An ambitious project to model the cerebral cortex in silicon is under way at Stanford. The man-made brain could help scientists understand how the most recently evolved part of our brain performs its complex computational feats, allowing us to understand language, recognize faces, and schedule the day. It could also lead to new neural prosthetics.
Silicon chips built to resemble the brain could shed light on its computational power.
“Brains do things in technically and conceptually novel ways–they can solve rather effortlessly issues which we cannot yet resolve with the largest and most modern digital machines,” says Rodney Douglas, a professor at the Institute of Neuroinformatics, in Zurich. “One of the ways to explore this is to develop hardware that goes in the same direction.”
Advertisement
Neurons communicate with a series of electrical pulses; chemical signals transiently change the electrical properties of individual cells, which in turn trigger an electrical change in the next neuron in the circuit. In the 1980s, Carver Mead, a pioneer in microelectronics at the California Institute of Technology, realized that the same transistors used to build computer chips could be used to build circuits that mimicked the electrical properties of neurons. Since then, scientists and engineers have been using these transistor-based neurons to build more-complicated neural circuits, modeling the retina, the cochlea (the part of the inner ear that translates sound waves into neural signals), and the hippocampus (a part of the brain crucial for memory). They call the process neuromorphing.
This story is only available to subscribers.
Don’t settle for half the story.
Get paywall-free access to technology news for the here and now.
Now Kwabena Boahen, a neuroengineer at Stanford University, is planning the most ambitious neuromorphic project to date: creating a silicon model of the cortex. The first-generation design will be composed of a circuit board with 16 chips, each containing a 256-by-256 array of silicon neurons. Groups of neurons can be set to have different electrical properties, mimicking different types of cells in the cortex. Engineers can also program specific connections between the cells to model the architecture in different parts of the cortex.
“We want to be able to explore different ideas, different connectivity patterns, different operations in these areas,” says Boahen. “It’s not really possible to explore that right now.” Boahen ultimately plans to build chips that other scientists can buy and use to test their own theories of how the cortex operates. That new knowledge can then be built into the next generation of chips.
“It’s very exciting,” says Terrence Sejnowski, head of the Computational Neurobiology Laboratory at the Salk Institute, in La Jolla, CA. “The technology has matured to the point where it’s possible to think about large-scale simulations.” For example, Sejnowski studies how the thalamus, a brain area thought to relay and integrate information from different parts of the brain, interacts with the cortex. “We can currently do small simulations of hundreds to thousands of neurons, but it would be great to be able to scale that up,” he says.
The million-neuron grid will have a processing speed equivalent to 300 teraflops, meaning that unlike computer-software simulations of the cortex, the hardwired silicon model will be able to run in real time. “Instead of running a thousand software instructions, it’s just current running through transistors, just like real neurons,” says Boahen.
Of course, the project will be a challenging one. “They will have to get a large number of chips to work together,” says Douglas. “To put together a structure on the scale Kwabena has in mind–no one has done that yet.” But it could become a turning point in the field. Douglas likens the current state of neuromorphic engineering to the early stages of computer-chip design. “People had been working on different types of logic gates, but it took a whole different worldview to build computer chips,” he says.
Engineers ultimately hope to use the information generated by the silicon cortex in a variety of ways–to build better neural prostheses, for example. “The real-time aspect of this technology allows us in principle to interface the silicon cortex with the real cortex or brain,” says Gert Cauwenberghs, a neuroengineer at the University of California, San Diego. “There is the promise, at least in the future, to build a prosthesis to replace some lost motor function or sensory function.”
