements. In [181], an iterative process employing an array of time-delayed CTMA detectors is proposed to achiev e joint detection and synchronization for TS-OOK communication. A similar architecture is proposed in [180] that pro vides an estimation of time-of-arriv al for time-hopping PPM modula- tion. Finally , the work by Iqbal et al. is worth highlighting as it proposes a simple modulation mode detection and clas- siﬁcation for intelligent nanonetworks where transmitters may switch between modulations type and order [182]. Note that the approaches discussed above hav e not been included in T ables V and VII, as they are not physical layer protocols, but proposals for hardware to support such protocols.  T able V provides a summary of the above-discussed pro- tocols for the physical layer in THz nanonetworks. It can be 21 T ABLE V: Summary of physical layer protocols Protocol Functionality Distinct Featur es Potential Applications Evaluation Metrics TS-OOK [165] Modulation - pulse-based - time spread 100-fs pulses - sync only in preamble - body-centric communication - wireless robotic materials - energy consumption - user inf. rate - aggregated inf. rate PHLAME [114] Modulation - TS-OOK with co-prime rates - minimizes collisions - body-centric communication - wireless robotic materials - energy consumption - collision probability - network throughput Shrestha et al. [166] Modulation - TS-OOK with co-prime rates - enhanced & generalized co-prime generation - body-centric communication - wireless robotic materials - collision probability - aggregated inf. rate SRH-TSOOK [167] Modulation - based on TS-OOK - random time between pulses - uniform average rate across users - body-centric communication - wireless robotic materials - collision probability - packet loss - network throughput ASRH-TSOOK [168] Modulation - based on TS-OOK - random time between pulses - adaptive rate per user - body-centric communication - wireless robotic materials - collision probability - packet loss V av ouris et al. [169] Modulation - PPM with time-spread pulses - extreme energy efﬁciency - body-centric communication - wireless robotic materials - energy consumption - information rate Zarepour et al. [170] Modulation - frequency hopping to av oid absorption peaks - body-centric communication - wireless robotic materials - signal-to-noise ratio - bit error rate - capacity Multi-band OOK [162] Modulation - continuous-wave OOK in multiple bands - high rates with simple transceiv ers - on-chip communication - energy consumption - information rate - silicon area D AMC [184] Modulation - continuous-wave multi-carrier - bands chosen depending on distance - on-chip communication - information rate Han et al. [171] Modulation - pulse-based version of D AMC - waveforms chosen based on distance - on-chip communication - software-deﬁned metamaterials - SINR - bit error rate - throughput Jornet et al. [172] Coding - low-weight channel coding - minimizes energy in TS-OOK - body-centric communication -wireless robotic materials - information rate - codeword error rate MEC [173] Coding - minimum energy channel coding - assumes multi-carrier OOK - on-chip communication - software-deﬁned metamaterials - energy consumption - transmission rate - error probability Chi et al. [174] Coding - minimum energy coding - preﬁx-free codes - body-centric communication - wireless robotic materials - energy consumption SBN [175] Coding - simple block codes - efﬁciency-reliability trade-of f - body-centric communication - wireless robotic materials - bit error rate - energy efﬁciency DS-OOK [176] Coding - direct sequence + OOK - on-chip communication - software-deﬁned metamaterials - multi-user interference - bit error rate observed how the physical layer is well-researched from a the- oretical and simulated perspectiv e. Howe v er , the main hurdle for their realization is the actual circuit implementation of the analog front-end of the transceiv er . THz signal generation with compact and ef ﬁcient means remains as a huge open challenge, especially in the case of the hundred-femtosecond-long pulses assumed in most of the works in the ﬁeld [160]. The literature mentions three main signal source technolo- gies for THz nanocommunications. One of them is photocon- ductiv e technology . For instance, pulse-based photoconductive sources might pro vide the required signal with high po wer and, as such, hav e been proposed for the experimental testbeds of graphene antennas [193]. Howe ver , these sources depend on a bulk y laser to excite the photo-carriers that turn into the THz signal and, therefore, are impractical for nanocommunications. A second alternativ e are electronic sources [48] that generate THz signals from the upcon version of lower frequency ones. Howe v er , these sources generally cannot provide sufﬁcient power with a compact form factor , although recent advances in nanoplasma switches may disprove that tenet [194]. A very promising technology in this ﬁeld is, again, graphene plasmonics. Graphene supports the propagation of tunable plasmons in the THz band, leading to unprecedented minia- turization and reconﬁgurability opportunities when operating in this frequency band. These properties have been studied when using graphene transistors as very compact THz sig- nal sources [23] exploiting the Dyakono v-Shur instability or also as direct modulators, translating changes in electrostatic biasing voltage into modulated plasmons [195]. Graphene antennas, as discussed above, can not only be miniaturized down to a few micrometers and still resonate in the THz band, but also deliv er joint frequency-beam reconﬁgurability with unprecedented simplicity [19], [177]. From the perspecti ve of the recei ver , the use of non-coherent detectors and CTMA approaches relax the synchronization requirements. Howe v er , synchronization keeps being the main challenge in impulse radio in general, and in THz nanocom- munication in particular . T o reach the promised Tbps barrier , sampling needs to occur at potentially very high speeds any- way , which goes against the simplicity and efﬁcienc y demands of most en visaged applications. V I I . C H A N N E L M O D E L I N G Channel characterization and modeling captures the changes that the electromagnetic w av es suffer as they propagate through a medium until reaching the receiver . In general, comprehensiv e models incorporate all possible sources of losses (e.g., spreading, blocking, dielectric losses), dispersion (e.g., due to multipath), and noise (e.g., thermal noise and interferences). By accounting for all these effects, channel modeling provides the physical layer with the necessary 22 considerations for the design of appropriate modulations and coding schemes that fulﬁll the application requirements. Channel modeling is critical for the dev elopment of THz band nanocommunication due to the important differences of THz propagation with respect to micro wa ve or optical propagation. The most striking peculiarity of THz channels is the appearance of molecular absorption effects, which create peaks of attenuation whose depth and frequency depend on the transmission distance and molecular composition of the medium, respectively . These effects limit the practicable band- width and, as we have seen in the previous section, may lead to the use of multi-carrier modulations for high-throughput applications. Another peculiarity of THz propag ation is that materials that were effecti vely transparent and ﬂat at the microw av e or ev en mmW ave regime start becoming lossy and producing rough scattering upon reﬂection. More details on these impairments are giv en in Section VII-A. Obviously , the propagation channel is highly dependent on the actual application scenario. In the follo wing subsections, we analyze the existing channel modeling efforts in two of the promising directions for THz nanocommunication. In Section VII-B, we revie w the w orks that characterize the channel within the body-centric applications. In Section VII-C, we discuss the attempts to model the THz channel within computing packages for on-chip communications. A summary of the papers discussed in the section is given in T able VI. A. THz Pr opagation Models In the THz band, phenomena that are generally ne glected become signiﬁcant as the wavelength reaches dimensions commensurate to the molecules found in the medium or the tiny irregularities of the surfaces upon which the wav es may reﬂect. For instance, the pioneering work by Piesie wicz et al. discusses how the resulting molecular absorption [196] could impair communication in these frequencies. Later , scattering produced by certain particles suspended on the environment was factored in [197]. Thus, propagation models in the THz band need to account for these effects. In [197], the path loss for Line-of-Sight (LoS) propagation is giv en by: A LoS ( f ) = A spr ( f ) A mol ( f ) A sca ( f ) , (1) where, on top of the typical spreading loss A spr ( f ) , we have the molecular absoprtion A mol ( f ) and the particle scattering loss A sca ( f ) . If waves reﬂect on rough surfaces, additional scattering would need to be accounted for [198]. Moreov er , in [199], the noise temperature T noise is modeled as: T noise = T sy s + T mol + T other , (2) where, on top of the typical system electronic noise T sy s , noise caused by molecular absorption T mol is considered. T other refers to any additional noise source. Molecular absorption is the process by which part of the wa ve energy excites molecules found along its path. This phenomenon occurs often in the atmosphere as many of the molecules comprised therein resonate in the THz re gime. From the communications channel perspecti ve, absorption manifests as (i) a frequency-selectiv e attenuation and (ii) added noise Figure 11: Molecular absorption for 1 cm (blue) and 10 cm (red) in a standard atmosphere as deﬁned by the International Organization for Standardization (ISO) in the standard ISO 2533:1975. The blue and red backgrounds indicate the av ailable bandwidth for distances of 1 cm and 10 cm, respecti vely , where available bandwidth is deﬁned as the frequency band where the attenuation caused by molecular absorption is below 10 dB. from the molecules residual radiation. Jornet et al. [199] were among the ﬁrst to incorporate a model of the molecular absorption and noise into a complete THz channel model. On the one hand, the attenuation A mol ( f ) caused by this molecular absorption is modeled as [199]: A mol ( f , d ) = e k A ( f ) d , (3) where k A ( f ) is the medium absorption coef ﬁcient that depends on the molecular composition of the medium, f is frequency , and d is the transmission distance. The result, illustrated in Figure 11, is a frequency-selecti ve attenuation that scales with the distance and can effecti vely limit the channel bandwidth. On the other hand, the molecular noise is modeled as [199]: T mol ( f , d ) = T 0 (1 − e − k A ( f ) d ) , (4) where T 0 is the reference temperature. Based on these models, Jornet et al. [199] studied the impact of absorption in terms of channel capacity for dif ferent medium compositions and distances. Later , Javed et al. [200] modeled the molecular absorption as a log-normally distributed attenuation (similar to ho w shadowing is generally accounted for) within a con- ventional log-distance path loss model. In [201], the authors focused on the most prominent transparency windows, i.e., bands where molecular attenuation is low , and performed a thorough capacity and throughput analysis both with and without energy constraints. Finally , Llatser et al. analyzed the effect of molecular absorption in the time domain and targeting short distances [202]. There, it was conﬁrmed that molecular absorption and its dispersi ve effects are generally ne gligible up to a few centimeters, as exempliﬁed in Figure 11. Besides molecular absorption, diffuse scattering caused by particles or rough surfaces commensurate to the THz w ave- length are also potential impairments in the THz band. These effects are also frequency-selectiv e and, therefore, hav e an impact upon the response of the channel. Kokk oniemi et al. provide a comprehensiv e channel model in both the time and frequency domains which accounts for the aforementioned effects, pro ving that they might not be negligible at certain 23 distances [197]. In the case of particle scattering, the attenu- ation A sca ( f ) is giv en by A sca ( f , d ) = e k S ( f ) d , (5) where the frequency dependence is modeled through the particle scattering coefﬁcient k S ( f ) = P j N j s σ j s [197]. This coefﬁcient requires kno wledge on the density of particles N and of the scattering cross section σ of each type of particle j . In the case of rough surface scattering, several measurement campaigns ha ve been carried out to analyze the response of materials such as wood, plaster , concrete, plastic, glass, or metal [203]–[205] and incorporate it into the channel model. The relati vely short wav elength of THz wa ves, besides leading to the impairments discussed above, also suggests the use of ray tracing techniques to develop comprehensi ve channel models ev en in nanocommunication scenarios. The work in [206] also argues that ray tracing could be particularly appropriate due to the high-gain antennas expected in THz applications. Moreover , it provides a comprehensiv e ray-based modeling methodology and ex empliﬁes its use in indoor chan- nel characterization. The methodology has been later extended to the particular case of THz wireless communications with graphene reﬂectarray antennas [207]. The same authors also discuss hybrid methodologies combining ray tracing and full- wa ve simulations to account for all effects accurately while be- ing computationally affordable [208]. Furthermore, the authors of [209] propose to expose the design parameters of graphene antennas in order to facilitate the design space exploration of graphene-enabled wireless channels. Finally , and since most works in the ﬁeld assume the co-existence of multiplexed links either spatially (via beaming) or temporally (via pulse-based modulations), Petrov et al. also employ ray-based methods to ev aluate interference and therefore derive SINR and spectral efﬁcienc y metrics [210]. W ith the requirement of multiple transmissions, the authors also estimate the optimal distance between recei ving nodes to maximize the area capacity and conclude that interference becomes more critical than molec- ular absorption in these multi-user scenarios. B. Intra-body Channels W ireless communication within the human body presents many exciting applications in the body-centric communication domain in as much as it poses a signiﬁcant challenge from the perspective of the propagation of THz signals. From the modeling perspecti ve, Equation (1) still holds. Ho wev er , the terms related to spreading loss, molecular absorption loss, and scattering loss will v ary . F or instance, the human body is composed of multiple tissues such as skin, fat, or blood, each with their own response to THz radiation. W ireless propagation is impaired not only by the dielectric loss of each