Reservoir computing using spin-VCSELs - the influence of timescales and data injection schemes

Abstract Reservoir computing implemented in photonic systems promises fast and energy efficient computations. Due to their particularly fast spin-flip dynamics, vertically emitting semiconductor lasers with two spin polarized charge carrier populations (spin-VCSEL), are good candidates for high-speed reservoir computing. However, spin-VCSELs are complex dynamical systems, and in order to properly utilise their dynamics for computation, a thorough understanding of the interplay of dynamical variables and external control parameters is needed. With our work, we numerically show the crucial impact of dynamic coupling and decay timescales on the prediction performance of a spin-VSCEL reservoir computer. We present numerical evidence of the critical impact of different data injection schemes and internal timescales. Injection schemes based on the phase of the injected light and on generating spin polarized charge-carriers are compared as a function of the spin carrier lifetimes. A central finding is that the internal relaxation dynamics of the charge carriers is only beneficial for the performance with encoding via the spin polarized current. If the data is encoded via an optical phase difference, these carrier dynamics, and with it the additional memory, cannot be utilised. We find strong correlations with the underlying delay induced bifurcation structure, which allows to transfer the results also to other physical reservoir computing systems.