Neural network computing using a large-area vertical-cavity surface-emitting laser

High-performance computing hardware is crucial for advanced neural network (NN) schemes. Photonics promises strong advantages in terms of parallelism, yet until now scalable and integrable concepts are scarce and partially rely on exotic substrates. The majority of large scale and parallel photonic NN demonstrations are neither standalone nor autonomous [1] , usually lacking fundamental NN constituents or requiring substantial interaction with a classical electronic computer. In this contribution, we implement a fully parallel photonic reservoir computer based on the spatially distributed modes of an efficient and fast semiconductor laser [2] . Crucially, all neural network connections are realized in hardware, and our laser-based and fully parallel NN comprising ~100 neurons produces results without pre-or post-processing.