tissue, but also the transition between tissues, thereby affecting A spr . Moreover , the composition very much depends on the position of transmitter and recei ver within the body and of the patient itself, affecting A mol . Finally , the presence of cells and other particles impact on the scattering term A sca . Among the ﬁrst works on this regard, the authors in [211] studied the attenuation of fat in the THz band. The characteris- tics of the fat layer were extracted from characterization works in the optics domain, yielding an attenuation factor of around 30 dB/mm. Later , the authors extended the work and published a complete model in [212] containing blood, skin, and fat. The molecular absorption of those tissues is used to determine the system noise and, then, the channel capacity . Further, Elayan et al. also considered a multi-layered model in the frequency and time domains and studied the response when on-body and intra-body de vices communicate [213]. They concluded that up to 30% of the incident power from outside the body may be reﬂected back and that the result is symmetrical. Finally , Zhang et al. provide a detailed model of artiﬁcial skin and perform a capacity analysis in the THz band [214]. The impact of the multiple phenomena occurring in intra- body channels on the noise temperature has been also studied in depth. Existing models e volv ed from the initial model from [212] where it was assumed that noise caused by molec- ular absorption dominates and other potential sources can be neglected, thereby obtaining an expression similar to Equation (2). A more comprehensi ve discussion was included in [225], where the authors modeled signal-independent body radiation noise using Planck’ s law , as well as the signal-dependent molecular absorption noise. A comparison between the two validated the assumption of dominance of molecular absorp- tion noise. Later , Elayan et al. revisited the model to include, besides the two sources mentioned above, other components such as thermal noise at the transceiv er circuitry or Doppler- shift-induced noise [216]. Finally , in [226], noise models are combined with interference models to deriv e SINR metrics tow ards accurately determining the capacity and throughput achiev able in intra-body networks. It is therefore suggested that nanomachine density can be a factor as important as the composition of the intra-body channel in assessing the viability of the communication. W e ﬁnally describe papers that extend existing models to account for changes in the transmission medium. The work of Zarepour et al. is worth mentioning as it considers time- varying channels in the THz band [217]. The key takea way is that nanocommunication channels are not static: the tempera- ture, pressure, or molecular composition of the medium may vary over time. They provide as example the composition of the blood when breathing in, which is clearly different than when breathing out. A similar example is analyzed in [227], where a nanonetwork for lung monitoring is explored. In that case, respiration clearly changes the volume and composition of the lung, and the authors adapt their design to that circum- stance. Even further , all the works described in this section can be used to determine the attenuation caused by vegetation in plant monitoring nanosensor networks, which can also vary ov er time due to the effects of photosynthesis [85]. C. On-chip Communication Channels A channel model that takes into consideration the pecu- liarities of the chip-scale scenario is fundamental to ev aluate the available bandwidth and to properly allocate po wer . The enclosed nature of the chip package suggests that propagation losses may be small, but also that multipath ef fects may be present. Additionally , the multiple metalization layers and the 24 T ABLE VI: Summary of channel modelling works Reference Scope Domain Method Analyzed Features Evaluation Metrics [199] General Frequency Analytical Molecular absorption, noise Path loss, capacity [202] General Frequency , time Analytical Molecular absorption Practicable bandwidth, pulse width [197] General Frequency , time Analytical Molecular absorption, particle scatter- ing, rough surface scattering Path loss, delay spread, coherence bandwidth [204], [215] General Frequency Experimental Rough surface scattering, diffraction Path loss (scattering power , diffraction angle) [203] Indoor Frequency Experimental Rough surface scattering Scattering power , path loss [206], [207] Indoor Frequency , time Analytical LoS and NLoS propagation, graphene reﬂectarray impact Path loss, delay spread, coherence bandwidth, capacity [210] Indoor Frequency Numerical Blocking, Interference SIR, SINR [200] Indoor , intra-body Frequency Analytical Molecular absorption, noise (fat, blood, bone) Path loss, capacity [211], [212] Intra-body Frequency Numerical Absorption in blood, skin, fat P ath loss, noise temperature, capacity [213] Intra-body Frequency Analytical, numerical Discontinuities on multi-layer medium Reﬂected power , reﬂectance [216] Intra-body Frequency Analytical Intra-body noise sources Noise spectral density [217] Intra-body Frequency , time Numerical T ime v ariation of medium composition SNR, BER [218] Chip-to- chip Frequency , time Experimental Re verberation Path loss, coherence bandwidth [219] On-chip Frequency Numerical Molecular absorption, dielectric losses Coupling (S 21 ) [220] On-chip Frequency Numerical Losses Coupling (S 21 ) [221], [222] On-chip Frequenc y Analytical, numerical Losses, interference Path loss, capacity [223], [224] On-chip Frequenc y Experimental Antenna orientation, position Insertion loss (S 11 ), Coupling (S 21 ) [164] Off-chip (SDM) Frequency , time Numerical Propagation path, SDM geometry Path loss, delay spread Through-Silicon V ias (TSV) present in today’ s chips may further challenge propagation [228]. Fortunately , the scenario is unique in that all these elements are ﬁxed and known beforehand. Thus, the channel model will be virtually time- in v ariant and quasi-deterministic. Moreov er , molecular absorp- tion and diffuse scattering are not problems in this controlled en vironment [219], [229]. Therefore, only the spreading loss term of Equation (1) stands. The model, howe ver , has to take into consideration dielectric losses caused by the chip package materials and by transitions between the different layers found within the chip structure. In terms of noise, the most prominent source is thermal noise and, thus, con ventional models can be employed. It is worth noting that the processor circuitry does not introduce noise as communication and computation sub- systems operate at disjoint frequencies. Thus far , few works have explored the chip-scale wireless channel down to the nanoscale and in the THz band. The theory is well laid out [228] and a wide variety of works exist in larger en vironments. THz propagation has been in vestig ated in small and enclosed environments such as across a computer motherboard [230], or within the metallic encasement of a laptop computer [218]. The experimental results, up to 300 GHz, have conﬁrmed that such systems act as reverbera- tion chambers due to the metallic enclosure, leading to rather low path loss but very high delay spreads. Down to the chip le vel, most characterization efforts hav e thus far been limited to mmW av e frequencies [229]. Analytical models [231], simulation-based studies [232], and actual mea- surement campaigns [223], [224] have been conducted under different assumptions. In most cases, path loss is modeled by an empirical log-scale approximation resulting in a deriv ed the path loss e xponent, rather than using a bottom up approach with Equation (1). More recently , a few authors have tried to study propagation in sub-THz frequencies, i.e., up to around 140 GHz in [233] and 200 GHz in [220]. At the time of writing, howe ver , the only channel model at THz frequencies is that of Chen et al. [221], [222], which employs ray tracing within a stratiﬁed model of the chip structure to extract path loss and dispersion metrics. Regardless, the main issue of the works mentioned above is that free-space propagation in an unpackaged chip is assumed for simplicity . Such a model generally falls short of capturing the enclosed nature of realistic on-chip communication envi- ronments. Hence, recent efforts are starting to model realistic ﬂip-chip packages with lossy bulk silicon in the mmW ave and sub-THz spectrum [233]. The main conclusion is that the losses introduced by the silicon substrate prevent the package to act as a rev erberation chamber . Howe ver , the price to pay is an unwanted path loss in excess of several tens of dBs. When scaling to the THz band, the results will likely worsen due to the smaller ef fectiv e aperture of the antennas. For potential solutions we refer the reader to the next subsection.  Channel modeling in THz nanocommunication has been well-researched in theory and simulations. Actual measure- ments are more complex to achiev e due to the lack of mature measuring equipment and the difﬁculty of accessing the nanoscale with enough accuracy . First experimental results hav e been achiev ed that conﬁrm molecular absorption and scattering effects and, in the case of on-chip wireless commu- nications, results up to sub-THz frequencies start becoming av ailable. Howe v er , there is still a lar ge room for impro ving the current characterizations experimentally . In the intra-body networks, researchers ha ve identiﬁed di- electric losses and molecular absorption as huge sources of attenuation and noise. Moreov er , these can vary o ver time e ven assuming ﬁxed transmitter-recei ver positions. It is yet unclear, hence, how nanomachines will ov ercome these issues and how protocols will adapt to these very adverse and changing 25 conditions while still being bio-compatible. In the on-chip scenario, a channel model in the THz band with a realistic chip package is still largely missing. Moreov er , the problem of relatively high attenuation due to the losses within the silicon remains unsolved. In this respect, T imoneda et al. propose to exploit the controlled nature of the chip scenario to actually design the wireless channel without affecting the reliability of the digital circuits [234]. T o that end, the thickness of silicon and the thermal interface material are introduced as design parameters and optimized to minimize path loss while maintaining an acceptable delay spread. Due to the similarities between the on-chip scenario and the intra-SDM network scenario, some of the knowledge of the former can be reused in the latter . First explorations in this regard [164] indeed show that, internally , the mmW ave wire- less propagation paths within the SDMs suffer similar effects than in on-chip channels: low loss, wav e-guiding effects, and relativ ely high delay spread. The challenge here is to extend those models to the THz band and conﬁrm the viability of wireless communication within the metamaterials. V I I I . S I M U L A T I O N A N D E X P E R I M E N TA T I O N T O O L S There are se veral tools for simulating the beha viour of nanonetworks operating the THz frequencies. The pioneering simulator is NanoSim [149], [235], an event-based ns-3 mod- ule for modeling nanonetworks based on electromagnetic com- munications in the THz band. In NanoSim, a nanonetwork can be comprised of nanonodes, nanorouters, and nanointerfaces. Nanonodes are small and simple devices with very limited energy , computational, and storage capabilities. Nanorouters hav e resources lar ger than the nanonodes and they are en vi- sioned to processing data coming from nanonodes, as well as controlling their behavior through control messages. Nanoin- terfaces act as g atew ays between the nano- and macro-scale world. All of them can be either static or different mobility models can be employed according to the application require- ments (i.e., constant acceleration, constant velocity , random walk, random direction, and random way-points). Moreo ver , their basic functionalities can be customized to the demands of a given ev aluation scenario. The nanonetwork in NanoSim consists of the network, link, and physical layers. On the network layer , random and selectiv e ﬂooding routing strategies are supported. Link layer currently supports Transparent-MA C (i.e., simple forwarding from network layer to the physical interface) and Smart-MA C protocols. Moreov er , NanoSim pro- vides a TS-OOK-based physical layer with se veral adjustable parameters such as pulse duration, pulse transmission interval, and transmission duration. Finally , the radio channel can be modeled based on a cut-of f transmission distances between nanonodes. In addition, the selectivity of the THz channel in both frequency and time domains can optionally be introduced based on the spectrum-aware channel modeling from [236]. V oui vre [237] ([238], [239] in its preliminary version) is a C++ THz nano-wireless simulation library dev eloped as both an extension for Dynamic Physical Rendering Simu- lator (DPRSim) and as a standalone discrete event simula- tor . DPRSim [240] has been de veloped in the scope of the Claytronics project for supporting simulations with a large number (up to millions) of Claytronics micro-robots (also known as catoms ). Original catoms do not have wireless trans- mission capabilities, as they are en visioned to communicate only through physical contact. V ouivre introduces wireless communication capability to the catoms. In particular , it can be used for simulating the THz radio channel and its concurrent accesses by catoms. The THz radio channel is modeled by a continuous distance-dependent attenuation contribution increased by a certain noise v alue caused by the concurrent transmissions, with the noise value tak en from [189], [199]. In addition, transmission delay in combination with total attenuation ha ve been utilized for determining pack et reception probability . Moreo ver , V oui vre implements a TS-OOK-based physical layer , while the upper layers hav e not been imple- mented, apart from the standard CSMA/CA scheme combined with the Friss propagation model in 2.4 GHz frequency for “allowing ulterior studies of hybrid systems”. BitSimulator [241] is another simulator speciﬁcally target- ing THz band nanonetworks. BitSimulator is implemented in C++ and utilizes a discrete ev ent model. At the physical layer , BitSimulator implements the TS-OOK scheme with 100 fs long pulses and per-frame conﬁgurable parameter β . The link layer is not implemented, as it is assumed that multiple frames can be temporally multiplex ed and the nodes have the capabil- ity of tracking the transmissions intended for them. Network layer implementation supports no routing, ﬂooding-based, and Stateless Linear-path Routing (SLR) [138]. T wo discovery modes are available: i) static, where neighbors are stored for each node and calculated at the beginning of the simulation; ii) dynamic, where neighbors are not stored, but computed at periodic time instances. Hence, the support for simple mo- bility exists in BitSimulator . In terms of channel