Developing brain-inspired AI

Sherbrooke, le 08 février 2024 – The Université de Sherbrooke’s new MEIE Chair in Neuromorphic Computing, which is funded by the Ministère de l’Économie, de l’Innovation et de l’Énergie (MEIE), will let two professor researchers from UdeS use brain function as inspiration to forge the future of artificial intelligence (AI). One goal of AI research is to have these systems process information in the same way a human brain would. This chair’s work will take place and benefit from the unique environment of the Interdisciplinary Institute for Technological Innovation (3IT), which is the nerve centre for neuromorphic research in Quebec that covers the full innovation chain.

The research of this chair will lead to different types of wearable devices such as smartwatches, sleep or heart rate trackers, mobile electroencephalography (EEG) devices, and hearing aids that could have considerably improved efficiency without the need to access centralized computing resources. The two chairholders are Sean Wood, a professor in the Department of Electrical and Computer Engineering, and Fabien Alibart, an associate professor in the Faculty of Engineering and a CNRS researcher who is a member of the Laboratoire international Nanotechnologies et Nanosystèmes (IRL-LN2) based in Sherbrooke.

The chair’s goal is to use what is called “edge AI” to develop innovative wearable devices that also use less energy. These two researchers have adopted a co-design approach to develop these neuromorphic systems, and their complementary expertise will fuel each other to drive the chair’s research into both hardware and software. Another goal is to optimize the pairing between these bio-inspired algorithms and their on-chip implementation.

Bio-inspired neural systems

Fabien Alibart’s main task will be developing the neuromorphic hardware components. “This work is at the interface between electronics and neuroscience. The hardware we develop for these new AI systems will be bio-inspired by actual neural communication systems, which is really exciting.” It is not easy to reproduce a neural network on conventional chips, as the language used in each case is completely different. The main challenge for these two researchers is therefore designing chips that mimic the actual biology of neural network architecture as closely as possible.

Innovating to simulate biological impulses

For his part, Sean Wood will oversee the algorithm development for these projects. “One challenge will be to create a computerized reproduction of how biological neural networks send electrical impulses through neurons and synapses. We believe that achieving this type of ‘electrical impulse’ will make a real difference in our work. Since these events are temporally isolated and parsimonious (i.e. they use no more parameters than necessary), we expect to see a significant reduction in energy consumption.”

“We want to overcome current limitations,” Sean Wood added. “Fabien’s research will help me adapt the algorithms to the hardware and vice versa. The software and hardware development will inform each other along the way.” In the first phase of the chair’s work, the neuromorphic algorithms will be influenced by hardware constraints, while the algorithms developed during this first phase will be used for the design of neuromorphic hardware.

“Responsible artificial intelligence is a promising field for Québec. This Chair in Neuromorphic Computing is a new resource that will complement the world-class quantum science ecosystem we are building with the DistriQ innovation zone,” says Pierre Fitzgibbon, Minister of Economy, Innovation and Energy, Minister Responsible for Regional Economic Development and Minister Responsible for the Metropolis and the Montréal Region.

Increasing the autonomy and reliability of electronic medical devices

Traditional artificial neural networks mainly use memory that is external to the processors. This aspect will also change with the use of local memory. Direct on-chip processing will have two enormous benefits: a significant decrease in power consumption and a reduction in latency, or the computation time between signal input and output. “It is a challenge to create electronic devices with autonomy, which is often limited due to the vast amount of information exchanged between the sensor networks and remote data processing centres,” said Fabien Alibart. “Pairing AI software with embedded sensors will lead to much more reliable functionalities that use less power. This is particularly important for devices that require extensive autonomy, such as heart rate monitors, sleep monitors, fall detectors, and devices that screen for critical medical conditions.”

3IT: Quebec’s centre for edge AI 

The neuromorphic processors developed at 3IT will contain the first neuromorphic embedded-memory chips. “This chair was developed thanks to UdeS’s expertise in creating collaborative continuums between the university and the private sector through its established international collaboration with the LN2 laboratory. The developments planned for the chair’s work will also be a springboard to position Quebec as a leader in neuromorphic-based AI. This initiative will let us train the next generation of specialists in this field who will benefit from an exemplary international training environment as they create lasting partnerships with a range of industrial players across the entire value chain of this sector of AI,” said Vincent Aimez, Vice-President, Partnerships and Knowledge Transfer.

– 30 –

Information:

Isabelle Huard, Media Relations Advisor
Communications Department | Université de Sherbrooke
819-821-8000, extension 63395 | medias@USherbrooke.ca