The brain's an incredibly rich and complex computational core that we don't really fully understand—but that isn't stopping IBM building a new form of computing architecture around what's happening inside our heads.
Building on a collaboration it ran with DARPA's SyNAPSE program back in 2011, where it developed a neurosynaptic computing chip, the company is taking things a step further to create a whole computational ecosystem inspired by the brain. That original chip simulated some of the brain's functions—including the inter relations between 530 billion neurons—but the idea now is to create something even more realistic.
That means creating a whole new computing architecture—which is exactly what IBM is doing. So, as it's just announced, it's building all kinds of weird and wonderful things: multi-threaded software simulators that replicate the way the brain processes data; a neuron model that uses deterministic and stochastic computations to make sense of the world; and programs made out of arrays of "corelets," each of which represents a discrete neurosynaptic core. This is the stuff of science fiction.
It's a lot to imagine, but the applications are mind-boggling: computers that can think and process data just—or at least, a little—like humans. It's still some way off, mind, especially given that even an 83,000-processor supercomputer can still only match 1 percent of your brain right now.
That doesn't deter IBM, though. Apparently, its long-term goal is to essentially build an actual silicon version of the brain. As they put it, that means they need a chip system with ten billion neurons and hundred trillion synapses, which consumes one kilowatt of power and occupies less than two litres in volume. That's what people refer to as a challenge. [IBM via Engadget]