Abstract
Probabilistic computing has gained attention for solving combinatorial optimization problems (COPs), mainly using the Ising model, which may not be suitable for complex COPs. Instead, this work proposes a multi-state probabilistic computing system based on the Potts model using stochastic threshold switching floating-body metal-oxide-semiconductor field-effect transistors (FB-MOSFETs) as the multi-state probabilistic bits (p-bits) to solve challenging COPs. The system employs drain voltage sharing and a one-hot sampling method to achieve controllable probabilistic behavior and scalable annealing. Experimental validations on spin glass and max-4-cut problems demonstrate that the system efficiently samples a tunable Boltzmann distribution while converging faster than traditional methods. Comparative analyses further highlight superior energy efficiency and decreased time-to-solution, underscoring the potential of multi-state probabilistic computing for large-scale, complex COPs using only MOSFET devices.