All-Optical Excitable Spiking Laser Neuron in InP Generic Integration Technology

Brain-inspired, neuromorphic devices implemented in integrated photonic hardware have attracted significant interest recently as part of efforts towards novel non-von Neumann computing paradigms that make use of the low loss, high-speed and parallel operations in optics. An all-optical spiking laser neuron fabricated on the indium-phosphide generic integration technology platform may be a practical alternative to other semi-integrated photonic and electronic-based spiking neuron implementations. Owing to the large number of predefined building blocks, a plethora of applications have benefitted already from the generic integration process. This technology platform has now been utilised for the first time to demonstrate an all-optical spiking laser neuron. This paper present and discusses the design and measurement of the ultra-fast and rich spiking dynamics in these devices. We show that under external pulse injection and operated slightly below the lasing threshold, the laser neuron exhibits an excitable mode, in addition to a self-spiking mode far above the threshold when no pulse is injected. In the excitable mode, the required injected pulse energy is much lower than that of the generated excited response, meeting an important requirement for neuron cascadability. In addition, we investigate excitability at different injection wavelengths below the lasing wavelength, as well as the ultra-fast temporal properties of the spiking response. All of the discussed characteristics point to the laser neuron being an important candidate for scaling up to future fully-connected, multi-wavelength all-optical photonic spiking neural networks in indium-phosphide generic integration technology.

Phase-space analysis of a two-section InP laser as an all-optical spiking neuron: dependency on control and design parameters

Using a rate-equation model we numerically evaluate the carrier concentration and photon number in an integrated two-section semiconductor laser, and analyse its dynamics in three-dimensional phase space. The simulation comprises compact model descriptions extracted from a commercially-available generic InP technology platform, allowing us to model an applied reverse-bias voltage to the saturable absorber. We use the model to study the influence of the injected gain current, reverse-bias voltage, and cavity mirror reflectivity on the excitable operation state, which is the operation mode desired for the laser to act as an all-optical integrated neuron. We show in phase-space that our model is capable of demonstrating four different operation modes, i.e. cw, self-pulsating and an on-set and excitable mode under optical pulse injection. In addition, we show that lowering the reflectivity of one of the cavity mirrors greatly enhances the control parameter space for excitable operation, enabling more relaxed operation parameter control and lower power consumption of an integrated two-section laser neuron.