modeling, a simple communicationRange parameter is used for specifying achiev able transmission range. In addition, collisions between frames are determined based on propagation delay and TS- OOK-speciﬁc bit values of the packets concurrently received at each nanonode [165]. If the number of collisions is above a set threshold, the packet is discarded. T eraSim [242] is a newer alternativ e to the NanoSim simu- lator and also implemented on top of ns-3 . The simulator sup- ports simulations of both major types of application scenarios, i.e., THz nano- and macro-scale communication. In T eraSim, the THz radio channel for a nanoscale scenario is modeled by applying frequency and distance dependent spreading and absorption loss, accounting for wav eforms with realistic band- width. The simulator consists of a common channel module, separate physical and link layers for each scenario, and tw o assisting modules, namely , THz antenna module and ener gy harvesting module, originally designed for the macro- and nanoscale scenario, respectiv ely . T eraSim allows the user to select sev eral channel attributes such as bandwidth, number of samples of the frequenc y-selectiv e channel, and detection threshold (i.e., noise ﬂoor). Collisions between two packets occur if two pulses of different TS-OOK receptions ov erlap in time (packet dropping based on SINR) or if the pulse in recep- tion o verlaps with the pulse of an ongoing transmission (packet in reception is dropped). On the physical layer, T eraSim im- 26 plements the TS-OOK modulation and channel coding scheme with user adjustable pulse and symbol durations. On the link layer in the nanoscenario, T eraSim implements the ALOHA protocol in which the data is sent if there is enough energy for the transmission. The recei ver receiv es the data if there is enough energy for reception and replies with an acknowl- edgment (A CK) packet. The packet is dropped upon exceeding the maximum number of retransmissions. In addition, T eraSim implements the CSMA handshake protocol, in which the usual Ready-to-Send (R TS) / Clear-to-Send (CTS) packet exchange occurs. A CTS packet is transmitted if the receiv er has enough energy to complete the reception. Same as before, the packet is dropped upon exceeding the maximum number of retransmissions. Upper layer protocols are not speciﬁcally tuned to THz nanocommunication, but utilize av ailable ns- 3 modules (e.g., User Datagram Protocol (UDP) client/server , IPv4 addressing). Similarly , node mobility support is based on the av ailable ns-3 modules. An interesting feature of T eraSim lies in the fact that it implements energy harvesting capability of the nanonodes. Hence, the current ener gy le vels of the nodes play a role in the network performance simulations, arguably making T eraSim the most suitable simulation tool for a variety of low-po wer THz nanoscale applications.  As outlined above, there are sev eral tools currently av ailable for THz nanocommunication and nanonetworking simulations. Being one of the most recently proposed simulators, T eraSim seemingly provides the most extensi ve capabilities, including the energy-harvesting module for nanonodes. BitSimulator, another recently proposed tool, trades-off the complexity for scalability , arguing that for many of the envisioned nanocom- munication applications scalability will be the primary require- ment. Howe ver , at this point the scalability vs. realism trade- off is merely a speculation, as it is not clear how scalable the outlined simulators are. In addition, it would be interesting to in vestigate the difference in simulation results of the different simulators in order to ev aluate their usability for different scenarios, as well as the reliability of the simulation results. In other words, large discrepancies in the results deri ved using different simulators could put into question the reliability of ﬁndings obtained using these simulators. In many of the en visioned applications, the only feasible powering option for the nanonodes will be through energy harvesting. This fact is only reﬂected in the T eraSim simulator . Even there, the ener gy storage capacity of the nanonodes is unlimited and the energy harvester is implemented as a simple process in which energy is harvested at a constant rate. Howe v er , the majority of nanoscale harvesters (e.g., exploiting piezo-electric ef fect of Zinc Oxide (ZnO) nano wires [10] or ultrasound-based power transfer [243]) charge the node in a non-linear way , with their harvesting rates being highly dependent on their current energy lev els and maximum stor- age capacity [244]. T o address these limitations, the insights from [245] could be utilized. [245] provides N3Sim, an ns-3-based simulator for molecular nanocommunication. It provides sev eral harvesting options (with harvesting being a synonym for collecting molecules from nanonodes’ local neighborhood), some of them realistically assuming that the nanonode’ s energy storage capacity is limited. In addition, the charging operation due to harvesting is in N3Sim a non- linear process. Moreov er , in the currently av ailable simulators the energy consumption of a nanonode is attributed to trans- mission and reception only . Howe ver , idling energy should be accounted for if the aim is accurate ener gy consumption modeling, as well as energy consumed due to for example information processing or data storage. Due to nanonode’ s highly constrained energy resources, accurate ener gy modeling should be of paramount importance, as will be discussed in the next section in more details. Furthermore, novel mobility models are needed for accurate simulations of nanonetworks for sev eral application scenarios. For e xample, as body-centric applications obviously assume in-body communication, hence ﬁne-grained models of human mobility are needed, as well as models for blood stream and various other in-body mobility effects (e.g., heart-beats). A very good initial step in this direction is BloodV oyagerS [246], a model of a human body’ s cardiovascular system, dev eloped with the idea of simulating nanonodes’ mobility . Similar tools are needed for other aspects of a human body . In addition, the integration of such mobility models with the current simulators will be needed for maximizing the beneﬁts and realism of the THz nanocommunication and nanonetworking simulations. Finally , in terms of experimentation facilities or experi- mental datasets, the areas of THz nanocommunication and nanonetworking are still uncharted. Publicly available exper- imental results of experimentation infrastructure would pre- sumably give a strong boost to the research in these domains, which has already been recognized in the community . One good example going in this direction is VISORSURF , a European Union (EU)-funded research project whose aim is to de velop a full stack of hardware and software tools for THz-based control of metamaterials [64], [75]. More initiati ves targeting dev elopment of hardware tools, integration into full prototypes, or generation of public datasets are needed for other nanocommunication scenarios. I X . A D D I T I O NA L C H A L L E N G E S , O P E N I S S U E S , A N D F U T U R E R E S E A R C H D I R E C T I O N S The optimization objectives for the abov e-discussed pro- tocols are indicated in T able VII. These objecti v es represent the design aims of the protocols and should not be mixed with the performance metrics used in the ev aluation of the protocols and listed in T ables III, IV, and V. The aim of T able VII is to help the reader in the selection of suitable protocols for a giv en application with speciﬁc requirements. In addition, the aim is to indicate the “missing pieces” in the existing protocols, i.e., the potential improvement directions. For example and as already mentioned, in terms of link-later protocols Akkari et al. [144] is the only one explicitly aiming at latency optimization. Hence, if an application requires certain bounds on latency , Akkari et al. [144] would naturally be the ﬁrst choice for the link layer . Moreover , giv en that there is only one proposal targeting latency optimization, new link layer protocols could be developed for its further optimization. In addition, there are several overarching challenges not directly related to the ones outlined in pre vious sections, which we discuss in the reminder of this section. 27 T ABLE VII: Optimization objectives for existing protocols in different layers of the protocol stack Protocol Network scalability Node density Latency Throughput Bidirectional trafﬁc Reliability Energy consumption Mobility Security Network layer protocols Xia et al. [118] X X X Rong et al. [119] X X Y u et al. [120] ( X ) X X X X ( X ) PESA WNSN [121] X X C ´ anov as-Carrasco et al. [123] ( X ) X X C ´ anov as-Carrasco et al. [124] X X X Liaskos et al. [125] X X X X Tsioliaridou et al. [126] X X X X Afsana et al. [127] X X X X Stelzner et al. [128] X X Buther et al. [129] X X X E 3 A [130] X X X X Pierobon et al. [122] X X X X Tsioliaridou et al. [137] X X X X Tsioliaridou et al. [138] X X X X Link layer protocols Akkari et al. [144] X X Alsheikh et al. [145] X X PHLAME [114] X X X X X DRIH-MA C [146] X X X X TCN [147] X X Xia et al. [148] X X Smart-MA C [149] X APIS [150] X X X W ang et al. [115] X X X EEWNSN-MA C [151] X X X ( X ) CSMA-MA C [153] X Mansoor et al. [154] X X X X BRS-MA C [108] X X X Dynamic MAC [155] X X X X Physical layer protocols TS-OOK [165] X X X PHLAME [114] X X X X X Shrestha et al. [166] X X X X X SRH-TSOOK [167] X X X X X ASRH-TSOOK [168] X X X X X V av ouris et al. [169] X X Zarepour et al. [170] X X Multi-band OOK [162] X X X X D AMC [184] X X X Han et al. [171] X X X Jornet et al. [172] X X X X MEC [173] X X Chi et al. [174] X X X SBN [175] X X DS-OOK [176] X X A. Common Design F rame works There is a need for common design frameworks in order to simplify the development of the THz nanonetworks for supporting different types of the en visioned applications. W e believ e this issue should be approached from the perspecti ves of both the applications and the nanodevices. In other words, in order to support a certain application in one of the speciﬁed application domains, one should be aware of the requirements that this application poses on the underlying nanonetwork. Simultaneously , one should consider the capabilities of the nanodevices expected to be used in that particular context, as it is hardly imaginable that the same types of devices will be used for enabling e.g., the on-chip, in-body , or software- deﬁned metamaterial-based applications. In Section III, we aimed at deriving the application re- quirements on the lev el of dif ferent application domains. Nev ertheless, we belie ve such deri v ations will be needed on the le vel of particular applications. The current literature on this issue is sparse, with only a fe w works mostly targeting applications in the domain of in-body communication [92], [94]–[96]. Further research is needed and should, in contrast to the existing works, aim at qualitatively specifying exhaustive sets of application requirements for all applications in all of the promising application domains. From the perspective of the nanodevices, we ar gue the full speciﬁcations of their features are needed, with primarily concerns being their size, av ailable energy , processing power , and memory . In this direction, it is worth emphasizing [247], where by follo wing the guidelines originally outlined in [5] the authors discuss a conceptual design of a nanodevice, consisting of a nanoprocessor , nanomemory , nanoantenna, and nanogenerator . Seemingly , the nanodevice will not be suitable for all the applications, for example because it assumes a nanogenerator based on ener gy harvesting or wireless power transfer . As such, it can be hardly characterized as feasible e.g., on-chip communication in which the energy is abundant. W e believ e the deriv ation of the mentioned perspectives will not only substantially simplify the dev elopment of the support- 28 ing nanonetworks along well-speciﬁed design requirements and constraints, but also highly contribute to the intuition and reasoning on the feasibility of different approaches for a particular set of applications. B. T ransport Layer Pr otocols In [26], [27], the authors state that, as Gbps and Tbps links become a reality , the netw ork throughput will increase dramatically . It will, therefore, be necessary to develop new transport layer solutions for mitigating network congestion problems. The authors in [26], [27] also state that the overhead of existing transport layer protocols requires minimization in order to reduce the performance constraints. This has been recognized by only a fe w early works on the topic [248]–[250]. In the scope of the VISORSURF project, Tsioliaridou et al. [248] consider a software-deﬁned metamaterial named the HyperSurface. The HyperSurf ace encompasses a hardware layer that can change its internal structure by tuning the state of its active elements, as well as a nanonetwork in which each nanonode controls a single activ e element. For carrying sensed information to the external world for processing, as well as conﬁguration commands from the external world to the HyperSurface, the authors propose HyperSurface Con- trol Protocol (HyperCP). The Hyper-CP uses L yapunov drift analysis for avoiding congested or out-of-power nanonetwork areas [249]. Speciﬁcally , the protocol aims at optimizing the network throughput by accounting for outdated nanonodes’ status information, as well as battery and latency constraints. Alam et al. [250] state that the existing studies on nanocom- munication and nanonetworking in body area nanonetworks focus on lower layers of the protocol stack, resulting in the upper layers (e.g., transport layer) remaining unexplored. They further argue that electromagnetic wav es will potentially be harmful to sensitiv e body areas. Moreover , they claim that molecular communication is slower and error-prone compared to electromagnetic one. Motiv ated by the above arguments, they propose an ener gy-efﬁcient transport layer protocol for hybrid body area nanonetworks (i.e., both electromagnetic and molecular communication supported). The protocol is based on a congestion control mechanism with v ery limited o verhead, in which the sender upon recei ving a “halt” signal suspends the packet transmission for a predeﬁned timeout period. These initial contributions are constrained to narrow appli- cation domains, speciﬁcally to software-deﬁned metamaterials and body-centric communication. T ransport layer protocols for other application domains with differing requirements are currently lacking. Even the two outlined protocols are lacking relev ant performance details, pertaining primarily to their scalability and protocol overhead. For example, for Hyper-CP it is unclear if the distribution of battery states and latency constraints among nanonodes is at all feasible, gi v en that it unav oidably causes signalling-related energy dissipation at the nanonodes. Similarly , for the protocol proposed in [250], halt signals can