ABSTRACT
Driving a laser or modulator directly from a photodetector is atypical for photonic systems because it lacks receiver circuitry. Even analog photonic links usually include a transimpedance receiver, although analog regenerators are generally not possible because signal-to-noise ratio (SNR) irreversibly degrades with each stage of a communication link. In a neural processing network, the neurons themselves reduce the SNR by applying nonlinear saturation or dynamics. The chapter discusses the processing-network node as a whole unit within an analog photonic processing network, focusing on the analog electronic link connecting wavelength division multiplexed weighted additon to a laser. Since the L-I curve includes saturation region at both small and large input amplitudes, a network of these units can in principle approximate any function or simulate any dynamical system. The device could serve as a template for future network-compatible laser neuron models, which could potentially operate at higher signal bandwidths with richer, or more efficient laser dynamics.
