Photonic neuromorphic computing using vertical cavity semiconductor lasers

Photonic realizations of neural network computing hardware are a promising approach to enable future scalability neuromorphic computing. The number special purpose and photonics has accelerated on such scale that one can now speak Cambrian explosion. Work along these lines includes (i) high performance for artificial neurons, (ii) the efficient scalable implementation network’s connections, (iii) strategies adjust connections during learning phase. In this review we provide an overview vertical-cavity surface-emitting lasers (VCSELs) how high-performance electro-optical components either implement or combined with additional photonic demonstrate points (i-iii). neurmorphic context, VCSELs exceptional interest as they compatible CMOS fabrication, readily achieve 30% wall-plug efficiency, >30 GHz modulation bandwidth multiply accumulate operations at sub-fJ energy. They hence highly energy ultra-fast. Crucially, react nonlinearly optical injection well electrical modulation, making them suitable all-optical neurons. Their cavities wavelength-limited, standard semiconductor growth lithography enables non-classical cavity configurations geometries. This excitable (i.e. spiking VCSELs) finely control their temporal spatial coherence, unlock terahertz bandwidths through spin-flip effects, even leverage quantum electrodynamics further boost efficiency. Finally, VCSEL arrays 2D integration, but emission vertical substrate makes ideally suited integrated networks leveraging 3D waveguides. Here, discuss spatially temporally multiplexed reservoirs, computation basis concepts advances in fabrication microlasers. outlook roadmap identifying possibilities some crucial milestones field.