Imagine a computer chip that learns like a human brain, remembering new things without forgetting old ones. Instead of moving data back and forth between a processor and memory, this tech does the math right inside the memory. It uses tiny magnetic switches tuned with ion beams to make AI run on very little power.
Deep neural networks require massive amounts of energy to move data between memory and processors. Additionally, edge AI often suffers from 'catastrophic forgetting' and hardware inconsistency.
A neuromorphic chip demonstrator featuring 25 nm MRAM cells in 15X15 arrays and associated Binary Neural Network algorithms for continual learning.
If you are a wearable device manufacturer dealing with short battery life in AI voice assistants — this project developed low power synapses that reduce energy consumption by performing computation within the memory. This allows for smarter devices that last longer on a single charge.
If you are an autonomous sensor developer dealing with the need for real-time learning at the edge — this project developed a neuromorphic chip that overcomes catastrophic forgetting. This enables vehicles to adapt to new road environments without losing previous training.
If you are a predictive maintenance provider dealing with high hardware variability in sensor arrays — this project developed an ion beam manufacturing solution that reduces device variability. This ensures consistent AI performance across thousands of deployed industrial sensors.
Based on available project data, specific pricing or cost per unit is not provided; however, the project focuses on reducing power consumption for edge AI.
The project developed a manufacturing solution based on ion beam processes to engineer magnetic properties, and has validated 15X15 crosbar arrays.
The project objective explicitly includes establishing an IP strategy as part of its steps toward full commercial readiness.
The project incorporates metaplasticity into Binary Neural Network algorithms and developed local learning algorithms to bridge deep learning and hardware.
The project runs from 2023-05-01 to 2025-10-31, with the goal of achieving full commercial readiness by the end of the period.
The project is led by a single French SME, Spin-Ion Technologies. With a 100% industry ratio and only one partner, the project is highly streamlined for commercialization rather than academic research, focusing directly on the transition from lab to market.
Contact Spin-Ion Technologies in France
Contact us to explore licensing opportunities for ion-beam engineered MRAM synapses.
