Deep learning control of THz QCLs

Abstract

Artificial neural networks are capable of fitting highly non-linear and complex systems. Such complicated systems can be found everywhere in nature, including the non-linear interaction between optical modes in laser resonators. In this work, we demonstrate artificial neural networks trained to model these complex interactions in the cavity of a Quantum Cascade Random Laser. The neural networks are able to predict modulation schemes for desired laser spectra in real-time. This radically novel approach makes it possible to adapt spectra to individual requirements without the need for lengthy and costly simulation and fabrication iterations.

Self‐Induced Mode‐Locking in Electrically Pumped Far‐Infrared Random Lasers

“…The RL modes compete for the available gain both in the spectral and spatial domains as an effect of spatial hole burning. [22] Then, a set of discrete modes having the lowest losses and the highest Q-factors are selected through optical feedback. For the 2d random resonators, a standard Gaussian-like distribution of the Q-values is found (Figure 1d), with the theoretical Q-factors of the resonator modes in Figure 1e spreading evenly around a mean Q tot = 37 for the pristine case, 36 for the EG mirror, and slightly lower (30) for the IG configurations.…”

“…A variety of approaches has been adopted to control RL modes [22] , such as tuning of the scatterer statistic arrangement and concentration, [23] external cavity configurations, [8] or the reshaping of their spectral emission via adaptive optical pumping. [24] Spontaneous mode-locking in RLs was theoretically predicted.…”