- Australian scientists are developing “DeepSouth”, the first supercomputer capable of processing information as fast as the human brain.
- Operating at 228 trillion synapses per second, DeepSouth will provide insights into neural efficiency while using just 20 watts.
- Analysts say understanding resource-efficient biological computation could enable more powerful artificial intelligence.
Brain-like supercomputer
Scientists down under are building what they say will be the first supercomputer capable of processing information at the same blazing speed as the human brain.
The new machine, dubbed “DeepSouth,” will carry out 228 trillion synaptic operations per second when it goes online next year, rivaling the estimated processing capability contained within our three-pound organic computers.
Researchers from Western Sydney University are developing DeepSouth to provide new insights into how brains can operate so efficiently using such little power.
It’s about intelligence AND power usage
“Our system will change that,” said Prof. André van Schaik, director at the university’s International Centre for Neuromorphic Systems.
Currently, efforts to simulate neural networks are “hampered by our inability to simulate brain-like networks at scale,” he said.
DeepSouth will use just 20 watts of power – the same as the human brain.
Understanding how brains achieve such computational complexity with little energy could someday allow engineers to build more powerful cyborg brains.
Revolutionizing neuroscience
The new supercomputer could revolutionize neuroscience research, analysts say.
“This will be the hardware to [better understand] the brain,” said Ralph Etienne-Cummings, an engineering professor at Johns Hopkins University.
Etienne-Cummings added that the machine may also enable rapid prototyping of new AI and engineering solutions requiring brain-like computation.
DeepSouth is just one of several projects trying to match human cortex capabilities. Others are exploring “biological computers” powered by neurons and brain cells.
