IBM claims first in 'neuromorphic' computing

IBM claims microprocessor that mimics human brain can perform 46 billion operations per second drawing just 70 milliwatts of power

IBM and researchers at Cornell University have developed what they claim is the world's first production-scale "neuromorphic" or "neurosynaptic" computing chip.

The development was revealed in a paper published in the journal Science.

Christened TrueNorth, the company claims that the chip works in a similar fashion to the human brain by using densely interconnected webs of up to 5.4 billion transistors laid-out like the human brain's neural networks or neurons, which communicate with each other via some 256 million synapses (junctions between two nerve cells).

The human brain works by recognising the different patterns that 100 billion neurons and 100 trillion synapses form to recognise different objects, thoughts, feelings and so on. IBM claims that TrueNorth has one million "neurons", which IBM claims is on a par with the brain of a bee.

Because it has a vast number of circuits working in parallel, TrueNorth is also able to perform some 46 billion operation per second, per watt of energy used, making it vastly more energy efficient than today's most power-efficient microprocessors.

While the microprocessor contains 5.4 billion transistors, it draws just 70 milliwatts of power - compared to the 35 watts or more typically consumed by a standard Intel i5 PC microprocessor.

Horst Simon, deputy director of the Lawrence Berkeley National Laboratory, compared the design to the advent of parallel supercomputers in the 1980s. "It is a remarkable achievement in terms of scalability and low power consumption," he told the New York Times.

IBM claims that the development of neurosynaptic microprocessors has been fast, with programmable neurons increasing between 2011 and 2014 from just 256 to one million, while programmable synapses have increased from 262,144 to 256 million. It has increased synaptic cores, meanwhile, from one to 4,096.