potentially be infeasible for low-ener gy nanonodes. In such a case, the transmission of data packets could unneces- sarily continue until the depletion of the transmitter’ s energy . In summary , transport layer protocols for nanonetworks are currently to a large extent unexplored. Even more, it is unclear if such protocols will at all be needed for many scenarios, as they will inevitably increase the protocol overhead, which for energy-constrained nanonodes could be infeasible. C. Reduced / Integr ated Protocol Stac k Along the above conclusions, in case of energy-constrained nanonodes, the protocol stack will hav e to be substantially condensed and integrated for maintaining feasible nanocom- munication and nanonetworking. This it predominantly due to the fact that more traditional networks trade-off a high ov erhead of the utilized protocols with the support for het- erogeneous application requirements. In nanocommunication and nanonetworking, this paradigm will potentially be shifted, which is currently largely unexplored with the only two exam- ples being the ones outlined below . Certainly , further insights are needed in terms of the design choices for nanonetworks for enabling different applications, pertaining to either reducing the complexity of the protocol stack by making it condensed and application speciﬁc, or providing a full stack at the cost of increased energy consumption, latency , and complexity . The ﬁrst example where the authors argue that the paradigm shift is needed is [251] (extended in [252]), in which a frame- work is proposed for enabling energy-harvesting nanonodes to communicate their locations (i.e., addresses) and sensed ev ents using only one wireless pulse. These wireless pulses feature two degrees of freedom pertaining to an y two of the following parameters: amplitude, pulse width, and transmitted energy (equaling amplitude multiplied by pulse width). The framew ork requires each nanonode to use a particular pulse width for their identiﬁcation, while the e vent types are identi- ﬁed by the amplitude/energy emitted by each nanonode. Similarly , the authors in [253] argue that even exotic (i.e., har d to inte grate) power supplies relying on energy harvesting can only scavenge energy for 1 pack et transmission per appr oximately 10 sec [244]. This makes the development of even basic pr otocols – such as addr essing and routing – highly challenging . T o mitigate this effect, the authors propose Bit- Surﬁng, a network adapter that does not generate physical data packets when transmitting information, but assigns meaning to symbols created by an external source. It does that by reading incoming symbols and waiting for desired messages to appear in the stream. Once they appear, short low-energy pulse is then emitted to notify neighboring nodes. The authors demonstrate BitSurﬁng’ s perpetual operation, as well as the ability to operate without link and transport layer protocols. Another example of a shallow protocol stack is given in the on-chip communication context, where packets need to be served to guarantee forward progress in the computation. This means that cores are self-thr ottling , hence their injection speed will be reduced if the network becomes congested [100]. As a result, the responsibilities of the transport layer are reduced and generally implemented at the architecture lev el. For instance, cache coherence protocols (which generate most of the trafﬁc in multiprocessors) typically implement end- to-end acknowledgment and timeouts to conﬁrm the reads and writes on shared data. In fact, on-chip communication 29 take advantage of the monolithic nature of the multiprocessor system to reduce the depth of the protocol stack. D. Ener gy Lifecycle Modeling From the abo ve discussions, it is clear that the energy consumption is one of the most stringent constraints that will potentially impede on the feasibility of nanocommunication and nanonetworking in the THz frequencies. In order to dev elop feasible nanocommunication and nanonetw orking pro- tocols, there is a need for accurate energy lifecycle modeling, especially for the nanonodes whose only powering option is through ener gy harvesting. As the practical implementations of the nanonodes are currently lacking, the development of accurate analytical energy lifecycle models can be seen as a fundamental step towards the design of feasible nanonetwork architectures and protocols [84]. This has to an extent been recognized in the research community . In the pioneering works on the topic [84], [244], an energy model for self-powered nanonodes is dev eloped with the aim of capturing the correlation between the nanonodes’ energy harvesting and the energy consumption processes. The energy harvesting process is realized by means of a piezoelectric nanogenerator , while the nanonode’ s energy consumption is quantiﬁed by assigning certain amounts of energy for the transmission and reception of “1” bits, under the assumption of the TS-OOK communication scheme being employed. A mathematical framework is then de veloped for deri ving packet deliv ery probability , end-to-end delay , and achiev able network throughput. A similar approach for energy lifecycle modeling has been taken in [254], [255], in which the authors addition- ally e valuate the contributions of packet sizes, repetitions, and code weights on the nanonode’ s energy consumption. The authors in [243] e xtended the work from [84], [244] by highlighting the effects of the employed network topol- ogy , as well as of different energy harvesting approaches and rates (i.e., blood currents and ultrasound-based power transfer). Their results show that a micrometer-sized piezo- electric system in lossy en vironments becomes inoperativ e for transmission distances over 1.5 mm. Similarly , the authors in [72], [256] reason that, as any other practical transcei ver , a nanoscale transceiv er will consume certain amounts of energy during its idling periods, in addition to the consumption due to transmission and reception. Once the idling energy is accounted for in the ov erall energy consumption modeling, the authors show that for feasible communication this energy consumption has to be nine orders of magnitude smaller than the energy consumed for reception during the same period. This is certainly a challenging requirement, as in current nanoscale systems the idling energy is in the best case scenario up to three orders of magnitude smaller than the corresponding energy in reception. The above mentioned contributions are only the initial steps in a promising direction. As discussed in [72], [256], certain energy will ine vitably be consumed when a nanonode is idling. By the same token, the nanodevice’ s energy will be distributed to different functions such as sensing, processing, etc., and not all of it can be used for transmission or reception. That being said, seemingly man y current works on the design of nanocommunication and nanonetworking protocols could be infeasible under more realistic energy consumption models, as they indeed make an assumption that the overall energy of a nanode vice can be utilized for transmission and reception of information. Example-wise, as the nanonodes are expected to experience intermittent on-off behavior , they will often have to wake-up after harv esting sufﬁcient ener gy [157]. This wake-up process per-se will consume a certain amount of energy , which the current energy models and consequently the protocols utilizing such models do not account for . As mentioned, in many applications the nanonodes will be po wered solely through harv esting en vironmental ener gy . Hence, it is ﬁrst worth pointing out that there is a variety of potentially feasible candidates for energy harvesting, for example blood currents [243] and ultrasound-based power transfer [257] in in-body communication applications, and Radio-Frequency (RF)-based power transfer in the domain of SDMs [258]. Similar to energy consumption modelling, there is a need for an accurate modelling of the charging of an energy-harv esting nanonode. Here, it is worth pointing out the effort in [257], where the authors outline an approach for ultrasound-based powering of piezoelectric ZnO nanowires- based nanonodes placed on peripheral nerves in the human body . In their characterization of the harvested energy (i.e., intensity of ultrasound), the authors consider a variety of relev ant parameters such as the stimulus depth, size of the energy-harv esting array , and duration of ultrasound pulses. Such a comprehensiv e characterization makes it easy for the proposed approach to be utilized in future works, among others in the domain of THz nanonetworks. Such efforts are needed for other types of energy harvesters and harvesting sources. In more general terms, accurate energy harvesting and consumption models would be highly beneﬁcial. E. End-to-End Ar chitectur es T o truly support many of the en visioned applications, seam- less integration of the nanonetworks with existing networking infrastructures will be needed [86]. Addressing this issue is not straightforward, as existing networks predominantly utilize carrier-based electromagnetic communication, while nanonet- works will seemingly have to rely on energy-constrained pulse-based communication. Thus, special gatew ay nodes be- tween the macro- and nano-worlds will be required, which has been only sporadically addressed in the literature to date. The authors in [259] argue that data acquisition from nanonetworks faces two challenges. First, the mismatch be- tween the demands of the nanonetworks and the av ailable bandwidth of the backhaul link with the macro-world reduces the bandwidth efﬁciency of the backhaul, as well as the energy efﬁcienc y of THz band nanonetworks. T o address this issue, the authors propose a polling mechanism for the backhaul tier which is composed of nanosinks that aggregate and transport data from nanonodes to the gateway . The mechanism is based on on-demand polling and accounts for the dynamic backhaul bandwidth and THz channel conditions. The mechanism is dev eloped with the goal of supporting applications in the domain of wireless robotic materials. 30 Sev eral works (i.e., [32], [34], [260], [261]) propose hier - archical network architectures for enabling body-centric com- munication and sensing only-based applications. Speciﬁcally , the aim in these works is to enable uplink communication between a nanonetwork deployed inside of a human body and external monitoring de vices through nanointerfaces (i.e., gate ways). Moreover , in [29], [30] the authors aim at enabling interconnection of sensing nanodevices with existing commu- nication networks, ev entually forming the IoNT. Howe v er , the current results mostly aim at enabling uplink communication from the nanonetworks to ward the macro- world. In order to unlock the full potential of the en visioned applications, downlink (predominantly control) communica- tion will also be required. This will enable applications ranging from software-controlled metamaterials to control of and ac- tuation using wireless robotic materials. Future in vestigations should also aim at minimizing the latency of such communi- cation, especially for control-related applications. F . Security , Integrity , and Privacy One question that did not receiv e substantial research at- tention is the security of THz band nanocommunication. The authors in both [262] and [263] argue that the security-related goals in THz band nanocommunication should be conﬁden- tiality (protection against malicious or unauthenticated users), integrity (protection against modiﬁcation), and av ailability (protection against disruption by a malicious user). There are sev eral challenges pertaining to achieving those goals, as in more details discussed in [262]. First, it will be necessary to dev elop methods for the establishment of shared encryption keys, as well as for rev oking them when needed. Second, the ov erhead of secure communication protocols, cryptographic algorithms, and access control and authentication methods will have to be minimized and potentially reconsidered for ultra-low po wer nanonetworks. Finally , as the prev ention of all malicious attacks can hardly be guaranteed, it will be important to at least de velop means for their detection, as well as strate gies for reacting to them. Ho we ver , the abov e-listed challenges are all but explored in the context of THz band nanocommunication and nanonetworking. G. Localization and T rac king Many of the en visioned applications supported by THz nanonetworks require localization or ev en tracking of the nanonodes. This is a highly challenging requirement, giv en that such localization and tracking capabilities will, due to the nature of the nanonodes, hav e to operate in a highly energy- constrained way , as well as provide very high accuracy (due to the small sizes of the nanonodes). Moreover , due to the low range of THz band nanocommunication, the localization capability will potentially have to be based on multi-hopping, which has a drawback in terms of propagation of localization errors with the number of hops [264]. There are only a handful of attempts in localizing THz- operating nanonodes. In [265], the authors propose two ranging- and hop-counting-based localization algorithms. The algorithms are envisioned to be used for estimating the loca- tions of all nanonodes deployed within a certain area. The ﬁrst technique uses ﬂooding-based forwarding to all nanonodes, where the locations of the two considered nanonodes are estimated by counting the number of hops between them. T o reduce the ov erhead and energy dissipation, the second algorithm works under the assumption that all nanonodes are grouped into clusters. Cluster heads communicate together and count the number of hops in order to localize different nodes within a cluster . Similarly , Zhou et al. [266] propose a pulse- based distance accumulation (PBD A) localization algorithm. The PDB A algorithm adopts TS-OOK pulses for estimating the distance between clustered nanonodes. The algorithms proposed in [265], [266] can potentially operate within the nanonodes’ ener gy constraints. Howe ver , their accuracy is intrinsically going to be relativ ely low due to the propagation of localization errors as the number of hops increases. One potential direction in enhancing localization accuracy , while at the same time maintaining its lo w ener gy consumption proﬁle, is to base the localization capabilities on backscat- tered signals. The feasibility and promising accuracy of such an approach has been demonstrated in [267]. The approach in [267] utilizes a backscattered signal from a nanonode (i.e., DR-Lens tag) for extracting the round-trip time-of-ﬂight (R T oF) between the tag and the localization anchor . The R T oF readings from multiple anchors are then used for estimating the corresponding distances to the nanonode, with the nanonode’ s location consequently being determined using linear least square algorithm. The yet unresolved challenges of developing such systems include angle and frequency-dependent response of the nanonodes, non-free space (e.g., in-body) propagation, and almost certainly a variety of hardware imperfections. H. Standar dization As stated in [27], the THz band is not yet regulated and it is up to the scientiﬁc community to jointly deﬁne the future of the paradigm. The IEEE P1906.1 standard [268] is, to the best of our knowledge, the only attempt going in this direction for THz-based nanocommunication. The standard pro vides a conceptual framework for future dev elopments of nanoscale communication networks. C ´ anov as-Carrasco et al. [269] pro- vided a re view of the latest IEEE P1906.1 recommendations, in which they outlined the main features and identiﬁed sev eral shortcomings of the standard, to which, they argue, further research efforts should be devoted. First, the characteristics and reference powering solutions of potential nanodevices hav e not been discussed in the standard. Second, the recom- mended values and ranges for respectively transmission po wer and SNR for reception hav e not been speciﬁed. Third, the standard does not specify the Open Systems Interconnection (OSI) layers 2 and 3 techniques, which hampers protocol interoperability . In order words, standardization efforts aiming at media access control, addressing schemes, ﬂo w control, error detection, and routing procedures are needed. Finally , the standard currently lacks recommendations about the intercon- nection between nanonetworks and existing communication networks. Substantial research efforts are certainly still needed for addressing the abov e-stated limitations of the standard. 