Periodic Reporting for period 4 - IONOS (An iono-electronic neuromorphic interface for communication with living systems)
Berichtszeitraum: 2023-05-01 bis 2024-10-31
While our understanding of the brain have made huge progresses, we are still inefficient in interfacing biological systems with electronics, both in terms of energy and integration potential. Pushed by the need to use conventional computers for building complex systems dedicated to brain interface applications, we have mostly capitalized on technologies and architectures inherits from microelectronic that are intrinsically not adapted to interface living systems. The IONOS project has shifted the brain interface paradigm by developing new technologies designed to interact intimately with biological cells and capitalizing heavily on bio-inspiration. To reach this goal, the IONOS project has explored how to sense, and compute biological signals from in-vitro neural cells’ assembly based on iono-electronic materials and devices. These emerging devices offer basics functionalities such as memory, ion-electron signal’s transduction, and amplification paving the way to a new field of device and circuit engineering that will reproduce key biological functions such as learning and spatio-temporal processing of information. This project demonstrated how these concepts associated to the bio-inspired computing paradigm could unlock our fundamental limitations for communicating with living neural cells. In the near future, the IONOS findings will very likely enable the proof of concept of how an artificial system can efficiently send, receive and compute information from a biological one, which constitutes the basic of communication.
In its first phase, the IONOS project progressed in parallel in the two main field of activity of the project.
(1) We have developped neuromorphic sensors for neural cells' activity recording based on iono-electronic materials. Passive and active sensors have been fabricated and charaterized for the recording of brain slices activity. Notably, we shew how organic materials can be used to create "plastic" sensors that could adapt to biological medium. We also explored how spatio-temporal sensing with dendritic like sensors can be used to process neural cells activity.
(2) We have put strong efforts for increasing the maturity level of neuromorphic hardware. In particular, we have developped a fully back end of line compatible process for memristive devices integration with CMOS. This work will allow us to access to neuromorphic hardware for processing bio-signals. Such neuromorphic circuits are expected to reduce drastically energy consumption and to provide a large parallelism for time dependent signal processing.
In the second phase, the IONOS project concentrated on merging sensing and computing toward the realization of innovative concepts for brain-machine ingterfaces.
(1) Dendritic PEDOT fibers where considered for implementing structural plasticity, which represent a new computing ingredient for neuromorphic engineering. Since dendritic PEDOT fibers can be employed as sensors, this proposition is opening new perspective to implement evolvable sensing platform, and more broadly to implement evolvable hardware and wetware
(2) Neuromorphic algorithms have been adapted to the particular context of spike sorting, which represents a very popular way of analyzing signals from microelectrode implants. Our future objective will be to realize the demonstration of real-time computing of bio-signals with ultra-low power neuromorphic platform.
(1) We have developped neuromorphic sensors for neural cells' activity recording based on iono-electronic materials. Passive and active sensors have been fabricated and charaterized for the recording of brain slices activity. Notably, we shew how organic materials can be used to create "plastic" sensors that could adapt to biological medium. We also explored how spatio-temporal sensing with dendritic like sensors can be used to process neural cells activity.
(2) We have put strong efforts for increasing the maturity level of neuromorphic hardware. In particular, we have developped a fully back end of line compatible process for memristive devices integration with CMOS. This work will allow us to access to neuromorphic hardware for processing bio-signals. Such neuromorphic circuits are expected to reduce drastically energy consumption and to provide a large parallelism for time dependent signal processing.
In the second phase, the IONOS project concentrated on merging sensing and computing toward the realization of innovative concepts for brain-machine ingterfaces.
(1) Dendritic PEDOT fibers where considered for implementing structural plasticity, which represent a new computing ingredient for neuromorphic engineering. Since dendritic PEDOT fibers can be employed as sensors, this proposition is opening new perspective to implement evolvable sensing platform, and more broadly to implement evolvable hardware and wetware
(2) Neuromorphic algorithms have been adapted to the particular context of spike sorting, which represents a very popular way of analyzing signals from microelectrode implants. Our future objective will be to realize the demonstration of real-time computing of bio-signals with ultra-low power neuromorphic platform.
1- Dendritic PEDOT fibers for structural plasticity implementation: this work open a new avenue for neuromorphic engineering and could also deeply influence how we implement computing by enabling bottom-up engineering approaches [25].
2- Demonstration of spiking neurons and memristors for unsupervised learning [15, in prep]. This realization will enable new cirucits design and applications in the broad context of AI deployment
3- Expanding memory in recurrent spiking networks [26] is paving the way for new approaches in neuromorphic computing that will rely on delays for implementing a distributed memory at the network level. This approach is conceptuallyt a clear departure with the conventional implementation of memory in hardware (i.e. via synaptic connections)
2- Demonstration of spiking neurons and memristors for unsupervised learning [15, in prep]. This realization will enable new cirucits design and applications in the broad context of AI deployment
3- Expanding memory in recurrent spiking networks [26] is paving the way for new approaches in neuromorphic computing that will rely on delays for implementing a distributed memory at the network level. This approach is conceptuallyt a clear departure with the conventional implementation of memory in hardware (i.e. via synaptic connections)