Bioengineers at the University of Stanford have developed a new type of circuit board modeled on the human brain. The Neurogrid, as it’s called, operates about 9,000 times faster using significantly less power than a typical PC. While no match for the power and capability of the human brain that inspired it, the new advance has major implications for robotics and computing.
The closer these bio-inspired chips get to real thing (i.e., an actual human brain) the better they will be at reproducing biological actions. Potential applications include prosthetic limbs that could mimic the speed and complexity of biological entities.
Leading the project is Kwabena Boahen, associate professor of bioengineering at Stanford. In a recent article for the Proceedings of the IEEE, Boahen highlights some of the main differences between the human-powered “computer” and the man-made one. A mouse brain for example is 9,000 times faster than a computer built to simulate its functions, and the computer takes 40,000 times more power to operate.
“From a pure energy perspective, the brain is hard to match,” Boahen states in an interview with Stanford News.
Boahen and his team represent one of the leading efforts in the field of neuromorphic research, which aims to reproduce brain functionality using a mix of silicon and software. Their iPad-sized system, called Neurogrid, is made up of a circuit board with 16 “Neurocore” chips. One Neurogrid board can simulate one million neurons and billions of synaptic connections in real-time. Power efficiency was a key design metric for the chips. The project is said to be able to simulate orders of magnitude more neurons and synapses than other bio-inspired computing devices using the footprint and power profile of a typical tablet computing device.
“The human brain, with 80,000 times more neurons than Neurogrid, consumes only three times as much power,” Boahen writes in the IEEE piece. “Achieving this level of energy efficiency while offering greater configurability and scale is the ultimate challenge neuromorphic engineers face.”
The million-neuron prototype was made possible with a grant from the National Institute of Health’s five-year Pioneer Award. Next up for the project will be reducing design costs and developing more straightforward compiler software. Boahen suggests that modern mass-scale manufacturing processes would lower the price significantly, from roughly $40,000 for the prototype down to about $400 per unit. The end goal is to make the technology accessible from a cost and usability standpoint.