McGill and Queen's University researchers have built an improved version of a computer that uses light to solve extremely hard problems more quickly and at larger scale than existing systems, without the need for cryogenic cooling.
Many of the most demanding real-world challenges in science and engineering, from predicting how proteins fold to optimizing shipping routes, belong to a class of problems known as non-deterministic polynomial (NP) problems. As these begin to include more variables, or scenarios, the number of possible solutions explodes, and the time required to solve them on conventional computers increases exponentially.
"This exponential scaling is a major bottleneck for progress in several fields," said David Plant, Tier I Canada Research Chair, and senior author of the study published in Nature and a professor of electrical engineering at McGill.
Built an Ising machine
To address this, the team built a photonic Ising machine - a non-traditional computing system that models complex problems using the physics of light. Unlike other non-traditional computing approaches that struggle with scaling, stability or the need for cryogenic operation, this photonic Ising machine runs at room temperature and is designed to remain stable as problems scale up.
Previous digital Ising solvers relied on computing-clusters to emulate the Ising model, demanding substantial resources and energy to perform the required calculations. Physical Ising machines suffer from stability issues and challenges in controlling a large number of physical parameters, restricting their practical applications in advanced computation to small scale problems.
To build the optical computer comprised of more than 20 ultra-sensitive optical components, the team developed new control algorithms and signal-processing methods to both accelerate computation, and keep the system stable.
"We built the entire photonic Ising machine," Charles St. Arnault, lead McGill Ph.D. student said. "This involved putting together multiple ultra-sensitive components, writing completely novel control algorithms, and digital signal processing to keep the system stable. We found that those signal processing algorithms sped up the computation, reducing the number of iterations required to reach the optimal solution."
The researchers report that their system is the largest-scale and most stable photonic Ising machine that exists so far. It also operates at the highest computational speed shown for a photonic system, reaching 212 giga-operations per second for a single computation core.
"This new photonic Ising machine allows us to solve problems of unprecedented scale for any analog Ising machine class," St. Arnault said.
As a proof of concept, the team used the platform to solve real-world problems, including protein folding, which is important for understanding diseases and designing drugs. They also showed performance that exceeds that of quantum annealers. Quantum annealers are one of the main competing technologies for solving hard optimization problems, but they are expensive, hard to scale and need cryogenic cooling.
Faster, cheaper solutions to complex problems
The researchers say this faster and more scalable optimization could accelerate drug discovery, improve vaccine development and reduce costs and emissions in logistics and transportation.
"This research matters because it opens a path to solving complex problems much faster, at lower cost and with less consumed power the researchers. The new photonic Ising machine runs at very high speed, at room temperature, and scales to problem sizes that were previously out of reach, even for quantum devices."
About the study
"Programmable 200 GOPS Hopfield-inspired photonic Ising machine" by Charles St-Arnault, David Plant, and colleagues at Queens University was published in Nature.
