Researchers Build Programmable Probabilistic Computer with One Million P-Bits
Tags Infrastructure · AI
A team of researchers demonstrated a programmable probabilistic computer built from one million p-bits (probabilistic bits) by networking multiple FPGAs into a single Ising machine. The system performs Gibbs sampling at over a trillion flips per second while keeping coupling weights in local on-chip memory. The paper, published on arXiv, establishes a quantitative design rule for scaling probabilistic computers beyond single-chip limits by identifying a critical timing ratio between boundary-exchange frequency and local p-bit update frequency. The system was demonstrated on spin glasses, Max-Cut, and Boolean satisfiability problems.
Technical significance
This work provides a concrete path to scaling probabilistic computing hardware beyond single-chip limits, with direct applications to optimization problems, sampling for machine learning, and Ising model simulation. The quantitative design rule (eta = f_comm/f_p-bit) gives engineers a principled framework for building larger distributed probabilistic systems.