Improving photonic Ising machines: a guide to reaching the ground state
Jacobus Lamers · VUBIR (Vrije Universiteit Brussel) · 2024
With ever increasing computational demands, the need for efficient computer architectures is more and more pressing.Photonic Ising machines are a promising way to tackle challenging optimization problems.The photonic Ising machines considered here encode the variables of the problem in some property of a laser in a feedback loop, such as the phase or the intensity and evolve them dynamically.After some time, this system relaxes to a steady state, corresponding to a solution of the problem.However, obtaining the optimal solution is not always guaranteed as the system can get trapped in local minima of the cost function.Previous research has revealed that the gain of the feedback loop and the coupling strength have a great influence on being able to reach the ground state.Here, using a stability analysis, we demonstrate that by choosing the gain correctly, we can make sure that the ground state can be reached by scanning the coupling strength slowly.Interestingly, this is not possible for all physical implementations of the photonic Ising machine and we find that photonic Ising machines using a Mach-Zehnder modulator have a distinct advantage.