This is the question that Steve Furber, the main designer of ARM chip, is aiming to answer. Furber, the ICL Professor of Computer Engineering at University of Manchester, has estimated that a single ARM core can simulate around 1000 spiking neurons. In his project, dubbed SpiNNaker, looks to link ARM cores in a manner similar to that of the human brain.
ARM processors, a small yet powerful chip that runs on simple instructions in order to process tasks quickly, have been around in some form since the 1980s, and require relatively small amounts of processing power. These characteristics make the chip the ideal choice for the SpiNNaker project.
The project entails creating a supercomputer cluster with 1 million processors (each processor emulating 1000 neurons) for a total of 1 billion artificial neurons. That might sound like a lot, but it’s only a fraction of the estimated 80-90 billion neurons that the typical brain has. As the article states, this also depends on “how old they are, and how much drinking and brown acid they have done.”
By combining multi-core chips with a network of chips, researchers hope to create a scaled functional analog of the human brain. The brain is considered a massively parallel system, where large problems are split into smaller sections, and those sections are solved in parallel, or simultaneously. By using 20-core chips and assigning one of those cores to be the “control,” Furber hopes to emulate the processing style of the brain.
Aside from his obvious familiarity with the ARM architecture, Furber has other reasons for his choice.
“Embedded processors can reduce the capital and energy costs of a given level of compute power by about an order of magnitude, thereby significantly reducing the ownership (and environmental) costs. The embedded processor technology employed in SpiNNaker delivers a similar performance to a PC from each 20-processor node, for a component cost of around $20 and a power consumption under 1 watt.”
The individual SpiNNaker chips (click that link) is linked to surrounding chips (see top diagram) via the Network On Chip functionality. This allows for high speed communications without extra peripherals. The ultimate goal of the project is to mimic the brain’s biological structure and functionality in order to better learn how neural pathways are utilized and how failures within the brain are overcome.
If you’re worrying about Skynet, don’t (yet.) The million core system is still around 1/100th of the processing power of the human brain. And by the time cores-per-chip reaches a number where we can easily match the processing power, it’s estimated that it will require 25kW of power, as opposed to the human brains 20Watt consumption.