Forget Quantum Computing, Neuromorphic Cores Are The New Hotness!

Alongside the growth of quantum computing, several companies have been quietly figuring out better ways to improve processing power with Neuromorphic Cores.

Forget Quantum Computing, Neuromorphic Cores Are The New Hotness!

It's a truism in the tech world that computing power doubles once every 18 months. Known as Moore's Law, this dates back from the moment that the computer industry switched from vacuum tubes to transistors in the 1950s, and over the last few decades has mostly consisted of squeezing ever more transistors onto a chip. But there are physical limits to both the size of transistors and the number of angels that can dance on the heads of these pins, so scientists and technology experts have been searching for the next big thing for quite some time.

Core Wars

The prospect of quantum computing – which allows for a huge increase in processing power – has been mostly theoretical for many years now, but recent advances in solid state physics have brought it ever closer to reality. However, alongside the growth of quantum technology, several companies have been quietly figuring out better ways to improve processing power. Dual and quad core processors have become increasingly common, and improvements in multithreaded programming and parallel processing techniques allow applications to run nonlinearly and keep pace with the increase in independent processing cores that they have to deal with.

And so it was only a matter of time before the number of cores on a processor began to increase exponentially alongside the number of transistors. This advance has been driven not just by technological developments, but also by developments in artificial intelligence and neural networks – for example, look at the recent Nature article by Google in which the company claimed to have 'solved' the ancient, classic game of Go based on training a highly advanced neural network. Given this, the news that IBM has developed a computer with a whopping 4,096 cores that is inspired by the structure of the human brain should not come as a huge surprise.

Neuromorphic Range of Power

The term for this kind of chip is neuromorphic, which essentially means 'brain-shaped'. Alongside the thousands of cores, IBM's new baby has 1 million programmable neurons, 256 million programmable synapses and an astounding 5.4 billion transistors. It's the largest and most advanced computer chip ever made, and is incredibly efficient, running at just 72 milliwatts at max load. This is about 176,000 times more efficient than a standard modern CPU running the same kind of workload, and it's even 769 times more efficient than other neuromorphic style approaches. The efficiency of the human brain is even greater than this, and as we move into the era of artificial general intelligence and the analysis of big data, moving toward brain-like efficiency is going to be crucial.

IBM has already tested the chip on a variety of brain-like artificial intelligence tasks, like image recognition, and passes them at comparable speed and using much less power than other chips. The research is part of the SyNAPSE (Systems of Neuromorphic Adaptive Plastic Scalable Electronics) project, which is part of the US defense research agency Darpa's efforts to create brain-like hardware. The name alludes to synapses, the junctions between biological neurons, and the idea is to get beyond the von Neumann method of computing, where the storage and handling of data is divided up between the machine's memory and its central processing unit. This requires a lot of shuttling of information between sections of the system, which is computationally expensive, as well as power-hungry. Neuromorphic chips, however, process all this information locally and resembles the structure of the brain, where the separation of storage and computation isn't quite so clear cut.

Neurally Challenging

Of course, many challenges remain in the development and deployment of neuromorphic chips. Traditional chip design, for example, relies on high-quality electromagnetic interference (EMI) testing to ensure that the circuits do not interfere with other electronic devices; with such a dramatic change in hardware structure, this may affect the EMI compliance design of the associated printed circuit boards. Much research, development and testing is going to have to occur before these chips are seen anywhere other than supercomputers in research labs.

Money is also an object, no matter how much the US government is putting in. For the next phase of the project, DARPA added $16.1 million more to the IBM effort, but this money isn't inexhaustible. Various companies have tried to jump on the bandwagon; the recently formed startup company BrainChip Inc. is attempting to be the ARM of neuromorphic cores, licensing its technology and circuit designs for use by semiconductor companies for a percentage in royalties. With such huge upheavals occurring in the technology sector, time will tell if they succeed.