Solving complex optimisation problems remains a significant challenge in modern computing, and researchers are increasingly exploring novel approaches beyond traditional digital systems. Nayem AL-Kayed from Queen’s University, Charles St-Arnault from McGill University, and Hugh Morison from Queen’s University, alongside colleagues including A. Aadhi and Chaoran Huang, present a new photonic Ising machine capable of performing 200 giga-operations per second. This innovative system, inspired by the principles of Hopfield networks, overcomes limitations of existing technologies by offering a scalable and reconfigurable platform for tackling NP-hard problems with up to 41,000 spins. The team demonstrates superior performance on benchmark problems, including Max-Cut, number partitioning, and even lattice protein folding, previously inaccessible to photonic systems, and importantly, leverages inherent noise and digital signal processing to enhance both speed and solution quality, paving the way for advancements in optimisation, neuromorphic computing, and artificial intelligence.
Thin-film Ising Machine for Optimization Problems
This research details the development and experimental validation of a novel photonic Ising machine, a type of analog computer designed to solve optimization problems. The team constructed a system using thin-film lithium niobate modulators to create connections between computational units called spins, successfully demonstrating the machine’s ability to tackle various optimization challenges, including square lattice problems, Max-Cut, number partitioning, and even a simplified protein folding problem. Sophisticated digital signal processing is employed for both transmitting and receiving signals, improving signal quality and minimizing errors, while intentional optical noise injection improves the machine’s ability to escape local minima in the optimization landscape. The performance of this photonic machine was compared to that of a digital quantum annealer, demonstrating competitive results, and the analog nature of the machine offers the potential for faster and more energy-efficient computation compared to digital approaches. Results demonstrate that the intentional injection of optical noise improves performance, and the digital signal processing stack is crucial for achieving high-speed and accurate operation.
Optoelectronic Oscillators Implement Room-Temperature Ising Machine
Researchers engineered a room-temperature Ising machine based on optoelectronic oscillators, creating a scalable platform for solving complex optimization problems. This system utilizes cascaded thin-film lithium niobate modulators, a semiconductor optical amplifier, and a digital signal processing engine within a recurrent time-encoded loop, implementing the Ising model’s energy function as a matrix-vector multiplication problem mapped onto optical and electronic signals for high-speed computation. The team implemented the Ising machine with two cascaded Mach-Zehnder modulators operating at 1310nm in a closed-loop feedback configuration, leveraging its nonlinear transfer function to approximate spin states and linearly modulating a fixed interaction matrix representing the problem’s couplings. Optical element-wise multiplication and electronic summation, facilitated by a quantum dot semiconductor optical amplifier and photodetector, combine these signals, while high-speed electro-optic feedback, completed via analog-to-digital and digital-to-analog converters operating at 256 GSa/s, enables iterative convergence toward the ground state.
To validate the system, the team investigated bifurcation, observing a transition from monostability to bistability in individual spins as feedback strength increased, confirming the presence of sufficient nonlinearity and feedback for supporting the required dynamical behavior. Further validation involved performing large-scale matrix-vector multiplication at 64 GBaud using randomly sampled matrices, demonstrating the system’s capability for high-speed spin-spin coupling and its potential for tackling complex optimization problems with up to 256 spins, and scaling to 41,000 spins in sparse configurations. The system also leverages inherent electrical noise and a dedicated optical noise source to emulate annealing, accelerating convergence and improving solution quality.
Optoelectronic Ising Machine Achieves Record Scale and Speed
Researchers have developed a room-temperature Ising machine based on optoelectronic oscillators, achieving a significant leap in spin configuration and computational speed. This system utilizes cascaded lithium niobate modulators, a semiconductor amplifier, and digital signal processing to create a recurrent loop capable of performing 200 giga-operations per second for spin coupling and nonlinear operations, supporting fully-connected problems with up to 256 spins, representing 65,536 couplings, and scaling to 41,000 spins with over 205,000 couplings in sparse configurations. Experiments demonstrate best-in-class solution quality for Max-Cut problems, achieving results for graphs containing 2,000 and 20,000 spins, and successfully obtaining ground-state solutions for challenging benchmarks including number partitioning and lattice protein folding. Measurements confirm the system’s ability to accurately perform matrix-vector multiplication, achieving 212 GOPS on a single wavelength channel at 106 GBaud, nearly doubling previously reported performance, with accuracy remaining stable at 96.
2 ±0. 4 % as matrix size increased from 32 to 200. Further characterization revealed that the system’s performance is influenced by baud rate, with accuracy decreasing as the symbol rate increased, but this degradation, caused by system noise and distortion, can be beneficial, acting as a form of annealing that helps the system escape local minima and converge toward optimal solutions. The system was successfully applied to benchmark problems, including the 2D square lattice graph and the Max-Cut problem on Gset graphs, achieving a solution quality of 99. 48% of the best-known cut value of 13289 cuts for a Gset graph with 104 nodes.
Scalable Photonic Ising Machine Solves Complex Problems
This research introduces a new approach to solving complex optimization problems using a room-temperature photonic Ising machine, demonstrating a scalable and stable platform based on optoelectronic oscillators. The team successfully built a system capable of handling problems with up to 256 spins, and simulations suggest scalability to 1024 spins, achieving high-quality solutions for benchmark problems like Max-Cut, number partitioning, and lattice protein folding. The demonstrated architecture distinguishes itself through its stability and speed, operating without the need for active stabilization and leveraging inherent noise to improve performance. The researchers highlight the benefits of integrating digital signal processing directly into the computational loop, enhancing convergence and solution quality, and suggest that commercially available DSP technology could reduce latency and increase speed by a factor of ten. Further improvements could be achieved through hypermultiplexing techniques, potentially enabling parallel computation and even greater throughput.