The world's first supercomputer capable of simulating networks at the scale of the human brain has been announced by researchers at the International Centre for Neuromorphic Systems (ICNS) at Western Sydney University.
DeepSouth uses a neuromorphic system which mimics biological processes, using hardware to efficiently emulate large networks of spiking neurons at 228 trillion synaptic operations per second - rivalling the estimated rate of operations in the human brain.
ICNS Director, Professor André van Schaik says DeepSouth stands apart from other supercomputers as it is purpose-built to operate like networks of neurons, requiring less power and enabling greater efficiencies. This contrasts with supercomputers optimised for more traditional computing loads, which are power hungry.
“Progress in our understanding of how brains compute using neurons is hampered by our inability to simulate brain like networks at scale. Simulating spiking neural networks on standard computers using Graphics Processing Units (GPUs) and multicore Central Processing Units (CPUs) is just too slow and power intensive. Our system will change that,” Professor van Schaik said.
“This platform will progress our understanding of the brain and develop brain-scale computing applications in diverse fields including sensing, biomedical, robotics, space, and large-scale AI applications.”
Professor van Schaik explained that practically this will lead to advances in smart devices, such as mobile phones, sensors for manufacturing and agriculture, and less power-hungry and smarter AI applications. It will also enable a better understanding of how a healthy or diseased human brain works.
Western Sydney University’s ICNS team collaborated with partners across the neuromorphic field in developing this ground-breaking project, with researchers from the University of Sydney, University of Melbourne, and University of Aachen, Germany.
The supercomputer is aptly named DeepSouth, paying homage to IBM's TrueNorth system, which initiated efforts to build machines simulating large networks of spiking neurons, and Deep Blue, which was the first computer to become a world chess champion. The name is also a nod to its geographical location.
DeepSouth will be based at Western Sydney University and is a key contributor to the growth of the region as a high-tech hub.
DeepSouth aims to be operational by April 2024.
Key Benefits of DeepSouth:
- Super-fast, large scale parallel processing using far less power: Our brains are able to process the equivalent of an exaflop — a billion-billion (1 followed by 18 zeros) mathematical operations per second — with just 20 watts of power. Using neuromorphic engineering that simulates the way our brain works, DeepSouth can process massive amounts of data quickly, using much less power, while being much smaller than other supercomputers.
- Scalability: The system is also scalable, allowing for the addition of more hardware to create a larger system or scaling down for smaller portable or more cost-effective applications.
- Reconfigurable: Leveraging Field Programmable Gate Arrays (FPGA) facilitates hardware reprogramming, enabling the addition of new neuron models, connectivity schemes, and learning rules—overcoming limitations seen in other neuromorphic computing systems with custom-designed hardware. DeepSouth will be remotely accessible with a front end that allows description of the neural models and design of the neural networks in the popular programming language Python. The development of this front-end enables researchers to use the platform without needing detailed knowledge of the hardware configuration.
- Commercial Availability: Leveraging commercially available hardware ensures continual improvements of the hardware, independent of the team designing the supercomputer, overcoming limitations seen in other neuromorphic computing systems with custom designed hardware. Custom chips take a large amount of time to design and manufacture and cost tens of millions of dollars each. Using commercial off-the-shelf configurable hardware means that the protype would be easy to replicate at data centres around the world.
- Artificial Intelligence: By mimicking the brain, we will be able to create more efficient ways of undertaking AI processes than our current models.
ENDS
