Probabilistic computers might one day tackle certain problems beyond standard computers while avoiding the hardware challenges that vex quantum computing. Scientists have now created the largest probabilistic computer yet, with 1 million “probabilistic bits,” or p‑bits, spread across multiple chips. Unlike regular bits that are fixed 0 or 1, and qubits that can be 0, 1, or any superposition, p‑bits flip between 0 and 1 with a tunable probability. When many such noisy bits operate in a correlated fashion they can solve stochastic problems such as optimization and sampling tasks.

Earlier machines held only eight p‑bits in 2019 and 7,200 p‑bits on a single chip by 2023. Scaling required linking many chips, a difficult task because the device relies on correlated fluctuations. The new system uses 18 field‑programmable gate arrays (FPGAs)—reconfigurable electronic chips—to implement the p‑bits in software, achieving more than a trillion flips per second. Kerem Çamsarı, associate professor at UC Santa Barbara, notes the machine communicates without global lockstep synchronization, a finding highlighted by Navid Anjum Aadit, a Stanford postdoc, who discovered a simple design rule for inter‑chip data exchange that balances speed and accuracy.

These results suggest a path to arbitrarily large probabilistic computers built from many chips, much like scaling conventional CPUs or GPUs. The approach works with various hardware; future work aims to use specialized stochastic memory such as MRAM or magnetic tunnel junctions for greater energy efficiency. Systems combining CMOS with dense stochastic memory technologies such asic memory technologies such as MRAM offer one of the most compelling paths forward, according to Aadit, opening the door to practical probabilistic computing for real‑world applications.