There is a shift towards alternative hardware architectures and specialized devices in the quest for enhanced computing performance. Among these, universal quantum computing emerges as a solution to reduce algorithmic complexity in challenging tasks by leveraging entangled states. An alternative strategy involves transitioning from electron-based to photon-based systems within classical or quantum-coherent optical computing. The rationale for this shift is evident: photons travel at light speed, generate minimal heat, enable high-density configurations, and can effectively interact with matter to harness nonlinear effects. Optical technologies have become ubiquitous, evidenced by their presence in everyday applications like fibre optic communications and optical disc readers. Yet, a significant challenge in integrating photonics with CMOS architecture is the energy-intensive and time-consuming conversion of photons to electrons, posing a substantial bottleneck in hybrid systems.
Despite these challenges, optical hardware is being utilized in computing, particularly in data centres, for intensive machine learning (ML) applications and large-scale optimization tasks. Neural network (NN) architectures are especially suited to optical hardware, potentially offering high efficiency, rapid computing times, and reduced energy consumption due to the inherent physical properties of photonic systems.
Our research examines analogue optical computing through the lens of universal dynamical system descriptions. We focus on physical optimizers that utilize bifurcation dynamics and threshold operations, aiming to solve nonlinear problems and thus extending beyond the linear operations where optics excels. This approach offers the advantage of ultrafast emulation of dynamics, enabling optical systems to rapidly replicate complex systems' behaviour beyond the capabilities of digital simulations.
