As we stand on the shoulders of giants, we’ve often looked to technological innovations to take us away from our biological beginnings. Nowadays, though, biology is often inspiring new artificial designs, and a new study led by engineers at MIT have taken this a step further than many others.
According to a new study published in Nature Materials, they’ve been tinkering around in the world of “neuromorphic computing”, which describes technological systems that replicate, to some degree, the human brain. Rather boldly, an accompanying MIT report on the paper explains that the team have created a new “artificial synapse”.
No, the researchers haven’t developed a fully operation human brain using nothing but mechanical and digital components, and we wouldn't advise holding your breath for such a revelation any time soon. They have, however, taken a significant (if baby) step toward replicating some of its functions on a chip.
Before we dive into what this study has actually accomplished, though, we need to get a few things straight.
First, neuromorphic computing. Conventional computer chips transfer information to and fro at regular intervals. This works fine for the most part, but if we wanted to transfer information whenever we felt like it, rather than at a tick-tock constant pace, we need something else.
Enter, neuromorphic chips. These chips aren’t new, and in fact versions have been around since the 1980s. Instead of using logic gates – which express information using a binary (0 or 1) output, depending on the voltage level – these chips use neuron-inspired blocks.
These allow information to be transmitted in pulses and patterns, independent of any set pace. Information is transmitted on a spectrum, or a gradient, rather than through a binary yes/no system, much like real neurons.
These chips use a lot less energy overall than their conventional equivalents, which makes them more efficient at processing information. Intel have recently made headlines for using these chips to make computing components that “resemble the brain”, but they only resemble one aspect of the brain, to be fair.
So what have the team at MIT done that’s new? Well that all comes down to the aforementioned “artificial synapse.”
Within your brain, and the rest of your central nervous system, you have electrically conductive cells known as neurons. These send information to and from each other via neurotransmitters – biochemical signaling molecules. The junction that allows this connection to happen is called a synapse.
Synapses control the strength of the electrical connection – the flow of ions – between neurons. An artificial synapse would do much the same, and they do exist at present. Right now, though, the material in which they travel across aren’t that good at controlling the current, which leads to inaccurate information transference across neuromorphic chips.
MIT thought they could do one better. They designed an artificial neuron made of single-crystalline silicon, whose atoms are arranged in a particularly ordered way. They found that this allowed for a much more precise flow of ions, and in simulations using handwriting samples, the correct information was transferred accurately 95.1 percent of the time.
Making these synapses wasn’t easy. At just 25 billionths of a meter across, they’re smaller than the Ebola virus is wide.
As you may have noticed, neuromorphic chips aren’t widespread in computing right now. Despite existing for almost 40 years, their true potential hasn’t been realized, and until they can be scaled up, they’ve yet to replace conventional logic gate chips for the most part.
Research like this, however, may change that. Combining the tiny dimensions of these artificial synapses with the efficiency of neuromorphic chips may allow us to design portable neural networks: advanced artificial intelligences we can fit in our pockets.