TDK and CEA Collaborate on AI Spin Memristor for Power Efficiency

TDK, in collaboration with CEA in France, has made a significant breakthrough in the field of neuromorphic AI devices by developing a spin memristor. This innovative technology, developed in partnership with Tohoku University in Japan, boasts a remarkable power consumption rate of just 1% compared to previous devices.

The spin memristor leverages the charge and spin of electronics within an MRAM element, enabling the creation of prototype learning neuromorphic AI devices in conjunction with the Centre for Innovative Integrated Electronic Systems (CIES) at Tohoku University.

In the era of digital transformation (DX), the rapid advancement of AI has led to a surge in energy consumption associated with big data and AI applications. This has brought to light various challenges, including the complexity of processing vast amounts of data and the escalating power consumption linked to AI development. TDK is committed to addressing these social and environmental issues through its cutting-edge innovations.

TDK's ambitious goal was to develop a device that mimics the synapses of the human brain, leading to the creation of the memristor. Unlike conventional memory elements that store data in binary form, the spin memristor can store data in analog form, mirroring the brain's functionality. This breakthrough enables complex computations to be carried out with minimal power consumption.

While existing memristors for neuromorphic devices encounter challenges such as resistance fluctuations over time and difficulties in data writing precision, TDK's spin memristor overcomes these obstacles. It offers resilience to environmental factors, long-term data retention, and reduced power consumption by minimizing leakage current in current devices.

The prototype AI circuit, featuring 3 memristor elements, 2 sets, and 4 chips, successfully demonstrated its operation through a sound separation test. This showcased the spin memristors' potential as fundamental components in AI circuits, capable of real-time learning in dynamic environments.

The manufacturing process of these revolutionary products involves the integration of semiconductor and spintronic manufacturing techniques. TDK has collaborated with Tohoku University to achieve this integration, drawing on their expertise in MRAM production, a technology akin to memristors.

“The collaboration between TDK and CEA has been exceptional, combining our unique strengths to drive innovative solutions for modern AI applications,” remarked Marc Duranton, Senior Fellow of CEA. Tetsuo Endoh, Director of CIES at Tohoku University, emphasized the importance of TDK's AI semiconductor development program in meeting the evolving demands of society.

For more information, visit www.tdk.com and www.cea.fr.