31 X . C O N C L U S I O N In this survey , we hav e outlined the most promising appli- cation domains that could be enabled by the nanonetworks operating in the THz frequencies. Moreover , we hav e deriv ed the requirements that such applications pose on the supporting nanonetworks, which could be utilized as a rule-of-thumb guidelines in the dev elopment of the supporting nanocommu- nication and nanonetworking protocols. W e have then outlined the current State-of-the-Art (SotA) nanocommunication and nanonetworking protocols, as well as the applicable channel models and experimentation tools. In addition, we ha ve dis- cussed their strengths and weaknesses, as well as summarized a set of potential directions for future research. Future efforts could target the dev elopment of THz nanonetwork proto- types and experimental testing infrastructures, dev elopment of protocols for primarily higher layers of the protocol stack, mitigation of mobility ef fects, enabling additional features such as security or localization of nanodevices, to name some. W e believ e that this survey demonstrates that THz band nanocommunication is a vivid and promising research domain, as Prof. Feynman suggested it will be more than 60 years ago. W e encourage the community to focus on resolving the indicated major challenges, so that the outlined set of exciting applications could become a reality in the near future. A C K N O W L E D G M E N T S The author Filip Lemic was supported by the EU Marie Skłodowska-Curie Actions Individual Fellowships (MSCA- IF) project Scalable Localization-enabled In-body T erahertz Nanonetwork (SCaLeITN, grant nr . 893760). The author Pieter Stroobant was supported by a PhD grant of Ghent University’ s Special Research Fund (BOF). This work was partly funded by the Ghent University’ s BOF/GOA project “ Autonomic Net- worked Multimedia Systems”. This work was also supported by the Spanish Ministry of Economy and Competitiv eness under contract TEC2017-90034-C2-1-R (ALLIANCE project) that receiv es funding from FEDER, and by the European Commission under grants No. 736876 and No. 863337. R E F E R E N C E S [1] R. P . Feynman, “There’s plenty of room at the bottom, ” California Institute of T ec hnology , Engineering and Science magazine , 1960. [2] Q. H. Abbasi, K. Y ang, N. Chopra, J. M. Jornet, N. A. Abuali, et al. , “Nano-communication for biomedical applications: A re view on the state-of-the-art from physical layers to novel networking concepts, ” IEEE Access , vol. 4, pp. 3920–3935, 2016. [3] S Hiyama et al. , “Molecular communication, ” Journal-Institute of Electr onics Information and Communication Engineers , vol. 89, no. 2, p. 162, 2006. [4] Y . Chen et al. , “A touch-communication framew ork for drug deliv- ery based on a transient microbot system, ” IEEE T ransactions on Nanobioscience , vol. 14, no. 4, pp. 397–408, 2015. [5] I. F . Akyildiz and J. M. Jornet, “Electromagnetic wireless nanosen- sor networks, ” Nano Communication Networks , vol. 1, no. 1, pp. 3– 19, 2010. [6] I. F . Ak yildiz, J. M. Jornet, and M. Pierobon, “Propagation models for nanocommunication networks, ” in Antennas and Pr opagation (EuCAP) , IEEE, 2010, pp. 1–5. [7] J. M. Jornet and I. F . Ak yildiz, “Graphene-based nano-antennas for electromagnetic nanocommunications in the terahertz band, ” in Pr oceedings of the F ourth Eur opean Conference on Antennas and Pr opagation , IEEE, 2010, pp. 1–5. [8] N. Rikhtegar and M. Keshtgary, “A brief survey on molecular and electromagnetic communications in nano-networks, ” International Journal of Computer Applications , vol. 79, no. 3, 2013. [9] J. M. Jornet, I. F . Akyildiz, et al. , “Fundamentals of electromagnetic nanonetworks in the terahertz band, ” F oundations and Tr ends® in Networking , vol. 7, no. 2-3, pp. 77–233, 2013. [10] Z. L. W ang, “T o wards self-po wered nanosystems: From nanogener- ators to nanopiezotronics, ” Advanced Functional Materials , vol. 18, no. 22, pp. 3553–3567, 2008. [11] G. W . Hanson, “Fundamental transmitting properties of carbon nan- otube antennas, ” IEEE T ransactions on antennas and pr opagation , vol. 53, no. 11, pp. 3426–3435, 2005. [12] I. Llatser, C. Kremers, A. Cabellos-Aparicio, J. M. Jornet, E. Alarc ´ on, and D. N. Chigrin, “Graphene-based nano-patch an- tenna for terahertz radiation, ” Photonics and Nanostructures- Fundamentals and Applications , vol. 10, no. 4, pp. 353–358, 2012. [13] M. R. Da Costa, O. Kibis, and M. Portnoi, “Carbon nanotubes as a basis for terahertz emitters and detectors, ” Microelectr onics Journal , vol. 40, no. 4-5, pp. 776–778, 2009. [14] A. V akil and N. Engheta, “Transformation optics using graphene, ” Science , vol. 332, no. 6035, pp. 1291–1294, 2011. [15] J. Horng, C.-F . Chen, B. Geng, C. Girit, Y . Zhang, Z. Hao, H. A. Bechtel, M. Martin, A. Zettl, M. F . Crommie, et al. , “Drude conductivity of dirac fermions in graphene, ” Physical Re view B , vol. 83, no. 16, p. 165 113, 2011. [16] S. A. Mikhailov, “Theory of the giant plasmon-enhanced second-harmonic generation in graphene and semiconductor two- dimensional electron systems, ” Physical Review B , vol. 84, no. 4, p. 045 432, 2011. [17] T . Zhou, Z. Cheng, et al. , “Miniaturized tunable terahertz antenna based on graphene, ” Micr owave and Optical T ec hnology Letters , vol. 56, no. 8, pp. 1792–1794, 2014. [18] J. Perruisseau-Carrier, M. T amagnone, J. S. Gomez-Diaz, and E. Carrasco, “Graphene antennas: Can integration and reconﬁgurability compensate for the loss?” In 2013 Eur opean Micr owave Confer ence , Ieee, 2013, pp. 369–372. [19] D. Correas-Serrano and J. S. Gomez-Diaz, “Graphene-based An- tennas for T erahertz Systems: A Review , ” FERMA T , 2017. arXiv: 1704.00371. [20] Y . Wu, D. B. F armer, F . Xia, and P . A vouris, “Graphene electronics: Materials, devices, and circuits, ” Proceedings of the IEEE , v ol. 101, no. 7, pp. 1620–1637, 2013. [21] S.-J. Han, A. V . Garcia, S. Oida, K. A. Jenkins, and W . Haensch, “Graphene radio frequency receiver integrated circuit, ” Natur e com- munications , vol. 5, no. 1, pp. 1–6, 2014. [22] D. Neumaier, S. Pindl, and M. C. Lemme, “Integrating graphene into semiconductor fabrication lines, ” Natur e Materials , vol. 18, no. 6, p. 525, 2019. [23] J. M. Jornet and I. F . Akyildiz, “Graphene-based plasmonic nano- transceiv er for terahertz band communication, ” in Proceedings of the EuCAP ’14 , 2014, pp. 492–6. [24] S. K eele et al. , “Guidelines for performing systematic literature revie ws in software engineering, ” T echnical report, V er . 2.3 EBSE T echnical Report. EBSE, T ech. Rep., 2007. [25] J. Cruz-Benito, “Systematic literature review & mapping, ” 2016. [26] I. F . Ak yildiz, J. M. Jornet, and C. Han, “T eraNets: Utra-broadband communication networks in the terahertz band, ” IEEE W ir eless Communications , vol. 21, no. 4, pp. 130–135, 2014. [27] ——, “T erahertz band: Next frontier for wireless communications, ” Physical Communication , vol. 12, pp. 16–32, 2014. [28] V . Petrov, A. Pyattaev , D. Moltchanov, and Y . K oucheryavy , “T er - ahertz band communications: Applications, research challenges, and standardization activities, ” in 2016 8th International Congr ess on Ultra Modern T elecommunications and Contr ol Systems and W orkshops (ICUMT) , IEEE, 2016, pp. 183–190. [29] I. F . Ak yildiz and J. M. Jornet, “The internet of nano-things, ” IEEE W ir eless Communications , vol. 17, no. 6, pp. 58–63, 2010. [30] J. M. Jornet et al. , “The internet of multimedia nano-things, ” Nano Communication Networks , vol. 3, no. 4, pp. 242–251, 2012. [31] S. Balasubramaniam and J. Kangasharju, “Realizing the internet of nano things: Challenges, solutions, and applications, ” Computer , vol. 46, no. 2, pp. 62–68, 2013. [32] I. F . Akyildiz, M. Pierobon, S. Balasubramaniam, and Y Kouch- eryavy , “The internet of bio-nano things, ” IEEE Communications Magazine , vol. 53, no. 3, pp. 32–40, 2015. [33] M. H. Miraz, M. Ali, P . S. Excell, and R. Picking, “A revie w on Internet of Things (IoT), Internet of everything (IoE) and Internet of 32 nano things (IoNT), ” in 2015 Internet T ec hnologies and Applications (IT A) , IEEE, 2015, pp. 219–224. [34] F . Dressler and S. Fischer, “Connecting in-body nano communica- tion with body area networks: Challenges and opportunities of the internet of nano things, ” Nano Communication Networks , vol. 6, no. 2, pp. 29–38, 2015. [35] R. Alsheikh, N. Akkari, and E. Fadel, “MAC protocols for wireless nano-sensor networks: Performance analysis and design guidelines, ” in Digital Information Processing and Communications (ICDIPC) , IEEE, 2016, pp. 129–134. [36] A. Rizwan, A. Zoha, R. Zhang, W . Ahmad, K. Arshad, N. A. Ali, A. Alomainy, M. A. Imran, and Q. H. Abbasi, “A review on the role of nano-communication in future healthcare systems: A big data analytics perspectiv e, ” IEEE Access , vol. 6, pp. 41 903–41 920, 2018. [37] S. Ghafoor, N. Boujnah, M. H. Rehmani, and A. Davy, “Mac protocols for terahertz communication: A comprehensiv e survey , ” IEEE Communications Surveys & Tutorials , 2020. [38] K. M. S. Huq, J. M. Jornet, W . H. Gerstacker, A. Al-Dulaimi, Z. Zhou, and J. Aulin, “Thz communications for mobile heterogeneous networks, ” IEEE Communications Magazine , vol. 56, no. 6, pp. 94– 95, 2018. [39] S. Mumtaz, J. M. Jornet, J. Aulin, W . H. Gerstack er, X. Dong, and B. Ai, “T erahertz communication for vehicular networks, ” IEEE T ransactions on V ehicular T echnology , vol. 66, no. 7, pp. 5617– 5625, 2017. [40] Z. Chen, X. Ma, B. Zhang, Y . Zhang, Z. Niu, N. Kuang, W . Chen, L. Li, and S. Li, “A survey on terahertz communications, ” China Communications , vol. 16, no. 2, pp. 1–35, 2019. [41] K. M. S. Huq, S. A. Busari, J. Rodriguez, V . Frascolla, W . Bazzi, and D. C. Sicker, “T erahertz-enabled wireless system for beyond-5g ultra-fast networks: A brief survey, ” IEEE Network , vol. 33, no. 4, pp. 89–95, 2019. [42] X. Ma, Z. Chen, W . Chen, Y . Chi, Z. Li, C. Han, and Q. W en, “Intel- ligent reﬂecting surface enhanced indoor terahertz communication systems, ” Nano Communication Networks , v ol. 24, p. 100 284, 2020. [43] M. Kuscu, E. Dinc, B. A. Bilgin, H. Ramezani, and O. B. Akan, “T ransmitter and receiv er architectures for molecular communica- tions: A surve y on physical design with modulation, coding, and detection techniques, ” Pr oceedings of the IEEE , vol. 107, no. 7, pp. 1302–1341, 2019. [44] O. B. Akan, H. Ramezani, T . Khan, N. A. Abbasi, and M. Kuscu, “Fundamentals of molecular information and communication sci- ence, ” Proceedings of the IEEE , vol. 105, no. 2, pp. 306–318, 2016. [45] J. W ang, B. Y in, and M. Peng, “Diffusion based molecular com- munication: Principle, key technologies, and challenges, ” China Communications , vol. 14, no. 2, pp. 1–18, 2017. [46] L. Felicetti, M. Femminella, G. Reali, and P . Li ` o, “ Applications of molecular communications to medicine: A survey , ” Nano Commu- nication Networks , vol. 7, pp. 27–45, 2016. [47] Y . Chahibi, “Molecular communication for drug delivery systems: A survey, ” Nano Communication Networks , vol. 11, pp. 90–102, 2017. [48] R. A. Lewis, “A re view of terahertz sources, ” J ournal of Physics D: Applied Physics , vol. 47, no. 37, p. 374 001, 2014. [49] E. Lallas, “Key roles of plasmonics in wireless thz nanocommuni- cations—a surve y , ” Applied Sciences , vol. 9, no. 24, p. 5488, 2019. [50] H. Elayan, R. M. Shubair, and A. Kiourti, “On graphene-based thz plasmonic nano-antennas, ” in 2016 16th Mediterranean Micr owave Symposium (MMS) , IEEE, 2016, pp. 1–3. [51] S. H. Naghavi, M. T . T aba, R. Han, M. A. Aseeri, A. Cathelin, and E. Afshari, “Filling the gap with sand: When cmos reaches thz, ” IEEE Solid-State Circuits Magazine , v ol. 11, no. 3, pp. 33–42, 2019. [52] K. Sengupta, T . Nagatsuma, and D. M. Mittleman, “T erahertz in- tegrated electronic and hybrid electronic–photonic systems, ” Natur e Electr onics , vol. 1, no. 12, pp. 622–635, 2018. [53] I. F . Akyildiz, “Nanonetworks: A new frontier in communications, ” in ACM Mobile computing and networking , ACM, 2012, pp. 1–2. [54] S. W alia, C. M. Shah, P . Gutruf, H. Nili, D. R. Chowdhury , W . Withayachumnankul, M. Bhaskaran, and S. Sriram, “Flexible metasurfaces and metamaterials: A re view of materials and fabrica- tion processes at micro-and nano-scales, ” Applied Physics Reviews , vol. 2, no. 1, p. 011 303, 2015. [55] W . Cai, U. K. Chettiar, et al. , “Optical cloaking with metamaterials, ” Natur e photonics , vol. 1, no. 4, p. 224, 2007. [56] X. W ang, Z. Huang, F . Han, and Z. Feng, “Reduction of aircraft engine noise by covering surface acoustic metamaterials on side- walls, ” in Pr oceedings of the 24th International Congress on Sound and V ibration (ICSV), London, UK , 2017, pp. 23–27. [57] A. Poddubny , I. Iorsh, P . Belov, and Y . Kivshar, “Hyperbolic metamaterials, ” Natur e photonics , vol. 7, no. 12, p. 948, 2013. [58] D. R. Smith, J. B. Pendry, et al. , “Metamaterials and negati ve refractiv e index, ” Science , vol. 305, no. 5685, pp. 788–792, 2004. [59] T . Zhai, J. Shi, S. Chen, and D. Liu, “Electromagnetic shielding and energy concentration using zero-index metamaterials, ” Applied physics expr ess , vol. 4, no. 7, p. 074 301, 2011. [60] C. Liaskos, A. Tsioliaridou, A. Pitsillides, I. F . Akyildiz, N. V . Kan- tartzis, A. X. Lalas, X. Dimitropoulos, S. Ioannidis, M. Kafesaki, and C. Soukoulis, “Design and development of software deﬁned metamaterials for nanonetworks, ” IEEE Cir cuits and Systems Mag- azine , vol. 15, no. 4, pp. 12–25, 2015. [61] S. Abadal, T .