Home | Digitize the circuits

Digitize the circuits

The brain is a powerful natural computer, processing millions of signals on the power of nothing more than a sandwich. To reverse-engineer it, you need to mimic that speed and efficiency.

If a scientist gives a mouse and a personal computer the same task, the mouse's brain will operate 9,000 times faster than the PC while using 40,000 times less energy.

One second of mouse time takes a PC two-and-a-half hours.

When it comes to simulating more complex neural patterns, a machine with such low speed and power efficiency limits the scope of what scientists can hope to build.

Kwabena Boahen, professor of bioengineering, has developed a circuit board that might take electronic brains to the next level. Called Neurogrid, the board resembles the brain in both energy efficiency and organization and has the computing capacity of 1 million neurons.

Kurt Hickman

Neurogrid simulates the function of one million neurons and billions of synaptic connections using significantly less power than a PC. Kwabena Boahen is working with other scientists to develop prosthetic limbs controlled by Neurogrid.

"Neurogrid is the first system to simulate a million neurons connected by billions of synapses in real time," Boahen said. "It does this using about three watts of power. That's orders of magnitude more efficient than a supercomputer, which normally is used to do these kinds of simulations and uses megawatts of power."

Neurogrid mimics the richly branching structure of neurons. With this design and power, scientists could model large-scale simulations of the brain. The more streamlined the design, the more complex simulations will be achievable.

"The whole brain – the visual cortex, the motor system, the decision-making – all of that could be put into one system," said Boahen, who is also a member of Stanford Bio-X. "We could simulate it all in real time with only milliwatts of power."

Right now, Neurogrid is about the size of an iPad. But within five years, Boahen and his team hope to scale it down to a tiny chip. When they do, that single chip could mirror the speed and efficiency of the human brain, at a fraction of the cost of a supercomputer that would normally be required to run brain-like simulations.