Description
Probabilistic computing aims to leverage randomness to solve difficult computational problems more efficiently than conventional methods. Photonic systems are particularly attractive owing to their ultrafast speeds, inherent parallelism, and access to quantum-limited noise sources. Existing photonic approaches have demonstrated applications ranging from random-number generation to Ising machine optimization, highlighting the versatility of nonlinear optical systems for probabilistic information processing.
Here we present a versatile platform for photonic computing based on polarization symmetry breaking in a nonlinear Kerr resonator. We drive our optical resonator with a 10GHz repetition rate pulsed laser source, allowing many picosecond duration pulses to recirculate at each roundtrip. With a slight change of driving frequency, each intracavity pulse undergoes polarization symmetry breaking, leading to an unbiased, random selection of one of two possible states. By encoding spin states in polarization with intensity-based readout, our platform eliminates the complex phase stabilization required by many existing implementations.
With no applied feedback, the system acts as an all-optical random-number generator seeded by vacuum fluctuations. By introducing a controllable bias, the same platform can also realize probabilistic bits (p-bits) with tuneable output statistics. We then show how an FPGA-based measurement-feedback system enables programmable spin Hamiltonians, allowing networks of p-bits to implement an optical Ising machine for combinatorial optimization. We will also discuss how the platform may naturally extend to higher dimensional systems, enabling vector spin solvers based on XY and Heisenberg spin systems.
| I am the presenting author | Yes |
|---|