-J. Cui, T . Low, and J. Georgiou, “Programmable metamaterials for software-deﬁned electromagnetic control: Cir- cuits, systems, and architectures, ” IEEE J ournal on Emerging and Selected T opics in Cir cuits and Systems , vol. 10, no. 1, pp. 6–19, 2020. [62] C. Liaskos, S. Nie, A. Tsioliaridou, A. Pitsillides, S. Ioannidis, and I. Akyildiz, “A New Wireless Communication Paradigm through Software-Controlled Metasurfaces, ” IEEE Communications Maga- zine , vol. 56, no. 9, pp. 162–169, 2018, I S S N : 15581896. D O I : 10 . 1109/MCOM.2018.1700659. eprint: 1806.01792. [63] S. Gong, X. Lu, D. T . Hoang, D. Niyato, L. Shu, D. I. Kim, and Y .-C. Liang, “T o wards smart radio environment for wireless communications via intelligent reﬂecting surfaces: A comprehensiv e survey , ” arXiv preprint , 2019. [64] C. Liaskos, S. Nie, A. Tsioliaridou, et al. , “A no vel communication paradigm for high capacity and security via programmable indoor wireless environments in next generation wireless systems, ” Ad Hoc Networks , vol. 87, pp. 1–16, 2019. [65] Q. W u and R. Zhang, “Intelligent reﬂecting surf ace enhanced wire- less network via joint acti ve and passiv e beamforming, ” IEEE T rans- actions on W ireless Communications , vol. 18, no. 11, pp. 5394– 5409, 2019. [66] C. Huang, A. Zappone, G. C. Alexandropoulos, M. Debbah, and C. Y uen, “Reconﬁgurable intelligent surfaces for energy efﬁciency in wireless communication, ” IEEE T ransactions on W ireless Com- munications , vol. 18, no. 8, pp. 4157–4170, 2019. [67] ¨ O. ¨ Ozdogan, E. Bj ¨ ornson, and E. G. Larsson, “Intelligent reﬂect- ing surfaces: Physics, propagation, and pathloss modeling, ” IEEE W ir eless Communications Letters , 2019. [68] W . T ang, J. Y . Dai, M. Chen, X. Li, Q. Cheng, S. Jin, K.-K. W ong, and T . J. Cui, “Programmable metasurface-based rf chain-free 8psk wireless transmitter, ” Electronics Letters , vol. 55, no. 7, pp. 417– 420, 2019. [69] L. Li, T .-J. Cui, W . Ji, S. Liu, J. Ding, X. W an, Y . Bo Li, M. Jiang, C. W . Qiu, and S. Zhang, “Electromagnetic reprogrammable coding- metasurface holograms, ” Natur e Communications , vol. 8, no. 1, pp. 1–7, 2017, IS S N : 20411723. D OI : 10.1038/s41467- 017- 00164- 9. [70] S. Abadal, C. Liaskos, A. Tsioliaridou, S. Ioannidis, A. Pitsillides, J. Sol ´ e-Pareta, E. Alarc ´ on, and A. Cabellos-Aparicio, “Computing and communications for the software-deﬁned metamaterial paradigm: A context analysis, ” IEEE access , vol. 5, pp. 6225–6235, 2017. [71] E. Bj ¨ ornson and L. Sanguinetti, “Power scaling la ws and near-ﬁeld behaviors of massive mimo and intelligent reﬂecting surf aces, ” arXiv pr eprint arXiv:2002.04960 , 2020. [72] F . Lemic, C. Han, and J. Famaey, “Idling energy modeling and re- duction in ener gy harvesting terahertz nanonetworks for controlling software-deﬁned metamaterials, ” IEEE Journal on Emerging and Selected T opics in Cir cuits and Systems , vol. 10, no. 1, pp. 88–99, 2020. [73] T . Saeed, S. Abadal, C. Liaskos, A. Pitsillides, H. T aghv aee, A. Cabellos-Aparicio, M. Lestas, and E. Alarc ´ on, “W orkload Char - acterization of Programmable Metasurfaces, ” in Proceeding of the NANOCOM ’19 , 2019. [74] H. T ataria, F . T ufvesson, and O. Edfors, “Real-time imple- mentation aspects of large intelligent surfaces, ” arXiv pr eprint arXiv:2003.01672 , 2020. [75] F . Liu, O. Tsilipakos, A. Pitilakis, A. C. T asolamprou, M. S. Mir- moosa, N. V . Kantartzis, D.-H. Kwon, M. Kafesaki, C. M. Soukoulis, and S. A. Tretyak ov, “Intelligent metasurfaces with continuously tunable local surface impedance for multiple reconﬁgurable func- tions, ” Physical Review Applied , vol. 11, no. 4, p. 044 024, 2019. 33 [76] H. T aghv aee, A. Cabellos-Aparicio, J. Georgiou, and S. Abadal, “Error analysis of programmable metasurfaces for beam steering, ” IEEE Journal on Emer ging and Selected T opics in Circuits and Systems , vol. 10, no. 1, pp. 62–74, 2020. [77] M. A. McEvo y and N. Correll, “Materials that couple sensing, actuation, computation, and communication, ” Science , vol. 347, no. 6228, p. 1 261 689, 2015. [78] M. McEvoy and N. Correll, “Shape change through programmable stiffness, ” in Experimental Robotics , Springer, 2016, pp. 893–907. [79] N. Correll, P . Dutta, R. Han, and K. Pister , “W ireless robotic mate- rials, ” in ACM Confer ence on Embedded Network Sensor Systems , A CM, 2017, pp. 1–6. [80] H. Hosseinmardi, A. Mysore, N. Farrow, N. Correll, and R. Han, “Distributed spatiotemporal gesture recognition in sensor arrays, ” ACM T ransactions on Autonomous and Adaptive Systems (TAAS) , vol. 10, no. 3, p. 17, 2015. [81] D. Hughes and N. Correll, “T exture recognition and localization in amorphous robotic skin, ” Bioinspiration & biomimetics , vol. 10, no. 5, p. 055 002, 2015. [82] G. P . Fettweis, “The tactile internet: Applications and challenges, ” IEEE V ehicular T echnology Magazine , v ol. 9, no. 1, pp. 64–70, 2014. [83] M. V an de Panne and E. Fiume, “Sensor-actuator networks, ” in Pr oceedings of the 20th annual conference on Computer graphics and interactive techniques , ACM, 1993, pp. 335–342. [84] J. M. Jornet, “A joint energy harvesting and consumption model for self-powered nano-devices in nanonetworks, ” in IEEE International Confer ence on Communications (ICC) , IEEE, 2012, pp. 6151–6156. [85] A. Afsharinejad, A. Davy, B. Jennings, and C. Brennan, “Per- formance analysis of plant monitoring nanosensor networks at THz frequencies, ” IEEE Internet of Things Journal , vol. 3, no. 1, pp. 59–69, 2016. [86] V . Petrov, “Feasibility study of the THz band for communications between wearable electronics, ” in Open Innovations Association (FR UCT), 2015 17th Conference of , IEEE, 2015, pp. 157–162. [87] J. Simonjan, J. M. Jornet, I. F . Akyildiz, and B. Rinner, “Nano- cameras: A key enabling technology for the internet of multimedia nano-things, ” in ACM Nanoscale Computing and Communication , A CM, 2018, p. 28. [88] W . Knap, G Cywinski, M Sypek, N Dyakonov a, D. Coquillat, K Szkudlarek, I Y ahniuk, C. Archier, B. Moulin, M. Triki, et al. , “THz imaging and wireless communication using nanotransistor based detectors: From basic physics to ﬁrst real world applications, ” in International Confer ence on T ranspar ent Optical Networks (IC- TON) , IEEE, 2017, pp. 1–5. [89] M. Sitti, H. Ceylan, W . Hu, J. Giltinan, et al. , “Biomedical appli- cations of untethered mobile milli/microrobots, ” Pr oceedings of the IEEE , vol. 103, no. 2, pp. 205–224, 2015. [90] N. Agoulmine, K. Kim, S. Kim, T . Rim, J.-S. Lee, and M. Meyyap- pan, “Enabling communication and cooperation in bio-nanosensor networks: T o ward innov ative healthcare solutions, ” IEEE Wir eless Communications , vol. 19, no. 5, 2012. [91] G. Piro, K. Y ang, G. Boggia, N. Chopra, L. A. Grieco, and A. Alo- mainy , “T erahertz communications in human tissues at the nanoscale for healthcare applications, ” IEEE T ransactions on Nanotechnology , vol. 14, no. 3, pp. 404–406, 2015. [92] N. A. Ali and M. Ab u-Elkheir, “Internet of nano-things health- care applications: Requirements, opportunities, and challenges, ” in W ir eless and Mobile Computing, Networking and Communications (W iMob) , IEEE, 2015, pp. 9–14. [93] M. A. Lewis and G. A. Bekey, “The behavioral self-organization of nanorobots using local rules, ” in IEEE/RSJ Intelligent Robots and Systems (IROS) , 1992, pp. 1333–1338. [94] R. Asorey-Cacheda, S. Canov as-Carrasco, A.-J. Garcia-Sanchez, and J. Garcia-Haro, “An analytical approach to ﬂow-guided nanocom- munication networks, ” Sensors , vol. 20, no. 5, p. 1332, 2020. [95] R. Asorey-Cacheda, A.-J. Garcia-Sanchez, and J. Garcia-Haro, “Throughput optimization in ﬂow-guided nanocommunication net- works, ” IEEE Access , vol. 8, pp. 142 875–142 891, 2020. [96] ——, “On the feasibility of ﬂow-guided nanocommunication net- works for some medical applications, ” in 1st ACM International W orkshop on Nanoscale Computing, Communication, and Applica- tions (NanoCoCoA’20) , A CM, 2020, pp. 1–7. [97] B. Bohnenstiehl, A. Stillmaker, J. Pimentel, T . Andreas, B. Liu, A. T ran, E. Adeagbo, and B. Baas, “KiloCore: A Fine-Grained 1,000-Processor Array for T ask-Parallel Applications, ” IEEE Micr o , vol. 37, no. 2, pp. 63–69, 2017. [98] S. Abadal, A. Mestres, J. T orrellas, E. Alarc ´ on, and A. Cabellos- Aparicio, “Medium Access Control in W ireless Network-on-Chip: A Context Analysis, ” IEEE Communications Magazine , vol. 56, no. 6, pp. 172–178, 2018. [99] R. Marculescu, U. Ogras, L.-S. Peh, N. Enright Jer ger, and Y . Hoskote, “Outstanding research problems in NoC design: system, microarchitecture, and circuit perspectiv es, ” IEEE T ransactions on Computer-Aided Design of Inte grated Circuits and Systems , v ol. 28, no. 1, pp. 3–21, 2009. [100] “On-chip networks from a networking perspective: congestion and scalability in many-core interconnects, ” in Pr oceedings of the SIG- COMM , 2012, pp. 407–18. [101] D. Bertozzi, G. Dimitrakopoulos, J. Flich, and S. Sonntag, “The fast ev olving landscape of on-chip communication, ” Design Automation for Embedded Systems , v ol. 19, no. 1, pp. 59–76, 2015, I S SN : 0929- 5585. D O I : 10.1007/s10617- 014- 9137- 6. [102] J. Kim, K. Choi, and G. Loh, “Exploiting new interconnect technolo- gies in on-chip communication, ” IEEE Journal on Emerging and Selected T opics in Circuits and Systems , v ol. 2, no. 2, pp. 124–136, 2012, I S SN : 21563357. D O I : 10.1109/JETCAS.2012.2201031. [103] A. Ganguly , K. Chang, S. Deb, P . P . Pande, B. Belzer, and C. T euscher, “Scalable hybrid wireless network-on-chip architectures for multicore systems, ” IEEE T ransactions on Computers , vol. 60, no. 10, pp. 1485–1502, 2011. [104] C. W ang, W .-H. Hu, and N. Bagherzadeh, “A wireless network- on-chip design for multicore platforms, ” in 2011 19th International Eur omicr o conference on parallel, distributed and network-based pr ocessing , IEEE, 2011, pp. 409–416. [105] S. Abadal, E. Alarc ´ on, A. Cabellos-Aparicio, M. C. Lemme, and M. Nemirovsky , “Graphene-enabled wireless communication for massiv e multicore architectures, ” IEEE Communications Magazine , vol. 51, no. 11, pp. 137–143, 2013. [106] S. Saxena, D. S. Manur, et al. , “A folded wireless network-on- chip using graphene based THz-band antennas, ” in ACM Nanoscale Computing and Communication , A CM, 2017, p. 29. [107] S. Saxena, D. S. Manur, N. Mansoor, and A. Ganguly, “Scalable and energy efﬁcient wireless inter chip interconnection fabrics using thz-band antennas, ” J ournal of P arallel and Distributed Computing , 2020. [108] A. Mestres, S. Abadal, J. T orrellas, E. Alarc ´ on, and A. Cabellos- Aparicio, “A MA C protocol for Reliable Broadcast Communications in W ireless Network-on-Chip, ” in Proceedings of the NoCArc ’16 , 2016, pp. 21–26. [109] H. Kwon et al. , “Rethinking NoCs for Spatial Neural Netw ork Accelerators, ” in Pr oceedings of the NoCS ’17 , 2017, Art. 19. [110] J. Postman and P . Chiang, “A survey addressing on-chip inter - connect: Energy and reliability considerations, ” ISRN Electronics , vol. 2012, 2012. [111] B. Lebiednik, S. Abadal, H. Kwon, and T . Krishna, “Architecting a Secure Wireless Network-on-Chip, ” in Pr oceedings of the NOCS ’18 , 2018. [112] O Y algashe v , M. Bakhouya, A. Chariete, and J. Gaber, “Performance ev aluation of nano-on-chip interconnect for SoCs, ” in High P erfor- mance Computing & Simulation (HPCS) , IEEE, 2014, pp. 512–515. [113] S. Abadal, B. Sheinman, O. Katz, O. Markish, D. Elad, Y . Fournier, D. Roca, M. Hanzich, G. Houzeaux, M. Nemirovsky , et al. , “Broadcast-enabled massiv e multicore architectures: A wireless RF approach, ” IEEE micr o , vol. 35, no. 5, pp. 52–61, 2015. [114] J. M. Jornet, J. C. Pujol, and J. S. Pareta, “Phlame: A physical layer aware mac protocol for electromagnetic nanonetworks in the terahertz band, ” Nano Communication Networks , vol. 3, no. 1, pp. 74–81, 2012. [115] P . W ang, J. M. Jornet, M. A. Malik, N. Akkari, and I. F . Akyildiz, “Energy and spectrum-aware MAC protocol for perpetual wireless nanosensor networks in the terahertz band, ” Ad Hoc Networks , vol. 11, no. 8, pp. 2541–2555, 2013. [116] H. Elayan, O. Amin, B. Shihada, R. M. Shubair, and M.-S. Alouini, “T erahertz band: The last piece of rf spectrum puzzle for communi- cation systems, ” IEEE Open Journal of the Communications Society , vol. 1, pp. 1–32, 2019. [117] J. N. Al-Karaki and A. E. Kamal, “Routing techniques in wireless sensor networks: A survey, ” IEEE wir eless communications , vol. 11, no. 6, pp. 6–28, 2004. [118] Q. Xia and J. M. Jornet, “Cross-layer analysis of optimal relaying strategies for terahertz-band communication networks, ” in W ir eless and Mobile Computing, Networking and Communications (W iMob), , IEEE, 2017, pp. 1–8. 34 [119] Z. Rong, M. S. Leeson, and M. D. Higgins, “Relay-assisted nanoscale communication in the THz band, ” Micro & Nano Letters , vol. 12, no. 6, pp. 373–376, 2017. [120] H. Y u, B. Ng, and W . K. Seah, “Forwarding schemes for EM-based wireless nanosensor networks in the terahertz band, ” in Nanoscale Computing and Communication , A CM, 2015, p. 17. [121] H.-H. Y en, “Energy aware and signal quality aware data aggregation touting in wireless nanosensor networks, ” in Consumer Communica- tions & Networking Conference (CCNC), 2017 14th IEEE Annual , IEEE, 2017, pp. 879–884. [122] M. Pierobon, J. M. Jornet, N. Akkari, S. Almasri, and I. F . Akyildiz, “A routing framework for energy harvesting wireless nanosensor networks in the terahertz band, ” W ir eless networks , vol. 20, no. 5, pp. 1169–1183, 2014. [123] S. Cano vas-Carrasco, A.-J. Garcia-Sanchez, and J. Garcia-Haro, “A nanoscale communication network scheme and energy model for a human hand scenario, ” Nano communication networks , vol. 15, pp. 17–27, 2018. [124] S. Canov as-Carrasco, R. M. Sandov al, A. Garcia-Sanchez, and J. Garcia-Haro, “Optimal transmission policy derivation for iont ﬂo w- guided nano-sensor networks, ” IEEE Internet of Things Journal , vol. 6, no. 2, pp. 2288–2298, 2019. D O I : 10 . 1109 / JIO T . 2019 . 