BLAST Synapses Show Promise for Neural Networks and Brain-Machine Interface Efforts
Operating well below the energy required of biological synapses, never mind electronic components, BLASTs could power future prosthetics.
Researchers from the University of Texas at Austin and Sandia National Laboratories have designed bio-compatible components for brain-like computers, manufactured from graphene and a membrane material called nafion.
"Computers that think like brains can do so much more than today's devices," claims Jean Anne Incorvia, assistant professor in the Cockrell School of Engineering and lead author of the paper detailing the artificial synapse component. "And by mimicking synapses, we can teach these devices to learn on the fly, without requiring huge training methods that take up so much power."
The synapse developed by the team, which aims to drive future computers capable of operating more like the human brain, is dubbed BLAST: Biocompatible Bilayer Graphene-based Artificial Synaptic Transistor. The device switches with an energy efficiency an order of magnitude lower than its predecessors, and even below the 1fJ estimated switching energy of biological synapses — making it ideal for high-efficiency systems.
A key feature of the device is its metaplasticity, which the team claims makes it perfect for generalizable deep neural network (DNN) workloads — and in testing, the prototype BLASTs proved capable of outperforming ideal linear synapses in image classification tasks.
The material from which the synaptic transistors are made has another feature: Biocompatibility, making it suitable for implantation. "Biocompatibility, flexibility, and softness of our artificial synapses is essential," says post-doc researcher and project co-lead Dmitry Kireev. "In the future, we envision their direct integration with the human brain, paving the way for futuristic brain prosthesis."
The team's work has been published in the journal Nature Communications under open-access terms.