Roadmap on Atto-Joule per Bit Modulators

Volker J. Sorger1,*, Rubab Amin1, Jacob B. Khurgin2 Zhizhen Ma1, Hamed Dalir3, Sikandar Khan1

1Department of Electrical and Computer Engineering, George Washington University 800 22nd St., Science & Engineering Hall, Washington, DC 20052, USA 2Department of Electrical and Computer Engineering, Johns Hopkins University, Baltimore, Maryland 21218, USA 3Omega Optics, Inc., 8500 Shoal Creek Blvd., Bldg. 4, Suite 200, Austin, TX 78757, USA

*corresponding author, E-mail: sorger@gwu.edu

Abstract Electro-optic modulation performs the conversion between the electrical and optical domain with applications in data communication for optical interconnects, but also for novel optical compute algorithms such as providing non-linearity at the output stage of optical perceptrons in neuromorphic analogue optical computing. While resembling an optical transistor, the weak light-matter-interaction makes modulators 105 times larger compared to their electronic counterparts. Since the clock frequency for photonics on-chip has a power-overhead sweet-slot around 10’s of GHz, ultrafast modulation may only be required in long-distance communication, but not for short on-chip links. Hence the search is open for power-efficient on-chip modulators beyond the solutions offered by foundries to date. Here we show a roadmap towards atto-Joule per bit efficient modulators on-chip as well as some experimental demonstrations of novel plasmon modulators with sub-1fJ/bit efficiencies. Our parametric study of placing different actively modulated materials into plasmonic vs. photonic optical modes shows that 2D materials overcompensate their miniscule modal overlap by their unity-high index change. Furthermore, we reveal that the metal used in plasmonic-based modulators not only serves as an electrical contact, but also enables low electrical series resistances leading to near-ideal capacitors. We then discuss the first experimental demonstration of a photon-plasmon-hybrid Graphene-based electro- absorption modulator on silicon. The device shows a sub-1V steep switching enabled by near-ideal electrostatics delivering a high 0.05dB/V-µm performance requiring only 110 aJ/bit. Improving on this design, we discuss a plasmonic slot-based Graphene modulator design, where the polarization of the plasmonic mode matches with Graphene’s in-plane dimension. Here a push-pull dual-gating scheme enables 2dB/V-µm efficient modulation allowing the device to be just 770 nm short for 3dB small signal modulation. Lastly, comparing the switching energy of transistors to modulators shows that modulators based on emerging material-based, plasmonic-Silicon hybrid integration perform on-par relative to their electronic counter parts. This in turn allows for a device-enabled two orders-of- magnitude improvement of electrical-optical co-integrated network-on-chips over electronic-only architectures. The latter opens technological opportunities in cognitive computing, dynamic data-driven applications system, and optical analogue compute engines to include neuromorphic photonic computing.

Keywords: electrooptic modulation, plasmonics, integrated optics, Graphene, energy efficiency, DDDAS, optical computing.
Main Body Electro-optic modulation is a key function in modern data communication as it performs the conversion between the electronic data originating from compute cores, to the optical domain of low-loss data routing. While this function is universally used around the globe in long-haul, metro, and short-haul communications [1], such as data centers [2,3], the case for on-chip optical interconnects was made [4] mainly to address the widening discrepancy between the data handling capability of electronic cores vs. delays and power overheads in the communication- handling network-on-chip [5-8].

The parallels between an electro-optic modulator (EOM) and a field-effect-transistor (FET) however are noticeable; both control a channel via an electrostatic gate. The discrepancy of the physical device lengths are, however, significant, as state-of-the-art silicon photonics modulators are of millimeter dimensions [9], while FETs are just 10 nanometer. The reason for this is known, and lies in the weak light-matter-interaction, and inefficient material ability to change its optical refractive index upon applying a gate bias [11-14]. Given the resemblance of an EOM to FETs, we interchangeably use the words ‘switching ‘ and ‘modulation’, while accepting the slight discrepancy – switching technically refers to a strict 2-level system, while modulation is an analogue function. Of course, in reality both EOMs and FETs have analogue transfer functions, hence justifying the wording definition in this work.

This manuscript focuses on charge-driven electro-absorption modulators only, as suppose to electric field-driven designs, such as those based on Franz-Keldysh, or

…(Full text truncated)…