Probabilistic Computers Driven by Stochastic Nanomagnets Hint at Major AI Efficiency Gains to Come
Proven in prototype with a "spintronic p-bit," this new approach to computing could deliver a three-orders-of-magnitude energy saving.
Researchers from the University of California Santa Barbara, Tohoku University, the Japan Science and Technology Agency (JST), the National Institutes for Quantum Science and Technology, and the Inamori Research Institute of Science (InaRIS) have come up with a new approach to deliver energy-efficient computing — by combining a traditional complementary metal-oxide semiconductor (CMOS) circuit with stochastic nanomagnets.
"Our constructed prototype demonstrated that excellent computational performance can be achieved by driving pseudo random number generators in a deterministic CMOS circuit with physical random numbers generated by a limited number of stochastic nanomagnets," co-corresponding author Shunsuke Fukami explains. "Specifically speaking, a limited number of probabilistic bits (p-bits) with a stochastic magnetic tunnel junction (s-MTJ), should be manufacturable with a near-future integration technology."
The team's creation, which was prototyped using a traditional CMOS field-programmable gate array (FPGA) connected to a physical random number generator "probabilistic bit" that uses stochastic nanomagnets, targets artificial intelligence and machine learning (AI and ML) workloads — and comes in the face of projections that the energy needs of the world's data centers will treble by 2030 owing to growing demand for AI technology.
Rather than delivering a deterministic solution, as with a traditional computer, the team's approach aims to offer a "probabilistic computer" based on spintronic rather than electronic technology — wherein information is encoded with magnetism rather than electrons. When running probabilistic algorithms, the team projects that a future system built around the prototype could deliver the same performance as a CMOS-based electronic system but with a four-order-of-magnitude reduction in size and a three-order-of-magnitude reduction in energy consumption.
"We anticipate future research and development will advance," Fukami says, "leading to the implementation in society of an innovative computing hardware that boasts exceptional computational performance and energy-saving capabilities."
The team's work has been published in the journal Nature Communications under open-access terms.