Probabilistic Computing Accelerated by Spike-Frequency Adaptation: Applications in Integer Factorization

We develop a local-heating scheme by drawing inspiration from spike-frequency adaptation (SFA) in neuroscience, which naturally fits the present probabilistic computing architecture. The SFA-based approach introduces negative feedback to the individual probabilistic bits (p-bits), effectively lowering the energy gradient in the Ising model and facilitating the system's escape from local minima in complex energy landscapes. Using integer factorization of semiprime numbers ranging from 16 to 30 bits as a case study, we demonstrate that the SFA algorithm significantly enhances computational efficiency, with an acceleration that exceeds linear scaling with the bit number. We further illustrate the practicality of our SFA algorithm through circuit simulations using stochastic magnetic tunnel junction-based p-bits. This approach not only accelerates integer factorization but also holds promise for addressing other largescale combinatorial optimization problems, thereby expanding the potential applications of probabilistic computing.