Researchers have been trying to build artificial synapses for years in hopes of getting close to the unrivaled computational performance of the human brain. A new approach has now succeeded in designing approaches 1,000 times smaller and 10,000 times faster than their biological counterparts.
Despite the tremendous success of deep learning over the past decade, this brain-inspired approach to AI is challenged by the fact that it works on devices that only look like real brains. That’s a big part of the reason why a human brain weighing just three pounds can pick up new tasks in seconds using the same amount of power as a light bulb, while training the largest neural networks takes weeks, megawatt hours of electricity, and racks of specialized processors.
This has led to increased interest in efforts to redesign the core machines that AI runs on. The idea is that by building computer chips whose components such as neurons and natural synapses function, we may be able to approach the maximum space and energy efficiency of the human brain. The hope is that these so-called “neural” processors could be better suited to powering AI than current computer chips.
Researchers from the Massachusetts Institute of Technology have shown that an unusual synthetic synapse design that mimics the brain’s reliance on ions moving around it can significantly outperform biological elements. A major breakthrough was finding a material that withstands intense electric fields, dramatically improving the speed at which ions can move.
“The speed was certainly surprising,” Murat Onin, who led the search, He said in a press release. “Normally, we would not apply such extreme fields across devices, so as not to turn them into ash. But instead, protons [which are equivalent to hydrogen ions] It ended up cruising at tremendous speeds through the hardware stack, specifically a million times faster than what we had before. “
while there a A variety of neural engineering approaches, one of the most promising is analog computing. This seeks to design components that can exploit internal physics to process information, which is more efficient and straightforward than executing complex logical operations like traditional chips.
So far, much research has focused on designing “memristors” – electronic components that control the flow of current based on how much charge was previously flowing.so through the device. This mimics the way connections between biological neurons increase or decrease in strength depending on the frequency with which they communicate, meaning that these devices can in principle be used to create networks with similar properties to biological neural networks.
Perhaps not surprisingly, these devices are created using memory technologies. but in a new way the paper is in it SciencesAnd the The MIT researchers argue that components optimized for long-term information storage are in fact unsuitable for performing the normal state transitions required to constantly adjust connection strength in an artificial neural network. That’s because the physical properties that ensure long retention times are not complementary to those that allow for high-speed switching.
That’s why the researchers instead designed a component whose conduction is regulated by inserting or removing protons into a channel made of phosphosilicate glass (PSG). To some extent, this mimics the behavior of biological synapses, which use ions to transmit signals across the gap between two neurons.
However, this is where the similarities areflies End. The device has two terminal terminals which are basically clamp inputs and outputs. A third party is used to apply an electric field, which induces the protons to move from a tank to a PSG channel or vice versa depending on the direction of the electric field. More protons in the channel increase its resistance.
Researchers came with this Universal design returns in 2020However, their previous devices used materials incompatible with chip design processes. But most importantly, switching to PSG significantly increased device switching speed. That’s because the nano-sized pores in its structure enable the protons to move very quickly through the material, and also because they can withstand very strong electric field pulses without being damaged.
More powerful electric fields give the protons a massive increase in speed and are key to the device’s ability to outperform biological synapses. In the brain, electric fields must remain relatively weak because anything over 1.23 volts (V) causes water to forms The bulk of cells to be divided into hydrogen and oxygen gas. This is largely why neuronal processes occur on the millisecond scale.
In contrast, the MIT team’s device is capable of operating at up to 10 volts in pulses as short as 5 nanoseconds. This allows the synthetic synapse to work 10,000 times faster than its biological counterparts. Furthermore, the devices are only nanometers wide, making them 1,000 times smaller than biological synapses.
experts Tell new world That the three-pin setup of the device, in contrast to the two types found in most neuron models, may make it difficult to operate certain types of neural networks. The fact that protons must be introduced using hydrogen gas also presents challenges when scaling up the technology.
There is a long way to go from the single synthetic synapse to the large networks capable of serious information processing. But the exceptional speed and small size of the components indicate that this is a promising direction in the search for new devices that can match or exceed the power of the human brain.
Image Credit: Ella Maru Studio / Murat Onin