Sub-mV tunable photonic p-bits for probabilistic computing

Abstract

Randomness, once dismissed as unwanted noise, is now emerging as a foundation for intelligent computation. Probabilistic bits (p-bits), which fluctuate between 0 and 1 with tunable probability, offer a route to solve complex problems through stochastic logic and energy-based optimization. Here, we present light-induced bias-tunable probabilistic-bit (LBP-bit) devices that generate entropy through light-induced charge polarity switching in a back-to-back junction. The probability of each device's stochastic bitstream can be precisely tuned with submillivolt bias without disturbing the underlying distribution. This unique separation of randomness generation (by light) and probability control (by bias) enables stable control of output probability, essential for scalable probabilistic computing (p-computing). The proposed p-computing framework demonstrates integer factorization as a representative example of probabilistic search on computationally intensive problems. Max-Cut problems, representative combinatorial optimization tasks, are evaluated, demonstrating that light in this device functions as the stochastic source enabling probabilistic computation.