Programmable 200 GOPS Hopfield-inspired photonic Ising machine

Area of Science:

• Physics
• Computer Science
• Engineering

Background:

• Ising machines are promising for NP-hard problems, but current physical implementations lack scalability, reconfigurability, speed, and stability.
• Quantum annealers face scalability limitations due to quadratic qubit requirements for dense graphs.
• Existing photonic Ising machines struggle with large-scale problem solving.

Purpose of the Study:

• To introduce a programmable, stable, room-temperature optoelectronic oscillator (OEO)-based Ising machine with linear scaling.
• To demonstrate a photonic Ising machine capable of solving large-scale combinatorial optimization problems.
• To explore the potential of integrating digital signal processing (DSP) into optical computing for enhanced performance.

Main Methods:

• Developed a recurrent time-encoded loop architecture using cascaded thin-film lithium niobate (TFLN) modulators, a semiconductor optical amplifier (SOA), and a DSP engine.
• Achieved linear scaling in spin representation for problem solving.
• Utilized inherent noise from high baud rates to escape local minima and accelerate convergence.

Main Results:

• Demonstrated a system capable of solving fully connected problems up to 256 spins and sparse problems with over 41,000 spins.
• Achieved potential computational speeds exceeding 200 giga operations per second (GOPS).
• Obtained best-in-class solution quality for max-cut problems (2,000 and 20,000 spins) and ground-state solutions for number partitioning and lattice protein folding benchmarks.

Conclusions:

• The developed OEO-based Ising machine offers a scalable, high-speed, room-temperature solution for complex optimization.
• Integrating DSP enhances convergence and solution quality, paving the way for advanced optical computing.
• This platform opens new frontiers in ultrafast computing for optimization, neuromorphic processing, and analogue artificial intelligence.