Stochastic computing with biased optical parametric oscillators

In quantum optics, optical parametric oscillators (OPOs) have long served as testbeds for exploring quantum states and generating randomness. Here, we demonstrate that by applying vacuum-level coherent biasing, OPOs can be engineered as versatile hardware elements for stochastic computing. Specifically, we present a time-multiplexed, biased OPO system that enables discriminative and generative machine learning tasks, such as uncertainty-aware classification and the generation of MNIST digits directly from quantum vacuum noise. Beyond machine learning, we propose the use of biased OPO arrays for implementing fundamental stochastic logic gates and solving combinatorial optimization problems. Finally, we discuss the potential of on-chip OPO systems, which promise substantial improvements in latency and energy efficiency, paving the way for integrated stochastic computing architectures.