FYI.

This story is over 5 years old.

Tech

How a Single Circuit Component Mimics Neurons in the Brain

Engineers score a major advance in neuromorphic computing.

Synapses, the chemical or electrical linkages in the brain connecting neurons to other neurons, have a very useful property. They are able to change over time in accordance with patterns of activity, a property known as synaptic plasticity. This is memory, generally, and how it all works is still pretty murky—which makes replicating it technologically, as in neuromorphic computing, murky as well.

As it turns out, we may have a more or less readymade circuit component that can accomplish much the same thing, according to researchers at the University of Massachusetts at Amherst. This is the long-theorized memristor, something that became a reality several years ago at HP Labs. As one might surmise, a memristor is a resistor that comes equipped with its memory, or is its own memory. Like the biological synapse, it changes (resistivity) with varying patterns of use.

Advertisement

In chemical synapses, the transfer of an impulse from one neuron to another occurs via neurotransmitters. The process is initiated as the electrical potential on one neuron increases and allows calcium ions to flow freely into the gap between neurons. The calcium in turn triggers proteins that open up tiny chambers featuring neurotransmitter payloads, which saturate the neural gap. The free-floating neurotransmitter then binds to molecules on the receiver neuron. Message sent.

Image: Yang et al

It's the diffusion of calcium that has the Amherst researchers interested. Certain metallic atoms, like silver and copper, diffuse through certain dielectric materials in a very similar way. They attempted to mirror the calcium diffusion process in synapses by sandwiching a thin dielectric film embedded with silver atoms between two electrodes. They found that by adding a bit of electric stimulation, the atoms diffuse through the electrode gap, like calcium ions between two neurons. The result of this diffusion is increased conductivity through the memristor, just as the calcium acts to pave the way for a signal to fire across a chemical synapse in the brain.

The upshot is that the memristor is able to maintain a particular state even in the absence of electricity. It needs to the electrical bump to get into that state, but not to persist it. So, we can imagine circuits wired up with memristors that "remember" their own activity and use that memory to influence future computations. This is different than just referencing a memory bank in the usual way; it's as if the processor itself is mutating into something different based on its history. The brain is up to something similar, which is why memristor-based circuits have potential in neuromorphic computing—computers that can truly mimic brains, that is.

"Devices with slower switching speeds might be used as neuromorphic emulators," Joshua Yang, an electrical and computer engineering professor at Amherst, told Nanotechweb. "These could allow us to make more bio-realistic artificial synapses with higher fidelity that could help neurobiologists better understand—and replicate—how real synapses work."

The Amherst group's work was published last week in the journal Nature Materials.