Researchers in France are developing a hybrid AI chip in a €3m project inspired by insects.
The project is led by Elisa Vianello, senior scientist and Edge AI programme coordinator at CEA-Leti, and aims to combine different AI techniques and memories for consumer robotics, implantable medical diagnostic microchips and wearable electronics.
Moving data between AI processors and memory is a key area of design that has been tackled with in-memory computing and highly integrated AI processors such as GraphCore and Cerebras.
However, these approaches require very high-density, high-resolution, non-volatile memory with unlimited endurance. Instead, the CEA-Leti project will combine a Bayesian Inference engine and a local spiking neural network module to process data in real time. This will use deterministic, probabilistic, volatile and non-volatile memories that Vianello is currently developing with fellow CEA-Leti scientists.
“My project is to take inspiration from insects’ nervous systems to relax hardware requirements in terms of memory density and reliability, and to build the new nanosystems we need to enable learning from a very limited volume of noisy data,” she said.
“Crickets make accurate decisions based on sluggish, imprecise, and unreliable neurons and synapses in order to escape their predators. Looking closely at their biology, we identified a diversity of memory-like functions at play in their sensory and nervous systems,” she said. “By combining these different functions, the cricket’s internal computing system achieves amazing performance and energy efficiency.”
The team aims to make networks of physical nanoscale memory devices that translate insect biological principles into physical principles to enable learning from very limited volumes of noisy data, such as data measured in real time from different sensors in video cameras, radar sensors, ECG, EMG, bio-impedance streams and potentially also brain signals through EEG sensors and neuro-probes.
“Since the ideal memory does not exist today, the project aims at building a hybrid synapse that co-integrates different memory technologies,” said Vianello.
“Elisa’s work will open up new research perspectives towards more energy-efficient embedded intelligence capable of online learning,” said Jean-René Lequepeys, deputy director and CTO at CEA-Leti. “It is a real technological and application breakthrough that will combine the latest developments in microelectronics using new generations of non-volatile memories and drawing inspiration from the living world. This research work is fully in line with the priorities of the institute and will open up great opportunities for world premieres and commercialization.”
Other projects have taken insect vision as the inspiration for spiking neural network chips.
Other articles on eeNews Europe
- Quantum project for digital twins
- Open source chip design language is based around Python
- Intel confirms German fabs in €80bn European investment
- Attack strategy fools lidar sensors on driverless cars
- SiFive raises $175m for RISC-V design as a double unicorn
- Report shows most successful UK universities for spinouts