UTRECHT, Netherlands — An international team of physicists has made a splash in the world of brain-like computers. Theoretical physicists at Utrecht University in the Netherlands and experimental physicists at Sogang University in South Korea built an artificial synapse based on a mix of water and salt. This new type of computing device uses fluidic ion channels, just like the neurons in our brains. This “neuromorphic” approach promises to be much more energy efficient than traditional computers.
“While artificial synapses capable of processing complex information already exist based on solid materials, we now show for the first time that this feat can also be accomplished using water and salt,” says lead study author Tim Kamsma, a doctoral candidate at the Institute for Theoretical Physics and the Mathematical Institute of Utrecht University, in a media release. “We are effectively replicating neuronal behavior using a system that employs the same medium as the brain.”
According to the study published in the journal Proceedings of the National Academy of Sciences, the key to this breakthrough is the unique design of the device’s microchannels. Imagine a tiny tapered channel, thinner than a human hair, filled with a rigid crystal structure of charged silicon dioxide nanospheres. The spaces between these spheres create a network of even smaller nanochannels that ions (electrically charged particles) can flow through, similar to how electrical signals travel between neurons in the brain.
When a voltage is applied across the channel, the difference in ion concentration from one end to the other allows the device to preferentially conduct current in one direction — a property called ion current rectification. But here’s the really cool part: the degree of rectification can be altered by changing the voltage, allowing the device to function as a “memristor,” essentially a resistor with memory.
Researchers developed a theoretical model to understand exactly how the ion transport works in these channels. They discovered that the memristor’s “memory” timescale is surprisingly dependent on ion diffusion time, even though the process is driven by voltage. This means that by simply changing the length of the channel during fabrication (which is easy to do), the memory retention time can be precisely tuned.
To put their device to the test, the team used it to tackle a classic neuromorphic computing problem: handwritten digit recognition. They encoded pixel images of numbers as voltage pulses and fed them into the fluidic memristor. Researchers found it remarkable that the device was able to transform the temporal voltage signals into unique output current signatures for each digit. Feeding these outputs into a simple neural network led to the successful classification of the numbers on par with other state-of-the-art neuromorphic platforms.
“This suggests the possibility of tailoring channels to retain and process information for varying durations, again akin to the synaptic mechanisms observed in our brains,” explains Kamsma.
What makes this iontronic approach so exciting is how closely it mimics the brain’s own computing processes. Our neurons communicate via ions flowing through water-filled channels, with connection strengths that can vary over time (known as synaptic plasticity). The fluidic memristor captures this dynamic behavior, something traditional electronic memristors have struggled to fully replicate.
Looking ahead, researchers believe this new device could form the building block for advanced neuromorphic chips. Multiple channels could potentially be linked together, just like neurons in the brain, to enable even more sophisticated computing. And thanks to the nanoscale fabrication, such chips would be extremely energy efficient compared to today’s power-hungry processors.
“It represents a crucial advancement toward computers not only capable of mimicking the communication patterns of the human brain but also utilizing the same medium,” concludes Kamsma. “Perhaps this will ultimately pave the way for computing systems that replicate the extraordinary capabilities of the human brain more faithfully.”
There are still a few kinks to be worked out in the memristor’s performance and the underlying theory. But this study undoubtedly represents an exciting leap forward in the field of neuromorphic engineering. By embracing the brain’s “wetware” approach — ions, water, and ever-changing connections — we may be on the cusp of a computing revolution. The future of artificial intelligence could be a lot more fluid than we ever imagined.
StudyFinds’ Matt Higgins contributed to this report.