Coherent Ising Machines for Combinatorial Optimization

Coherent Ising Machines (CIMs) represent a class of specialised computing architectures that map combinatorial optimisation problems onto the search for the ground state of an Ising Hamiltonian. By exploiting coupled physical systems—most commonly networks of optical parametric oscillators or equivalently engineered oscillatory elements—CIMs perform collective evolution of binary variables (spins) toward low‐energy configurations. Designed to tackle NP-hard tasks such as Max-Cut, graph colouring, travelling-salesman routing and scheduling, CIM platforms offer room-temperature operation, high parallelism and rapid convergence. Advances in hardware and control have broadened the range of implementable problems, with implementations spanning free-space and integrated photonics, opto-electronic oscillators, microelectromechanical systems and memristor crossbar arrays. Complementary theoretical developments in energy‐landscape analysis and annealing schedules have deepened understanding of CIM dynamics, informing strategies to navigate complex optimisation landscapes with phase transitions from smooth descent to rugged search. Together, these innovations underscore the potential of CIMs as versatile solvers for large-scale real-world problems.

Research from Nature Portfolio

A quantum-inspired parallel annealing approach implemented in an analogue memristor crossbar array has demonstrated full exploitation of device parallelism and analogue storage. By mapping quadratic binary problems onto conductance states of memristors and simultaneously updating all spins, this system achieves substantial improvements in time-to-solution and energy-efficiency over previous simulated-annealing and Ising-machine implementations, successfully solving unweighted and weighted Max-Cut instances and small travelling-salesman benchmarks. In a complementary development, a fully programmable CIM based on opto-electronic oscillators with self-feedback has been realised using injection-locked feedback loops instead of nonlinear optical gain. This compact setup encodes spins in oscillator intensity, simplifies hardware by removing large cavities and nonlinear optics, and delivers competitive performance on Max-Cut problems with up to one hundred spins, offering enhanced stability, reduced footprint and cost advantages.

Research from all publishers

A rigorous theoretical framework has been established through the geometric landscape annealing analysis of CIMs. By employing tools from random matrix theory, replica methods and supersymmetry‐breaking, researchers have characterised the evolution of the high-dimensional energy landscape as laser gain is ramped, identifying phase transitions from flat to rough to rigid regimes and deriving optimal annealing schedules that align with numerical experiments. Parallel to these theoretical advances, an experimental microelectromechanical Ising machine utilising a network of lithium niobate MEMS oscillators has been demonstrated. Operating above 9 GHz without requiring second-harmonic injection locking, the system employs a novel grouping algorithm to guarantee approximation ratios of 0.878 for Max-Cut and 0.658 for graph colouring, achieving fast sub-optimal solutions with minimal hardware complexity and pointing to scalable high-speed optimisation engines.

Coherent Ising Machines for Combinatorial Optimization publication trend

The graph below shows the total number of articles in coherent ising machines for combinatorial optimization across all publications each year (not limited to Nature Index journals).

Technical terms

Coherent Ising Machine (CIM): A physical computing system that encodes binary optimisation problems into coupled oscillators or optical elements, seeking the ground state of an Ising Hamiltonian.

Ising Model: A mathematical representation of interacting binary variables (spins) with an energy function that captures pairwise couplings and external fields, used to formulate combinatorial optimisation tasks.

Combinatorial Optimisation: The process of finding an optimal object from a finite but large set of discrete configurations, often characterised by NP-hard complexity.

Optical Parametric Oscillator (OPO): A nonlinear optical device that generates coherent light fields used in many CIM implementations to represent spin states via phase or amplitude.

Memristor: A non-volatile two-terminal electrical component whose conductance can be modulated and retained, enabling analogue storage and parallel computation for optimisation tasks.

References

1. Efficient combinatorial optimization by quantum-inspired parallel annealing in analogue memristor crossbar. Nature Communications (2023).
2. Geometric Landscape Annealing as an Optimization Principle Underlying the Coherent Ising Machine. Physical Review X (2024).
3. MEMS Oscillators‐Network‐Based Ising Machine with Grouping Method. Advanced Science (2024).
4. A poor man’s coherent Ising machine based on opto-electronic feedback systems for solving optimization problems. Nature Communications (2019).