Imagine a computer brain that doesn't forget how to do old tasks when it learns a new one. It works like a human brain by marking some memories as 'extra important' so they aren't accidentally overwritten. This is achieved using tiny magnetic switches that can be tuned to be harder or easier to change.
Current AI suffers from 'catastrophic forgetting,' where learning a new task erases previous knowledge. This forces expensive retraining and limits the ability of AI to operate autonomously in changing environments.
A low-power artificial synapse based on spintronics and magneto-ionics, and a corresponding ANN demonstrator for multitask learning.
If you are a medical software provider dealing with AI that forgets old disease patterns when trained on new ones — this project developed a neuromorphic hardware demonstrator that uses metaplasticity to maintain a hierarchy of learned information.
If you are an IoT manufacturer dealing with high power consumption in on-device learning — this project developed low-power spintronics nanodevices that reduce the energy needed for artificial synapses.
If you are a robotics developer dealing with catastrophic forgetting in multi-task environments — this project developed a new class of hardware using magneto-ionics to prevent the loss of previously learned skills.
Based on available project data, specific unit costs are not provided, though the project received an EU contribution of EUR 2,999,750 for development.
The project involves 3 industrial partners and uses standard processes like sputtering and atomic layer deposition (ALD), suggesting a path toward scalability, though it is currently at the demonstrator stage.
Based on available project data, specific licensing terms are not mentioned; however, the consortium includes 9 partners across 5 countries.
The project is developing specific ANN learning schemes adapted to the device physics to ensure the hardware and software work together to mitigate catastrophic forgetting.
The project period runs from 2023-02-01 to 2027-07-31.
The consortium is well-balanced for technology transfer, featuring a 33% industry ratio with 3 industrial partners, including 2 SMEs. With 9 partners across 5 countries (FR, CH, CZ, DE, IT), the project combines academic research from 3 universities and 3 research institutes with industrial application capabilities.
Contact Universite Paris-Saclay
Contact us to explore licensing opportunities for metaplastic spintronic synapses.
