Building neuromorphic chips with traditional VLSI technology still lags nature’s efforts on many fronts – not least power consumption. Now, Oak Ridge National Laboratory researchers working with soft materials and biomolecules have demonstrated ‘synaptic mimics’ which, among other things, work at much lower consumption, could potentially interface directly with the brain, and have shown some learning capabilities.
Working at Center for Nanophase Materials Sciences, the researchers were able to fabricate ultra-thin membranes and embed them with biomolecules that transport ions in response to voltage changes. The synaptic mimics process and store information in the same location, unlike conventional computers in which information is continually transferred between a central processing unit and memory, which consumes a lot of energy.
“Through experiments and modelling, we were the first to demonstrate that our synaptic mimics, when configured in neuromorphic circuits, emulate key functions of real biological synapses, such as computing, learning and remembering,” said ORNL’s Patrick Collier, one of the researchers involved in the project. Their paper (Memristive Ion Channel-Doped Biomembranes as Synaptic Mimics) was published in the journal ACS Nano. See figure one from the paper below.
The authors report having presented a “fundamentally different class” of iono-neuromorphic, soft, two-terminal biomolecular memristor that mimics the physical structure, switching mechanism, and ion transport of biosynapses. Here’s an excerpt from the paper’s conclusion
“Compared to solid-state memristors, neuromorphic architectures based on stimuli-responsive biomolecules offer distinct advantages. Our self-assembled biomolecular memristors consume significantly less power (∼0.1−10 nW), and they are relatively inexpensive and easily scalable via droplet-based printingor microfluidic methods. Further, circuity based on stimuli-responsive ion channels offers the possibility for adaptive learning and memory, as well as sensing, of many types of physical and chemical stimulations, possibly even at the same time and in the same membrane.
“Thus, the results presented here forecast an alternative paradigm for neuromorphic hardware using materials that could be integrated into synthetic multifunctional structures and interfaced with bio- logical tissues to provide adaptive sensing, signal processing, smart edge computing, and memory.”
Link to paper: https://pubs.acs.org/doi/10.1021/acsnano.8b01282
Link to ORNL summary: https://www.ornl.gov/news/computing-mimicking-neurons
Feature Image Caption & Source: Computing building blocks of soft materials may someday directly interface with the brain, according to researchers at Oak Ridge National Laboratory and the University of Tennessee. Credit: Joseph Najem, Oak Ridge National Laboratory/U.S. Dept. of Energy