Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain
the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in
Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles
and JavaScript.
Unconventional computing with optical simulation of spin Hamiltonians
Guest Edited by Natalia Berloff (University of Cambridge, UK) and Prof Alireza Marandi (California Institute of Technology, USA), in collaboration with our Editorial Board Member Prof Mohammad Ali Miri (City University of New York, USA).
Lattice spin models, e.g., XY, Ising and Potts models, are widely utilized in statistical mechanics and in condensed matter physics for exploring magnetism. These models are important tools for exploring phase transitions and critical phenomena. In addition, spin Hamiltonians have been celebrated in the context of computer science as interesting models that can represent a large range of computationally-hard optimization problems. Subsequently, over the years there has been an interest in realizing physical systems that are governed by spin-like Hamiltonians for unconventional computing applications.
In recent years, several works revealed that networks of coupled optical oscillators, e.g., lasers and optical parametric oscillators, show a great promise for emulating a classical spin model. In such systems, the evolution of the oscillator network is toward an equilibrium amplitude and phase pattern that could represent the ground state of the corresponding spin model Hamiltonian. This Collection aims to curate, as a single resource, interesting research articles on the subject of “Unconventional Computing with Optical Simulation of Spin Hamiltonians” from a broad pool of scientists and accelerate the formation of a roadmap for future research directions of the field.
Analog Ising machines are promising fast computing schemes for some difficult optimization problems, yet their analog nature is known to cause errors and inhibit computational performance. Here, the authors investigate how the choice of nonlinear transfer functions partly suppresses errors caused by analog amplitude inhomogeneity, which leads to order-of-magnitude differences in the computation time.
Spatial photonic Ising machines (SPIMs), a variant of optical Ising machines, are promising for large-scale problems but have limitations on the type of problems that can be mapped on them. Here, a variant of SPIMs is demonstrated that can realize anti-ferromagnetic Ising problems with some further experimental simplifications that have potentials for future large-scale Ising machines.
The strong nonlinearity and absence of particle conservation leads to non-equilibrium nature of exciton-polariton condensates. Here, an unsupervised machine learning approach is employed to map phases of a polariton condensate lattice, and classify unique polarization patterns
The advantage of unconventional computing architectures is commonly demonstrated by solving an NP-hard problem, but some instances are easier to solve than others. Here, an optimisation simplicity criterion is proposed that classifies the complexity of instances on optical or electronic neuromorphic computers.
Coherently-coupled optical systems with different types of couplings between coherent centres are promising platforms for optical simulations of spin glasses and, therefore, for optical optimization. Here, an experimental implementation of dissipative and dispersive couplings is realized between two photon Bose-Einstein condensates, extending the range of physical models that can be addressed with optical simulators.
Physics-inspired algorithms are being developed to solve NP-hard problems while alleviating issues with scaling and trapping in local minima. Here, a method to search the global minimum of the Potts Hamiltonian with a photonic-inspired model is proposed.
Hybrid computing seeks to divide operations based on the strengths of digital, analogue or physical architectures. Here, approximate solutions to the multi-state Potts model are found using a physical Ising solver, networked degenerate optical parametric oscillators, repeatedly with learning processes.
Ising machines aim to find the global minimum of Ising energy in the fewest time steps. Here, simulated thermal fluctuations are added to simulated bifurcation to quickly escape from local minima.
Coherent Ising machines represent an optical approach to unconventional computing that fail if inhomogeneity of the analog spins’ amplitude becomes too great. Here, an approach to amplitude control is shown to improve their performance even when Zeeman terms are included.
Optical Ising machines provide a promising approach to solve complex optimization problems and hence are of broad interest in physics society. This paper constructs a nonlinear optical Ising machine with spatial light modulators to find distinct phase transitions, which demonstrates a platform for solving optimization problem in more efficient way.