2906015. [125] C. Liaskos and A. Tsioliaridou, “A promise of realizable, ultra- scalable communications at nano-scale: A multi-modal nano- machine architecture, ” IEEE Tr ansactions on Computers , vol. 64, no. 5, pp. 1282–1295, 2015. [126] A. Tsioliaridou, C. Liaskos, S. Ioannidis, and A. Pitsillides, “Lightweight, self-tuning data dissemination for dense nanonet- works, ” Nano Communication Networks , vol. 8, pp. 2–15, 2016. [127] F . Afsana, S. Mamun, M. Kaiser, and M. Ahmed, “Outage ca- pacity analysis of cluster-based forwarding scheme for body area network using nano-electromagnetic communication, ” in Electrical Information and Communication T echnology (EICT) , IEEE, 2015, pp. 383–388. [128] M. Stelzner, F . Busse, and S. Ebers, “In-body nanonetwork routing based on MANET and THz, ” in ACM Nanoscale Computing and Communication , ACM, 2018, p. 9. [129] F . B ¨ uther, I. Traupe, and S. Ebers, “Hop count routing: A routing algorithm for resource constrained, identity-free medical nanonet- works, ” in ACM Nanoscale Computing and Communication , A CM, 2018, p. 17. [130] F . Al-T urjman, “A cognitive routing protocol for bio-inspired net- working in the Internet of Nano-Things (IoNT), ” Mobile Networks and Applications , pp. 1–15, 2017. [131] M. Radetzki, C. Feng, X. Zhao, and A. Jantsch, “Methods for fault tolerance in networks-on-chip, ” ACM Computing Surveys (CSUR) , vol. 46, no. 1, p. 8, 2013. [132] M. Fattah, A. Airola, R. Ausavarungnirun, N. Mirzaei, P . Lilje- berg, J. Plosila, S. Mohammadi, T . Pahikkala, O. Mutlu, and H. T enhunen, “A low-ov erhead, fully-distributed, guaranteed-deliv ery routing algorithm for faulty network-on-chips, ” in Proceedings of the 9th International Symposium on Networks-on-Chip , ser . NOCS ’15, V ancouver , BC, Canada: A CM, 2015, 18:1–18:8. [133] Y . Fukushima, M. Fukushi, and S. Horiguchi, “Fault-tolerant routing algorithm for network on chip without virtual channels, ” in 2009 24th IEEE International Symposium on Defect and F ault T oler ance in VLSI Systems , 2009, pp. 313–321. [134] S. Jovano vic, C. T anoug ast, S. W eber , and C. Bobda, “A new deadlock-free fault-tolerant routing algorithm for noc intercon- nections, ” in Field Progr ammable Logic and Applications , 2009, pp. 326–331. [135] M. Ebrahimi, M. Daneshtalab, J. Plosila, and F . Mehdipour, “Md: Minimal path-based fault-tolerant routing in on-chip networks, ” in Design Automation Conference (ASP-DA C), 2013 18th Asia and South P aciﬁc , IEEE, 2013, pp. 35–40. [136] T . Saeed, C. Skitsas, D. K ouzapas, M. Lestas, V . Soteriou, A. Philippou, S. Abadal, C. Liaskos, L. Petrou, J. Georgiou, et al. , “Fault adaptive routing in metasurface controller networks, ” in 2018 11th International W orkshop on Network on Chip Ar chitectur es (NoCAr c) , IEEE, 2018, pp. 1–6. [137] A. Tsioliaridou, C. Liaskos, S. Ioannidis, and A. Pitsillides, “COR ONA: A coordinate and routing system for nanonetworks, ” in Nanoscale Computing and Communication , A CM, 2015, p. 18. [138] A. Tsioliaridou, C. Liaskos, E. Dedu, and S. Ioannidis, “Pack et routing in 3D nanonetworks: A lightweight, linear-path scheme, ” Nano communication networks , vol. 12, pp. 63–71, 2017. [139] S. W ang and T . Jin, “W ireless network-on-chip: A survey, ” The Journal of Engineering , vol. 2014, no. 3, pp. 98–104, 2014. [140] M. Stelzner, F . Dressler, and S. Fischer, “Function centric nano- networking: Addressing nano machines in a medical application scenario, ” Nano communication networks , vol. 14, pp. 29–39, 2017. [141] S. Xu, Y . Qin, C. Xu, Y . W ei, R. Y ang, and Z. L. W ang, “Self- powered nanowire devices, ” Natur e nanotechnology , vol. 5, no. 5, p. 366, 2010. [142] S. Mohrehkesh and M. C. W eigle, “Optimizing ener gy consumption in terahertz band nanonetworks, ” IEEE Journal on Selected Areas in Communications , vol. 32, no. 12, pp. 2432–2441, 2014. [143] A. I. Al-Sulaifanie, S. Biswas, and B. Khorsheed Al-Sulaifanie, “AH-MA C: adaptiv e hierarchical MA C protocol for low-rate wire- less sensor network applications, ” Journal of Sensors , vol. 2017, 2017. [144] N. Akkari, P . W ang, J. M. Jornet, E. Fadel, L. Elrefaei, M. G. A. Malik, S. Almasri, and I. F . Ak yildiz, “Distributed timely throughput optimal scheduling for the Internet of Nano-Things, ” IEEE Internet of Things Journal , vol. 3, no. 6, pp. 1202–1212, 2016. [145] R. Alsheikh, N. Akkari, and E. Fadel, “Grid based energy-aw are MA C protocol for wireless nanosensor network, ” in New T echnolo- gies, Mobility and Security (NTMS) , IEEE, 2016, pp. 1–5. [146] S. Mohrehkesh, M. C. W eigle, and S. K. Das, “DRIH-MA C: A distributed recei ver-initiated harvesting-a ware MA C for nanonet- works, ” IEEE T ransactions on Molecular , Biological and Multi- Scale Communications , vol. 1, no. 1, pp. 97–110, 2015. [147] S. D’Oro, L. Galluccio, G. Morabito, and S. Palazzo, “A timing channel-based MA C protocol for energy-efﬁcient nanonetworks, ” Nano Communication Networks , vol. 6, no. 2, pp. 39–50, 2015. [148] Q. Xia, Z. Hossain, M. Medle y , and J. M. Jornet, “A link-layer syn- chronization and medium access control protocol for terahertz-band communication networks, ” in 2015 IEEE Global Communications Confer ence (GLOBECOM) , IEEE, 2015, pp. 1–7. [149] G. Piro, L. A. Grieco, G. Boggia, and P . Camarda, “Nano-Sim: Simulating electromagnetic-based nanonetworks in the network sim- ulator 3, ” in ICST Confer ence on Simulation T ools and T echniques , ICST (Institute for Computer Sciences, Social-Informatics and . . ., 2013, pp. 203–210. [150] H. Y u, B. Ng, and W . K. Seah, “Pulse arriv al scheduling for nanonetworks under limited iot access bandwidth, ” in 2017 IEEE 42nd Conference on Local Computer Networks (LCN) , IEEE, 2017, pp. 18–26. [151] N. Rikhtegar , M. Keshtgari, and Z. Ronaghi, “EEWNSN: Energy efﬁcient wireless nano sensor netw ork MA C protocol for communi- cations in the terahertz band, ” W ireless P ersonal Communications , vol. 97, no. 1, pp. 521–537, 2017. [152] V Srikanth, S. Chaluvadi, et al. , “Energy efﬁcient, scalable and reliable mac protocol for electromagnetic communication among nano de vices, ” International J ournal of Distributed and P ar allel Systems , vol. 3, no. 1, p. 249, 2012. [153] S. J. Lee, H. Choi, and S. Kim, “Slotted csma/ca based energy ef- ﬁcinet mac protocol design in nanonetworks, ” International Journal of W ir eless & Mobile Networks (IJWMN) , vol. 10, 2018. [154] N. Mansoor , S. Shamim, and A. Ganguly , “A Demand-A ware Predictiv e Dynamic Bandwidth Allocation Mechanism for Wireless Network-on-Chip, ” in Proceedings of the SLIP ’16 , 2016. [155] N. Mansoor and A. Ganguly, “Reconﬁgurable W ireless Network-on- Chip with a Dynamic Medium Access Mechanism, ” in Pr oceedings of the NoCS ’15 , 2015, Article 13. [156] S. Mohrehkesh and M. C. W eigle, “RIH-MA C: Receiver-initiated harvesting-aw are MA C for nanonetworks, ” in A CM Nanoscale Com- puting and Communication , A CM, 2014, p. 6. [157] F . Lemic, R. Usman Akbar, J. Marquez-Barja, and J. Famaey, “ As- sessing the reliability of energy harvesting terahertz nanonetworks for controlling software-deﬁned metamaterials, ” in A CM Nanoscale Computing and Communication (NanoCom 2019) , A CM, 2019. [158] X. W u and K. Sengupta, “Dynamic W a veform Shaping W ith Picosecond Time Widths, ” IEEE Journal of Solid-State Circuits , vol. 52, no. 2, pp. 389–405, 2017. [159] S. Lee, R. Dong, T . Y oshida, S. Amakawa, S. Hara, A. Kasamatsu, J. Sato, and M. Fujishima, “An 80Gb/s 300GHz-Band Single-Chip CMOS T ranscei ver, ” in Pr oceedings of the ISSCC ’19 , IEEE, 2019, pp. 170–172. [160] H. Aggraw al, P . Chen, and M. M. Assefzadeh, “Gone in a Picosec- ond: T echniques for the Generation and Detection of Picosecond Pulses and Their Applications, ” IEEE Microwave Magazine , vol. 17, no. 12, pp. 24–38, 2016. 35 [161] R. Han et al. , “Filling the gap: Silicon terahertz integrated circuits offer our best bet, ” IEEE Micr owave Ma gazine , vol. 20, no. 4, pp. 80–93, 2019. [162] X. Y u, J. Baylon, P . W ettin, D. Heo, P . Pratim Pande, and S. Mirabbasi, “ Architecture and Design of Multi-Channel Millimeter- W ave Wireless Network-on-Chip, ” IEEE Design & T est , vol. 31, no. 6, pp. 19–28, 2014. [163] S. Hu, Y .-Z. Xiong, B. Zhang, L. W ang, T .-G. Lim, M. Je, and M. Madihian, “A sige bicmos transmitter/receiver chipset with on- chip siw antennas for terahertz applications, ” IEEE Journal of Solid- State Circuits , vol. 47, no. 11, pp. 2654–2664, 2012. [164] A. C. T asolamprou, A. Pitilakis, S. Abadal, O. Tsilipakos, X. T imoneda, H. T aghvaee, M. S. Mirmoosa, F . Liu, C. Liaskos, A. Tsioliaridou, S. Ioannidis, N. V . Kantartzis, D. Manessis, J. Georgiou, A. Cabellos-Aparicio, E. Alarc ´ on, A. Pitsillides, I. F . Akyildiz, S. A. T retyakov, E. N. Economou, M. Kafesaki, and C. M. Soukoulis., “Exploration of intercell wireless millimeter-wav e communication in the landscape of intelligent metasurfaces, ” IEEE Access , vol. PP, no. 99, pp. 1–1, 2019. [165] J. M. Jornet and I. F . Akyildiz, “Femtosecond-long pulse-based modulation for terahertz band communication in nanonetworks, ” IEEE Communications , vol. 62, no. 5, pp. 1742–1754, 2014. [166] A. P . Shrestha, S.-J. Y oo, H. J. Choi, and K. S. Kwak, “Enhanced rate division multiple access for electromagnetic nanonetworks, ” IEEE Sensors Journal , vol. 16, no. 19, pp. 7287–7296, 2016. [167] H. Mabed, “Enhanced spread in time ON-OFF ke ying technique for dense terahertz nanonetworks, ” in Computers and Communications (ISCC), 2017 IEEE Symposium on , IEEE, 2017, pp. 710–716. [168] H. Mabed and J. Bour geois, “A ﬂexible medium access control protocol for dense terahertz nanonetworks, ” in A CM Nanoscale Computing and Communication , A CM, 2018, p. 10. [169] A. K. V avouris, F . D. Dervisi, V . K. Papanikolaou, and G. K. Karagiannidis, “An energy efﬁcient modulation scheme for body- centric nano-communications in the THz band, ” in Modern Cir cuits and Systems T ec hnologies (MOCAST) , IEEE, 2018, pp. 1–4. [170] E. Zarepour , M. Hassan, C. T . Chou, and A. A. Adesina, “Frequency hopping strategies for improving terahertz sensor network perfor- mance over composition varying channels, ” in W orld of W ir eless, Mobile and Multimedia Networks (W oWMoM), 2014 IEEE 15th International Symposium on a , IEEE, 2014, pp. 1–9. [171] C. Han, A. O. Bicen, and I. F . Akyildiz, “Multi-wideband wave- form design for distance-adaptiv e wireless communications in the terahertz band, ” IEEE T ransactions on Signal Processing , vol. 64, no. 4, pp. 910–922, 2015. [172] J. M. Jornet, “Low-weight error-pre vention codes for electromag- netic nanonetworks in the terahertz band, ” Nano Communication Networks , vol. 5, no. 1-2, pp. 35–44, 2014. [173] M. K ocaoglu and O. B. Akan, “Minimum energy channel codes for nanoscale wireless communications, ” IEEE T ransactions on W ir eless Communications , vol. 12, no. 4, pp. 1492–1500, 2013. [174] K. Chi, Y .-h. Zhu, X. Jiang, and V . C. Leung, “Ener gy-efﬁcient preﬁx-free codes for wireless nano-sensor networks using OOK modulation, ” IEEE T ransactions on W ir eless Communications , vol. 13, no. 5, pp. 2670–2682, 2014. [175] M. A. Zainuddin, E. Dedu, and J. Bourgeois, “SBN: Simple block nanocode for nanocommunications, ” in A CM Nanoscale Computing and Communication , ACM, 2016, p. 4. [176] P . Singh, B. W . Kim, and S.-Y . Jung, “Ds-ook for terahertz band nanonetworks, ” National Academy Science Letters , pp. 1–4, 2020. [177] S. E. Hosseininejad, S. Abadal, M. Neshat, R. Faraji-Dana, M. C. Lemme, C. Suessmeier, P . Haring Bol ´ ıvar, E. Alarc ´ on, and A. Cabellos-Aparicio, “MA C-Oriented Programmable T erahertz PHY via Graphene-based Y agi-Uda Antennas, ” in Pr oceedings of the WCNC ’18 , 2018. [178] J. Lin and M. A. W eitnauer, “Pulse-level beam-switching for tera- hertz networks, ” W ir eless Networks , vol. 7, pp. 1–16, 2018. [179] R. G. Cid-fuentes, J. M. Jornet, I. F . Akyildiz, and E. Alarc ´ on, “A Receiver Architecture for Pulse-based Electromagnetic Nanonet- works in the T erahertz Band, ” in Pr oceedings of the ICC ’12 , 2012, pp. 4937–4942. [180] P . Singh, B. Kim, and S.-Y . Jung, “Non-coherent symbol detec- tion with T oA estimation for nanocommunication networks, ” 2019, pp. 1–5. [181] A. Gupta, M. Medle y , and J. M. Jornet, “Joint synchronization and symbol detection design for pulse-based communications in the THz band, ” in 2015 IEEE Global Communications Confer ence (GLOBECOM) , IEEE, 2015, pp. 1–7. [182] M. O. Iqbal, M. M. U. Rahman, M. A. Imran, et al. , “Modulation mode detection and classiﬁcation for in vivo nano-scale communi- cation systems operating in terahertz band, ” IEEE transactions on nanobioscience , vol. 18, no. 1, pp. 10–17, 2018. [183] S. Laha, S. Kaya, D. W . Matolak, W . Rayess, D. DiT omaso, and A. K odi, “A New Frontier in Ultralow Power W ireless Links: Network- on-Chip and Chip-to-Chip Interconnects, ” IEEE T ransactions on Computer-Aided Design of Integr ated Circuits and Systems , v ol. 34, no. 2, pp. 186–198, 2015. [184] C. Han and I. F . Akyildiz, “Distance-aware multi-carrier (damc) modulation in terahertz band communication, ” in 2014 IEEE In- ternational Confer ence on Communications (ICC) , IEEE, 2014, pp. 5461–5467. [185] E. Zarepour, M. Hassan, C. T . Chou, and S. Bayat, “Performance analysis of carrier-less modulation schemes for wireless nanosensor networks, ” in IEEE International Conference on Nanotechnology (IEEE-NANO) , IEEE, 2015, pp. 45–50. [186] K W itrisal, G. Leus, G. J. M. Janssen, M. Pausini, F . Troesch, T . Zasowski, and J. Romme, “Noncoherent ultra-wideband systems, ” IEEE Signal Pr ocessing Magazine , vol. 26, no. 4, pp. 48–66, 2009. [187] Z. Hossain, C. N. Mollica, et al. , “Stochastic interference modeling and experimental validation for pulse-based terahertz communica- tion, ” IEEE T ransactions on W ir eless Communications , vol. 18, no. 8, pp. 4103–4115, 2019. [188] P . a. Singh, “TH-PPM with non-coherent detection for multiple access in electromagnetic wireless nanocommunications, ” Nano Communication Networks , vol. 17, pp. 1–13, 2018. [189] J. M. Jornet and I. F . Akyildiz, “Low-weight channel coding for interference mitigation in electromagnetic nanonetworks in the tera- hertz band, ” in IEEE International Conference on Communications (ICC) , IEEE, 2011, pp. 1–6. [190] K. Chi, Y .-h. Zhu, X. Jiang, and X. T ian, “Optimal coding for transmission energy minimization in wireless nanosensor networks, ” Nano Communication Networks , vol. 4, no. 3, pp. 120–130, 2013. [191] M. A. Zainuddin, E. Dedu, and J. Bourgeois, “Low-weight code comparison for electromagnetic wireless nanocommunication, ” IEEE Internet of Things Journal , vol. 3, no. 1, pp. 38–48, 2016. [192] X. Y ao, D. Ma, and C. Han, “ECP: A probing-based error control strategy for THz-based nanonetworks with energy harvesting, ” IEEE Access , vol. 7, pp. 25 616–25 626, 2019. [193] A. Cabellos-Aparicio, I. Llatser, E. Alarc ´ on, A. Hsu, and T . P alacios, “Use of THz Photoconductive Sources to Characterize T unable Graphene RF Plasmonic Antennas, ” IEEE T ransactions on Nan- otechnology , vol. 14, no. 2, pp. 390–396, 2015. eprint: 1401.6878. [194] M. Samizadeh Nikoo, A. Jafari, N. Perera, M. Zhu, G. Santoruvo, and E. Matioli, “Nanoplasma-enabled picosecond switches for ul- trafast electronics, ” Natur e , vol. 579, no. 7800, pp. 534–539, 2020, I S SN : 14764687. D O I : 10.1038/s41586- 020- 2118- y. [195] B. Sensale-Rodriguez, R. Y an, M. M. Kelly, T . Fang, K. T ahy, W . S. Hwang, D. Jena, L. Liu, and H. G. Xing, “Broadband graphene terahertz modulators enabled by intraband transitions, ” Natur e communications , vol. 3, p. 780, 2012. [196] R. Piesiewicz, T . Kleine-Ostmann, N. Krumbholz, D. Mittleman, M. K och, J. Schoebel, and T . Kurner, “Short-Range Ultra-Broadband T erahertz Communications: Concepts and Perspectives, ” IEEE An- tennas and Pr opa gation Magazine , v ol. 49, no. 6, pp. 24–39, 2007. [197] J. Kokk oniemi, J. Lehtomaki, K. Umebayashi, and M. Juntti, “Fre- quency and Time Domain Channel Models for Nanonetworks in T erahertz Band, ” IEEE Tr ansactions on Antennas and Pr opagation , vol. 63, no. 2, pp. 678–691, 2015. [198] R. Piesiewicz, C. Jansen, S. Wietzke, D. Mittleman, M. K och, and T . K ¨ urner, “Properties of Building and Plastic Materials in the THz Range, ” International J ournal of Infrar ed and Millimeter W aves , vol. 28, no. 5, pp. 363–371, 2007. [199] J. M. Jornet and I. F . Akyildiz, “Channel modeling and capacity analysis for electromagnetic wireless nanonetworks in the terahertz band, ” IEEE T r ansactions on W ir eless Communications , vol. 10, no. 10, pp. 3211–3221, 2011. [200] I. Javed and I. H. Naqvi, “Frequency band selection and channel modeling for WNSN applications using simplenano, ” in IEEE International Confer ence on Communications (ICC) , Ieee, 2013, pp. 5732–5736. [201] P . Boronin, V . Petrov, D. Moltchanov, et al. , “Capacity and throughput analysis of nanoscale machine communication through transparency windows in the terahertz band, ” Nano Communication Networks , vol. 5, no. 3, pp. 72–82, 2014. 