Delay-based reservoir computing with spin-VCSELs: Interplay between internal dynamics and performance

Machine learning setups that are able to process data in the optical domain are ideal for on -chip hardware implementations [1]. Due to the fact that the training of hardware based solutions is complicated, a delay-based reservoir computing (RC) realization, where only the output weights need to be trained via linear regression, is very promising [2]. In this paper we investigate vertical cavity surface emitting laser with two mode emission (spin-VCSEL) as the nonlinear node for a delay-based RC setup. These lasers have the ability to exibit reprodicible and high speed dynamics [3] and are thus ideal candidates to increase the data injection rates which are limited by the clocktime [4], [5]. The focus of our numerical investigations is on the interplay between the internal charge carrier dynamics of the spin-VCSEL and its performance when operated in a delay-based RC setup with optically-injected phase-modulated data injection.