Sub 100nW volatile nano-metal-oxide memristor as synaptic-like encoder of neuronal spikes

📝 Abstract

Advanced neural interfaces mediate a bio-electronic link between the nervous system and microelectronic devices, bearing great potential as innovative therapy for various diseases. Spikes from a large number of neurons are recorded leading to creation of big data that require on-line processing under most stringent conditions, such as minimal power dissipation and on-chip space occupancy. Here, we present a new concept where the inherent volatile properties of a nano-scale memristive device are used to detect and compress information on neural spikes as recorded by a multi-electrode array. Simultaneously, and similarly to a biological synapse, information on spike amplitude and frequency is transduced in metastable resistive state transitions of the device, which is inherently capable of self-resetting and of continuous encoding of spiking activity. Furthermore, operating the memristor in a very high resistive state range reduces its average in-operando power dissipation to less than 100 nW, demonstrating the potential to build highly scalable, yet energy-efficient on-node processors for advanced neural interfaces.

💡 Analysis

Advanced neural interfaces mediate a bio-electronic link between the nervous system and microelectronic devices, bearing great potential as innovative therapy for various diseases. Spikes from a large number of neurons are recorded leading to creation of big data that require on-line processing under most stringent conditions, such as minimal power dissipation and on-chip space occupancy. Here, we present a new concept where the inherent volatile properties of a nano-scale memristive device are used to detect and compress information on neural spikes as recorded by a multi-electrode array. Simultaneously, and similarly to a biological synapse, information on spike amplitude and frequency is transduced in metastable resistive state transitions of the device, which is inherently capable of self-resetting and of continuous encoding of spiking activity. Furthermore, operating the memristor in a very high resistive state range reduces its average in-operando power dissipation to less than 100 nW, demonstrating the potential to build highly scalable, yet energy-efficient on-node processors for advanced neural interfaces.

📄 Content

Sub 100 nW volatile nano-metal-oxide memristor as synaptic-like encoder of neuronal spikes Isha Guptaa,*, Alexantrou Serba, Ali Khiata, Ralf Zeitlerb, Stefano Vassanellic, Themistoklis Prodromakisa. aDepartment of Electronics and Computer Science, Faculty of Physical Science and Engineering, University of Southampton, University Road, SO17 1BJ, Southampton, United Kingdom. bMax Planck Institute for Intelligent Systems, Heisenbergstr,3,70569 Stuttgart, Germany. cDepartment of Biomedical Sciences, University of Padova, Via Francesco Marzolo 3, Padova 35131, Italy.

Advanced neural interfaces mediate a bio-electronic link between the nervous system and microelectronic devices, bearing great potential as innovative therapy for various diseases. Spikes from a large number of neurons are recorded leading to creation of big data that require on-line processing under most stringent conditions, such as minimal power dissipation and on-chip space occupancy. Here, we present a new concept where the inherent volatile properties of a nano-scale memristive device are used to detect and compress information on neural spikes as recorded by a multi-electrode array. Simultaneously, and similarly to a biological synapse, information on spike amplitude and frequency is transduced in metastable resistive state transitions of the device, which is inherently capable of self-resetting and of continuous encoding of spiking activity. Furthermore, operating the memristor in a very high resistive state range reduces its average in- operando power dissipation to less than 100 nW, demonstrating the potential to build highly scalable, yet energy-efficient on-node processors for advanced neural interfaces.
Reverse engineering the human brain and decoding the underlying information processes of biological systems requires integrated efforts from researchers with different scientific backgrounds1. Towards enabling this vision, advances in neural recording techniques2,3,4,5,6 target the reliable acquisition of electrophysiological data from multiple neurons in-vitro and in-vivo. This has impacted our understanding of information processing by brain microcircuits7 and brought new prospects for novel therapies based on adaptive neural stimulation8. To date, state-of-art implementations can simultaneously record in-vivo9 from up to thousand sites and from up to 30k10 sites in-vitro using Complementary Metal Oxide Semiconductor (CMOS) based High Density Microelectrode Arrays (HDMEA’s). Such advances in micro-sensors technology have been paralleled by considerable progress in neural processing microsystems11,12 which are capable of detecting neural spiking activity on- node13,14. The relevant spike-detected information is then transmitted off-line wirelessly and techniques such as the Template Matching System (TMS) or Principle Component Analysis (PCA) 15 are used off- line for spike-sorting16. These methods, by mapping the recorded neural activity to the source active neurons, offer insights in neural coding principles17 and support novel neuroprosthetic applications18,19,20,8. Thus, further advances in the fast developing field of implantable neural interfaces21 are hampered by key bottlenecks in the processing of neuronal spikes including: a) computational power required to process the ever increasing volume of neural signals (Gb/s range presently) on-node and in real-time 22,23,24,25, b) bandwidth26 and, c) scalability. Recently, we proposed a new spike-detection approach27 based on metal-oxide resistive switching memory devices, also known as memristors28,29,30. Fundamentally, memristive devices undergo non- volatile resistive state transitions as a function of the integral of the input voltage, thus behaving as thresholded input integrators31. Taking advantage of this property, we demonstrated that TiOx -based memristive devices can be employed for spike-detection27, as extracellular neural spikes recorded from retinal ganglion cells32,33,34 were encoded in gradual, non-volatile resistive state transitions, whereas the sub-threshold events (i.e. noise) were naturally filtered-off27. This property makes these devices suitable as noise-suppressing integrating sensors and are thus termed as ‘Memristive Integrating Sensors (MIS).’ Non-volatility, however, was strongly limiting detection performance, as after saturation of the resistive state the devices, it failed to register any significant neural activity35. Consequently, performance was optimised by manual operation through frequent resets to the initial devices’ resistive state,36 which however impacts negatively on the overall power consumption.
In this work we advance on our previous findings by exploiting an often overlooked crucial property of memristive devices that is ‘volatility’37,38,39,40,41. This approach recalls the way of operation of biological synapses that translate spiking frequency in gradual changes of postsynaptic con

This content is AI-processed based on ArXiv data.