36 [202] I. Llatser, A. Mestres, S. Abadal, E. Alarc ´ on, et al. , “T ime and frequency domain analysis of molecular absorption in short-range terahertz communications, ” IEEE Antennas and W ir eless Propa ga- tion Letters , vol. 14, pp. 350–353, 2015. [203] C. Jansen, S. Priebe, and C Moller, “Diffuse scattering from rough surfaces in THz communication channels, ” IEEE T erahertz Science and T echnology , vol. 1, no. 2, pp. 462–472, 2011. [204] J. K okkoniemi, V . Petrov, D. Moltchanov, J. Lehtomaki, Y . K ouch- eryavy , and M. Juntti, “Wideband T erahertz Band Reﬂection and Diffuse Scattering Measurements for Beyond 5G Indoor W ireless Networks, ” in Proceedings of the EW ’16 , 2016. [205] A. Fricke, T . Kurner, M. Achir, and P . Le Bars, “A model for the reﬂection of terahertz signals from printed circuit board surfaces, ” in Proceedings of the EuCAP ’17 , 2017, pp. 711–715. [206] C. Han, A. Ozan Bicen, and I. F . Akyildiz, “Multi-Ray Channel Modeling and Wideband Characterization for W ireless Commu- nications in the T erahertz Band, ” IEEE T ransactions on W ir eless Communications , vol. 14, no. 5, pp. 2402–2412, 2015. [207] C. Han and I. F . Akyildiz, “Three-Dimensional End-to-End Model- ing and Analysis for Graphene-Enabled T erahertz Band Communi- cations, ” IEEE T ransactions on V ehicular T echnology , vol. 66, no. 7, pp. 5626–5634, 2017. [208] C. Han and Y . Chen, “Propagation Modeling for Wireless Commu- nications in the T erahertz Band, ” IEEE Communications Magazine , vol. 56, no. 6, pp. 96–101, 2018. [209] S. Abadal, A. Mestres, I. Llatser, E. Alarc ´ on, and A. Cabellos- aparicio, “A V ertical Methodology for the Design Space Exploration of Graphene-enabled W ireless Communications, ” in Pr oceedings of the NANOCOM ’15 , 2015. [210] V . Petrov et al. , “Interference and SINR in millimeter wave and terahertz communication systems with blocking and directional antennas, ” IEEE T ransactions on W ireless Communications , vol. 16, no. 3, pp. 1791–1808, 2017. [211] K. Y ang, A. Pellegrini, A. Brizzi, A. Alomainy , and Y . Hao, “Numerical analysis of the communication channel path loss at the THz band inside the fat tissue, ” in Micr owave W orkshop Series on RF and Wir eless T echnologies for Biomedical and Healthcare Applications (IMWS-BIO), 2013 IEEE MTT -S International , IEEE, 2013, pp. 1–3. [212] K. Y ang, A. Pellegrini, M. O. Munoz, et al. , “Numerical analysis and characterization of THz propagation channel for body-centric nano-communications, ” IEEE T ransactions on T erahertz Science and technology , vol. 5, no. 3, pp. 419–426, 2015. [213] H. Elayan, R. M. Shubair, et al. , “Multi-layer intrabody terahertz wav e propagation model for nanobiosensing applications, ” Nano Communication Networks , vol. 14, pp. 9–15, 2017. [214] R. Zhang, Q. H. Abbasi, and A. Alomainy, “Po wer distrib ution and performance analysis of terahertz communication in artiﬁcial skin, ” in Pr oceedings of the Sixth Annual A CM International Conference on Nanoscale Computing and Communication , 2019, pp. 1–5. [215] J. Kokkoniemi, P . Rintanen, J. Lehtom ¨ aki, and M. Juntti, “Diffrac- tion effects in terahertz band - Measurements and analysis, ” in Pr oceedings of the GLOBECOM ’16 , 2016. [216] H. Elayan, C. Stefanini, R. M. Shubair, and J. M. Jornet, “Stochastic noise model for intra-body terahertz nanoscale communication, ” in ACM Nanoscale Computing and Communication , A CM, 2018, p. 8. [217] E. Zarepour, M. Hassan, C. T . Chou, A. A. Adesina, and M. E. W arkiani, “Reliability analysis of time-varying wireless nanoscale sensor networks, ” in IEEE International Conference on Nanotech- nology (IEEE-NANO) , IEEE, 2015, pp. 63–68. [218] J. Fu, P . Juyal, and A. Zaji ´ c, “Modeling of 300 ghz chip-to-chip wireless channels in metal enclosures, ” IEEE, 2020. [219] M. Opoku Agyeman, Q.-T . Vien, A. Ahmadnia, et al. , “A Resilient 2-D W a veguide Communication Fabric for Hybrid Wired-W ireless NoC Design, ” IEEE T ransactions on P arallel and Distributed Sys- tems , vol. 28, no. 2, pp. 359–373, 2016. [220] I. El Masri, T . Le Gouguec, P .-M. Martin, R. Allanic, and C. Quendo, “Em analysis of a propagation channel in the sub-thz band for many-core architectures, ” in 2019 49th European Microwave Confer ence (EuMC) , IEEE, 2019, pp. 972–975. [221] Y . Chen, X. Cai, C. Han, and T . Link, “W av e Propagation Modeling for mmW a ve and T erahertz Wireless Networks-on-Chip Communi- cations, ” in Pr oceedings of the ICC ’19 , IEEE, 2019. [222] Y . Chen and C. Han, “Channel modeling and characterization for wireless networks-on-chip communications in the millimeter wav e and terahertz bands, ” IEEE T ransactions on Molecular , Biological and Multi-Scale Communications , 2019. [223] Y . P . Zhang, Z. M. Chen, and M. Sun, “Propagation Mechanisms of Radio W aves Over Intra-Chip Channels W ith Integrated An- tennas: Frequency-Domain Measurements and Time-Domain Anal- ysis, ” IEEE T ransactions on Antennas and Propa gation , vol. 55, no. 10, pp. 2900–2906, 2007. [224] R. S. Narde, J. V enkataraman, et al. , “Intra-and Inter-Chip Trans- mission of Millimeter-W ave Interconnects in NoC-based Multi-Chip Systems, ” IEEE Access , vol. PP, pp. 1–1, 2019. [225] R. Zhang, K. Y ang, A. Alomainy, Q. H. Abbasi, K. Qaraqe, and R. M. Shubair, “Modelling of the terahertz communication channel for in-vi vo nano-networks in the presence of noise, ” in Microwave Symposium (MMS), 2016 16th Mediterranean , IEEE, 2016, pp. 1–4. [226] R. Zhang, K. Y ang, Q. H. Abbasi, K. A. Qaraqe, and A. Alomainy, “Analytical characterisation of the terahertz in-viv o nano-network in the presence of interference based on TS-OOK communication scheme, ” IEEE Access , vol. 5, pp. 10 172–10 181, 2017. [227] E. Zarepour, N. Hassan, M. Hassan, C. T . Chou, and M. E. W arkiani, “Design and analysis of a wireless nanosensor network for monitoring human lung cells, ” in EAI International Conference on Body Ar ea Networks , ICST (Institute for Computer Sciences, Social-Informatics and . . ., 2015, pp. 139–145. [228] D. Matolak, S Kaya, and A Kodi, “Channel modeling for wire- less networks-on-chips, ” IEEE Communications Magazine , vol. 51, no. 6, pp. 180–186, 2013. [229] S. Abadal, C. Han, and J. M. Jornet, “W a ve propagation and channel modeling in chip-scale wireless communications: A survey from millimeter-w ave to terahertz and optics, ” IEEE Access , vol. 8, pp. 278–293, 2019. [230] S. Kim and A. Zajic, “Characterization of 300 GHz Wireless Chan- nel on a Computer Motherboard, ” IEEE T ransactions on Antennas and Propa gation , vol. 64, no. 12, pp. 5411–5423, 2016. [231] L. Y an and G. W . Hanson, “W ave propagation mechanisms for intra-chip communications, ” IEEE T ransactions on Antennas and Pr opagation , vol. 57, no. 9, pp. 2715–2724, 2009. [232] W . Rayess, D. W . Matolak, S. Kaya, and A. K. Kodi, “Antennas and Channel Characteristics for W ireless Networks on Chips, ” W ir eless P ersonal Communications , vol. 95, no. 4, pp. 5039–5056, 2017. [233] X. Timoneda, A. Cabellos-Aparicio, D. Manessis, E. Alarc ´ on, and S. Abadal, “Channel Characterization for Chip-scale W ireless Communications within Computing Packages, ” in Proceedings of the NOCS ’18 , 2018. [234] X. Timoneda, S. Abadal, A. Franques, D. Manessis, J. Zhou, J. T or- rellas, E. Alarc ´ on, and A. Cabellos-Aparicio, “Engineer the channel and adapt to it: Enabling wireless intra-chip communication, ” IEEE T ransactions on Communications , 2020. [235] G. Piro, L. A. Grieco, G. Boggia, and P . Camarda, “Simulating wireless nano sensor networks in the ns-3 platform, ” in Advanced In- formation Networking and Applications W orkshops (W AINA) , IEEE, 2013, pp. 67–74. [236] N. Baldo and M. Miozzo, “Spectrum-a ware channel and PHY layer modeling for ns3, ” in F ourth International ICST Conference on P erformance Evaluation Methodologies and T ools , ICST (Institute for Computer Sciences, Social-Informatics and . . ., 2009, p. 2. [237] N. Boillot, D. Dhoutaut, and J. Bourgeois, “Scalable simulation of wireless electro-magnetic nanonetworks, ” in Embedded and Ubiq- uitous Computing (EUC) , IEEE, 2015, pp. 83–89. [238] ——, “Using nano-wireless communications in micro-robots appli- cations, ” in ACM Nanoscale Computing and Communication , ACM, 2014, p. 10. [239] ——, “Large scale mems robots cooperative map building based on realistic simulation of nano-wireless communications, ” Nano Communication Networks , vol. 6, no. 2, pp. 51–73, 2015. [240] B. D. Rister, J. Campbell, P . Pillai, and T . C. Mowry, “Integrated debugging of large modular robot ensembles, ” in IEEE Robotics and Automation , IEEE, 2007, pp. 2227–2234. [241] D. Dhoutaut, T . Arrabal, and E. Dedu, “Bit simulator, an electro- magnetic nanonetworks simulator , ” in ACM Nanoscale Computing and Communication , ACM, 2018, p. 30. [242] Z. Hossain, Q. Xia, and J. M. Jornet, “T eraSim: An ns-3 extension to simulate terahertz-band communication networks, ” Nano Com- munication Networks , vol. 17, pp. 36–44, 2018. [243] S. Canovas-Carrasco, A.-J. Garcia-Sanchez, and J. Garcia-Haro, “On the nature of energy-feasible wireless nanosensor networks, ” Sensors , vol. 18, no. 5, p. 1356, 2018. [244] J. M. Jornet and I. F . Akyildiz, “Joint energy harvesting and com- munication analysis for perpetual wireless nanosensor networks in 37 the terahertz band, ” IEEE T ransactions on Nanotechnology , vol. 11, no. 3, p. 570, 2012. [245] I. Llatser, D. Demiray , A. Cabellos-Aparicio, D. T . Altilar, and E. Alarc ´ on, “N3Sim: Simulation framework for diffusion-based molec- ular communication nanonetworks, ” Simulation Modelling Practice and Theory , vol. 42, pp. 210–222, 2014. [246] R. Geyer, M. Stelzner, F . B ¨ uther, and S. Ebers, “BloodV oyagerS: Simulation of the w ork en vironment of medical nanobots, ” in ACM Nanoscale Computing and Communication , ACM, 2018, p. 5. [247] S. Canov as-Carrasco, A.-J. Garcia-Sanchez, F . Garcia-Sanchez, and J. Garcia-Haro, “Conceptual design of a nano-networking device, ” Sensors , vol. 16, no. 12, p. 2104, 2016. [248] A. Tsioliaridou, C. Liask os, A. Pitsillides, and S. Ioannidis, “A nov el protocol for network-controlled metasurfaces, ” in ACM Nanoscale Computing and Communication , A CM, 2017, p. 3. [249] L. Georgiadis, M. J. Neely, L. T assiulas, et al. , “Resource allocation and cross-layer control in wireless networks, ” F oundations and T r ends® in Networking , vol. 1, no. 1, pp. 1–144, 2006. [250] M. Alam, N. Nurain, S. T airin, and A. A. Al Islam, “Energy- efﬁcient transport layer protocol for hybrid communication in body area nanonetworks, ” in Humanitarian T echnology Conference (R10- HTC), 2017 IEEE Re gion 10 , IEEE, 2017, pp. 674–677. [251] N. Hassan, C. T . Chou, and M. Hassan, “Event and node identiﬁca- tion from a single-pulse transmission in self-powered nanosensor networks, ” in ACM Nanoscale Computing and Communication , A CM, 2017, p. 2. [252] N. Hassan, C. T . Chou, and M. Hassan, “Eneutral iont: Energy- neutral e vent monitoring for internet of nano things, ” IEEE Internet of Things Journal , vol. 6, no. 2, pp. 2379–2389, 2019. [253] A. Tsioliaridou, C. Liaskos, and S. Ioannidis, “Bitsurﬁng: Wireless communications with outsourced symbol generation, ” in Computer Aided Modeling and Design of Communication Links and Networks (CAMAD) , IEEE, 2018, pp. 1–7. [254] S. Mohrehkesh and M. C. W eigle, “Optimizing communication energy consumption in perpetual wireless nanosensor networks, ” in Global Communications Confer ence (GLOBECOM), 2013 IEEE , IEEE, 2013, pp. 545–550. [255] S. Mohrehkesh et al. , “Energy harvesting in electromagnetic nanonetworks, ” Computer , vol. 50, no. 2, pp. 59–67, 2017. [256] F . Lemic, J. Marquez-Barja, and J. Famaey, “Modeling and reducing idling energy consumption in energy harvesting terahertz nanonet- works, ” in IEEE Global Communications Conference , IEEE, 2019. [257] M. Donohoe, B. Jennings, J. M. Jornet, and S. Balasubramaniam, “Nanodevice arrays for peripheral nerve fascicle activ ation using ul- trasound energy-harv esting, ” IEEE T ransactions on Nanotechnolo gy , vol. 16, no. 6, pp. 919–930, 2017. [258] Z. Rong, M. S. Leeson, M. D. Higgins, and Y . Lu, “Simultaneous wireless information and power transfer for af relaying nanonet- works in the terahertz band, ” Nano communication networks , vol. 14, pp. 1–8, 2017. [259] H. Y u, B. Ng, and W . K. Seah, “On-demand probabilistic polling for nanonetworks under dynamic IoT backhaul network conditions, ” IEEE Internet of Things Journal , vol. 4, no. 6, pp. 2217–2227, 2017. [260] G. Piro, G. Boggia, and L. A. Grieco, “On the design of an energy-harv esting protocol stack for body area nano-networks, ” Nano Communication Networks , vol. 6, no. 2, pp. 74–84, 2015. [261] J. Jarmakiewicz, K. Parobczak, and K. Ma ´ slanka, “On the Internet of Nano Things in healthcare network, ” in Military Communications and Information Systems (ICMCIS) , IEEE, 2016, pp. 1–6. [262] F . Dressler and F . Kargl, “T o wards security in nano-communication: Challenges and opportunities, ” Nano communication networks , vol. 3, no. 3, pp. 151–160, 2012. [263] H. F . Atlam, R. J. W alters, and G. B. Wills, “Internet of nano things: Security issues and applications, ” in Cloud and Big Data Computing , A CM, 2018, pp. 71–77. [264] H. Karl and A. Willig, Protocols and ar chitectur es for wireless sensor networks . John W iley & Sons, 2007. [265] H. Tran-Dang, N. Krommenacker , and P . Charpentier , “Localization algorithms based on hop counting for wireless nano-sensor net- works, ” in Indoor P ositioning and Indoor Navigation (IPIN) , IEEE, 2014, pp. 300–306. [266] L. Zhou, G. Han, and L. Liu, “Pulse-based distance accumulation localization algorithm for wireless nanosensor networks, ” IEEE Access , vol. 5, pp. 14 380–14 390, 2017. [267] M. El-Absi, A. A. Abbas, A. Abuelhaija, et al. , “High-accuracy indoor localization based on chipless RFID systems at THz band, ” IEEE Access , vol. 6, pp. 54 355–54 368, 2018. [268] “Ieee recommended practice for nanoscale and molecular commu- nication framework, ” IEEE Std 1906.1-2015 , pp. 1–64, 2016. [269] S. Canovas-Carrasco, A.-J. Garcia-Sanchez, and J. Garcia-Haro, “The IEEE 1906.1 standard: Nanocommunications as a ne w source of data, ” in ITU Kaleidoscope: Challenges for a Data-Driven Society (ITU K), 2017 , IEEE, 2017, pp. 1–7.

Survey on Terahertz Nanocommunication and Networking: A Top-Down Perspective

Original Paper

Comments & Academic Discussion

Leave a Comment

Original Paper

Related Papers

Comments & Academic Discussion

Leave a Comment