Supercomputer simulating the human brain will be put into use in 2024

An "artistic photo" of the "Shennan" supercomputer. Image source: British "New Scientist" weekly website

According to a report on the website of the British New Scientist magazine on the 12th, a supercomputer that can fully simulate human brain synapses will be put into use in Australia next year. This neuromorphic supercomputer named "Deep South" can perform 228 trillion synaptic operations per second, which is equivalent to the number of synaptic operations in the human brain estimated by scientists, and will help us understand how the human brain processes large amounts of information while consuming relatively little energy.

The research team pointed out that similar neuromorphic computers have been available before, but Shennan will be the largest so far. Shennan was built by the International Center for Neuromorphic Systems in Sydney, Australia, in collaboration with Intel and Dell. Unlike ordinary computers, Shennan's hardware chips can implement spiking neural networks, thereby modeling the way synapses process information in the brain.

Andrew Van Scheik, the project leader of Deep South, pointed out that this will be the first time they have simulated the activity of a human brain-sized spike neural network in real time. Although Deep South is not as powerful as existing supercomputers, it will help advance the understanding of neuromorphic computing and biological brains, thereby gaining a better understanding of how the brain works.

Supercomputers are energy hogs, while the human brain consumes no more energy than a lightbulb. This difference is partly due to the different ways data is processed: as traditional computers perform operations, data is constantly moved between the processor and memory; neuromorphic architectures perform many operations in parallel, greatly reducing the movement of data. Since data movement is one of the most energy-intensive parts, neuromorphic architectures can significantly reduce energy consumption. In addition, spiking neural networks are event-driven, which means that neuromorphic systems only react to changes in inputs rather than running continuously in the background like traditional computers, further reducing energy consumption.

Researchers say Shennan will help advance neuroscience research and pave the way for more energy-efficient computing. If the technology can be miniaturized, it could improve the autonomy of drones and robots.