The world of computing just got a heck of a lot more exciting thanks to IBM’s incredibly powerful brain-inspired chip which was unveiled on Thursday. While their prototype single-core system, released back in 2011 as part of the Systems of Neuromorphic Adaptive Plastic Scalable Electronics (SyNAPSE) project, was impressive, this new chip blows the old one out of the water.
The human brain tops the computing chart as the most efficient organizational system in the world, so it’s no wonder IBM and collaborators chose to emulate its capabilities for their new system. This so-called “cognitive computing” aims to mimic the brain’s abilities for perception, action and cognition.
Traditional computing systems can be likened to the left brain; they’re analytical and mathematical with superb number-crunching abilities. But if we want something that could be used in more sophisticated systems, say artificial intelligence, then we’re going to need something more right brain-like as well, which is exactly what IBM have been working towards.
This new chip, which is the size of a postage-stamp, addresses the right brain functions of sensory processing and pattern recognition. The idea is to be able to process, respond to and “learn” from information gleaned from the environment. If successfully combined with a traditional “left brain” system, which is what IBM will be attempting over the coming years, we could have a “holistic computing intelligence” with vast capabilities in our hands.
The product, which has been coined “TrueNorth,” achieves this through a staggering network of 1 million programmable neurons, 256 million configurable synapses (connections) and over 4,000 neurosynaptic cores. To put this into perspective, the prototype had just 256 neurons, 260,000 synapses and one core. That’s a giant leap in just 3 years. According to wired.com, these neurons, or “spiking neurons,” essentially allow the chip to encode data as patterns of pulses, which is much like one of the many ways scientists believe the brain stores information. Details of the chip can be found in Science.
IBM has put the abilities of this chip to the test in various artificial intelligence tasks, such as image recognition. One test, for example, involved presenting the chip with a variety of images, and it was found to be able to recognize a variety of objects with around 80% accuracy. Remarkably, the system was able to do all of this on just 63 mW of power.
Like the 2011 version, this chip is just a prototype. IBM hopes to eventually produce a neuro-synaptic chip system with a whopping 10 billion neurons and 100 trillion synapses that can process information whilst consuming just 1 kW of power. Eventually, multiple chips will be strung together on a chip board to create a huge network.
IBM envisages that the technology could have a variety of applications, such as vision assistance for the blind, health monitoring and transportation such as self-driving cars.