NUS Builds 144-Device Spintronic Processor, Cutting Optimization Energy Use 58.3%
Updated
Updated · Newswise · Jul 2
NUS Builds 144-Device Spintronic Processor, Cutting Optimization Energy Use 58.3%
3 articles · Updated · Newswise · Jul 2
Summary
Two Nature Communications studies from NUS showed probabilistic spintronic processors can solve optimization problems faster and with less power, including a 3.2-fold speedup over a CPU and 58.3% energy savings.
The first system used 144 magnetic tunnel junction random-number generators in a parallel probabilistic Ising processor for quadratic assignment problems, where it consistently returned feasible, high-quality solutions as problem sizes grew.
D-Wave quantum annealers, used as a benchmark in those tests, struggled to produce feasible solutions on larger instances, positioning the spintronic approach as a nearer-term alternative to quantum hardware.
A second 250-device probabilistic Ising machine delivered 10-fold acceleration on sparsely connected graphs, while simulated quantum annealing improved solution quality 20 times over conventional simulated annealing.
The team plans to scale the hardware through chiplet-based architectures for AI, logistics, scheduling, financial modeling, communications and chip design.
By turning randomness into a resource, could this new hardware fundamentally change how we approach computing and AI?
Is spintronic computing the quiet competitor that will solve optimization problems before quantum computers are ready?
Spintronic Processors Achieve 3.2x Speedup and 58% Energy Savings: A Practical Alternative to Quantum Computing for Complex Optimization
Overview
Researchers at the National University of Singapore have developed a new probabilistic spintronic processor that uses tiny magnetic devices and randomness as a computing resource. This breakthrough, recently published, introduces a practical and energy-efficient way to solve complex optimization problems, offering a strong alternative to traditional and quantum computing. By integrating stochastic magnetic devices with parallel architectures and advanced annealing algorithms, the processor accelerates optimization tasks while reducing energy consumption. This innovation marks a significant step forward in computing, making it possible to tackle challenging problems faster and with less power.