Skip to main content

One Million ARM Cores Linked to Simulate Brain

Typically most references relating to ARM's architecture centers around the mobile sector including smartphones and tablets. But the UK's Manchester University is doing something entirely different with ARM-based cores: it's simulating the brain.

In conjunction with Southampton, Cambridge and Sheffield Universities, the Manchester University has created a massively parallel computer called SpiNNaker, short for Spiking Neural Network architecture, using up to one million cores based on an old ARM instruction set architecture. The project has been backed by a £5 million (about $8 million) grant from the British government.

Naturally the project isn't replicating the human brain at a 1:1 scale – there are around 100 billion neurons with 1,000 trillion connections taking place within our skulls. That said, the 1 million ARM cores will represent only 1-percent of the human equivalent. Instead of neurons transmitting information along analog electrical spikes, cores will use packets of descriptive data. To process the information, SpiNNaker will use virtual neurons. In the end, it can transmit and process data just as fast as the brain but with fewer physical components.

"We don't know how the brain works as an information-processing system, and we do need to find out," said Manchester's Professor Steve Furber who has been studying brain function and architecture for several years. "We hope that our machine will enable significant progress towards achieving this understanding."

According to Fubar, SpiNNaker's 130-nm CMOS processor was manufactured by Taiwan-based UMC. Each silicon die consists of 18 cores, memory and requires 1-watt of power. There are also around 100 million transistors scattered across 55 32-kb SRAM blocks. Micron Technology provided the accompanying 1 Gbit DDR SDRAM memory die which operates up to 166 MHz.

ARM has been supporting the SpiNNaker project since 2005.