A Reconfigurable Photonic Signal Processor Based on Spiking VCSEL Neurons

Given that a single neuron in the brain is capable of performing complex operations, replicating the high-fidelity functions of biological neurons is a critical area of research within photonic brain-like computing. Here, we report the numerical demonstration of a reconfigurable photonic signal processor to perform versatile functionalities, including all-optical logical computations, image processing, and automotive obstacle avoidance. The photonic signal processor consists of a vertical-cavity surface-emitting laser (VCSEL) neuron with dual-polarization optical injection. The reconfigurability is offered by the experimentally controllable injection parameters, i.e., the injection strength and frequency detuning. On the one hand, by modulating the injection intensity, different binary inputs are temporally encoded: a reduction in intensity signifies a “0”, while maintaining a constant injection intensity denotes a “1”. By doing so, the outputs are encoded as “1” (a single spike) or “0” (no spike) for the result of photonic logic operations, where the orthogonal polarization modes of the VCSEL act as input channels for Boolean values, facilitating the Exclusive OR (XOR) and Negated OR (NOR) operations as well as Gray code implementation in a single delay-free step. On the other hand, by manipulating the frequency detuning, we further demonstrate the adjustable threshold characteristics of the proposed VCSEL neuron. Leveraging this feature, we implement the Canny edge detection algorithm in a photonic neuron and accomplish multi-dimensional obstacle avoidance tasks including car following, side objects or pedestrians, and overtaking. Our approach perfectly simulates the biological XOR neuron in principle and expands the application potential of a single photonic neuron with high reconfigurability, highlighting a valuable reference for future research on photonic chip integration based on VCSEL neurons.