Researchers in the UK turned on the most powerful supercomputer designed to mimic the human brain for the first time last week, marking a big step toward replicating the power of the brain with silicon.
Using computers to mimic the brain, also known as neuromorphic computing is a rapidly growing area of computer science research that focuses on developing system architectures and specialized computer chips that replicate the way the human brain processes information. Not only will this allow neuroscientists to create unprecedented models of the brain, but it will also allow roboticists to create robots that can navigate complex environments using computer vision.
One of the biggest technical challenges in this regard is creating hardware that can handle the massive parallel processing capacity that defines the fleshy computer in our skulls. The Manchester supercomputer consists of one million processor cores that are capable of performing 200 trillion operations per second and has been under construction at the University of Manchester since 2006.
At the heart of the machine is a special computer chip that consists of 18 tightly-packed processor cores. Over 55,000 of these chips were linked together to endow the supercomputer with a million processor cores altogether that are coordinated using a new type of computer design known as a “Spiking Neural Network Architecture” (SpiNNaker).
In the human brain, neurons communicate with each other through brief flashes, or “spikes,” of electrochemical energy. Individual spikes don’t contain much information, but when hundreds of millions of spikes are happening simultaneously, it enables the sophisticated information processing that allows you to understand this sentence.
The SpiNNaker supercomputer borrows from the brain’s design insofar as the processor cores send billions of small packets of information to tens of thousands of destinations within the system. This is different from the way a normal computer routes information, which usually involves much larger packets of information because they must also tell the system how to route the information. The SpiNNaker supercomputer is able to drastically reduce packet size because the transmission of information is handled by the custom chips.
Put more simply: if normal computers are like sending mail through the postal service—complete with sending and receiving addresses, and a stamp—the SpiNNaker supercomputer is like sending mail if everyone had their own dedicated postman who knows the routes so well there’s no need to address the mail.
“SpiNNaker completely re-thinks the way conventional computers work,” Steve Furber, a computer scientist at the University of Manchester, said in a statement. “We’ve essentially created a machine that works more like a brain than a traditional computer, which is extremely exciting.” Furber and his colleagues aim to eventually model up to a billion neurons in real time using the SpiNNaker supercomputer. While this is significantly more neurons than any other computer is capable of modeling today, it still represents only one percent of the human brain’s 100 billion neurons.
Still, there is a great deal researchers can learn with the SpiNNaker supercomputer. One of its most promising use cases is modeling isolated brain networks. The computer has already been used to create an 80,000 neuron model of the cortex, the outer layer of the brain responsible for processing sensory information. It has also been used to control a robot that can process visual information in real time to navigate a complex terrain.
“Neuroscientists can now use SpiNNaker to help unlock some of the secrets of how the human brain works by running unprecedentedly large scale simulations,” Furber said.