Abstract
Most unconventional computing hardware forces you to reshape your problem before you can solve it – embedding, sparsification, quadratisation – and each transformation costs you solution quality.
Our photonic probabilistic computer takes a different approach. Virtual hardware connections mean we can tackle higher-order, densely connected and dynamically changing combinatorial optimisation problems in their native form.
In this paper we show how that architectural advantage plays out in practice, predicting orders-of-magnitude speedups over digital annealing units on hard spin-glass problems — with favourable scaling as problem size grows.
Full pre-print available at: https://arxiv.org/abs/2605.06037