ENGINEERING OF NEUROMORPHIC SYSTEMS
The research group « Engineering of neuromorphic systems » aims at designing integrated circuits (IC) and instrumentation systems which components and architecture are neuro-mimetic (i.e. mimic biological neural systems). This research activity appeared in IXL in 1993, and since then has been inter-disciplinary oriented. National and international collaborations exist in the group with neuroscientists as well as computer scientists and physicists.
In its current organization, two research directions are mainly identified in the group:
1) |
Silicon
neurons: design of analog and mixed neuromimetic IC devices; |
2) |
Systems
for artificial and hybrid neural networks: instrumentation platforms
based on software or hardware artificial neurons and dedicated to
computational or experimental neuroscience. |
Silicon neurons
Design of custom analog and mixed ASICs:
- |
Conductance-based
models of neurons and synapses |
- |
Mathematical
functions emulated by analog circuitry (see Gallery) |
- |
Neurons
and synapses models parameters are eventually tunable (on-chip
or off-chip storage) |
- |
Spikes
are analog or event-coded digital signals and time-stamped for
neural network communication (see Sensemaker) |
- |
Off-chip
synaptic plasticity computation for small networks (see Facets) |
- |
IC
functions library optimized for design re-use (see Analog
design re-use) |
=>
Real-time and
continuous simulation
of adaptive spiking neural networks modelled at the conductance level.
Systems for artificial and hybrid neural networks
Engineering of complete platforms computing software and hardware neural networks:
| - | HW/SF
systems for real-time simulation of neural networks with programmable
adaptive properties such as STDP (see SenseMaker,
Facets) |
| - | Closed-loop
hybrid systems, that bi-directionally interconnect in real-time
artificial and living neurons through intra-cellular or extra-cellular
connections (Hybrid Systems, Neurobit,
Neuro-vers-IT) |
=>
Instrumentation tools for computational and experimental neuroscience
using biomimetic and modular artificial neural networks.