Researchers at MIT may have finally found the key to creating a computer that works like the human brain – or at least as close to it as is currently possible.
In a breakthrough piece of research in the field of neuromorphic computing, a team at MIT designed a chip with artificial synapses.
In a human brain, synapses work as bridges for nerve cells to communicate by firing electrical or chemical signals across to a cell receptor. It’s this process that allows our brains to work and manage millions of tasks without us even needing to think about them. It’s also how we make conscious movements and decisions and is believed to be how we form memories.
In the MIT team’s chip, the creation of artificial synapses allows for the precise control of electrical currents flowing along them. This process mimics how synapses regulate neurons in the human brain and is key to unlocking the full potential of building artificial brains.
Currently, computers can’t contend with the computing power of the human brain – even if it seems AI is smarter than us all. In the field of machine learning, trying to simulate all the processes our brains perform subconsciously is a gargantuan task. Ask an AI to crunch some numbers while rendering a film, and it could do it, but ask it to perform a million tiny computations continually while doing that, and it’ll slow to a crawl.
This is because computer chips work in a binary state, sending information as pulses of power instead. The human brain similarly uses electrical pulses to work but, instead of binary signals, it uses synapses to regulate signals and activate set neurons to carry out certain tasks.
Building this system into a computer has proved incredibly complex. There have been numerous attempts, but their architecture has been too complex to create efficient chips. One such method, which utilised two conductive layers separated by an amorphous “switching medium” that works like the brain’s synapses, has now been modified by MIT researchers.
A single chip could replace all of these servers
“Once you apply a voltage to represent some data with your artificial neuron, you have to erase and be able to write it again in the exact same way,” said lead researcher Jeehwan Kim.
MIT’s neuromorphic chip uses a similar model, but now has lattices of silicon germanium with one-dimensional channels so ions can flow through structured tunnels instead of completely freely.
“Ultimately we want a chip as big as a fingernail to replace one big supercomputer,” Kim said. “This [research] opens a stepping stone to produce real artificial [intelligence] hardware.”