Grant funds hybrid system that could spur vast energy improvements to computing

The key commodities of classical computing, those silicon wafers and central processing units powering the surge in artificial intelligence (AI) hardware, are designed to overcome thermal fluctuations that might otherwise overwhelm the signals of interest. This also applies to advanced quantum computing where every material and component involved is cryogenically frozen, so the quantum signal is larger than any thermal fluctuations. But what if computer engineers could exploit that random noise, rather than fighting it, to maximize the efficiency of the AI hardware.

It might resemble what happens in biology.

“Biology has figured out how to exploit those fluctuations, that noise,” said Shantanu Chakrabartty, the Clifford W. Murphy Professor in the Preston M. Green Department of Electrical & Systems Engineering and vice dean for research in the McKelvey School of Engineering at Washington University in St. Louis. “If we exploit principles much like biology does, and we should be able to get much greater energy efficiency,” he added.

Chakrabartty already delves into the “neurobiological approach” to computer system architectures, he is part of a group of computer scientists that specializes in application of neuromorphic, or neurally inspired systems.

But to take advantage of the emerging field of thermodynamic computing, using those thermal fluctuations to power and drive processes, they must also employ some tricks from the quantum computing realm, specifically a concept known as quantum tunneling, a method of quantum mechanics that employs randomness to “tunnel” directly to the most optimized solution of any problem. Previously, Chakrabartty’s group has demonstrated fundamental single electron devices that operate using the Fowler-Nordheim quantum tunneling physics.

With more than half-million-dollar research grant from the National Science Foundation, Chakrabartty and his team will find a way create a hybrid system of that neuromorphic architecture plus quantum to create a room-temperature “single electron neuromorphic processor” where every electron matters. This would require using thermodynamic computing techniques to power and drive quantum fluctuations and will allow the researchers to account for every joule of energy that might be lost during computation.

 A related question that this research could also help answer is whether similar principles are in play within biological synapses and other cellular mechanisms. That’s a controversial topic, noted Chakrabartty. But this is also a big mystery in computing, why can’t the most sophisticated technology outperform simple biological mechanisms?

For instance, an insect brain with 1 million neurons consumes 10 microwatts of power less than a simple coin cell battery can provide “but does remarkable things compared to silicon systems,” Chakrabartty noted.

“Why is there a huge energy gap between silicon systems versus what biology does?”

Answering this question could be the key toward finding new and much more efficient ways to make the current AI energy-guzzlers more efficient in the future.

Click on the topics below for more stories in those areas

• Research
• Electrical & Systems Engineering