Bayesian optimization of quantum cascade detectors

A Bayesian optimization algorithm in combination with a scattering based simulation approach is used for the optimization of quantum cascade detectors (QCDs). QCDs operate in the mid-infrared and terahertz regime and are, together with quantum cascade lasers, appropriate for the integration into on-chip applications such as gas sensors. Our modeling approach is based on a rate equation model and a Kirchhoff resistance network for noise modeling, using scattering rates calculated with Fermi’s golden rule, or alternatively extracted from an ensemble Monte Carlo transport approach. The appropriate surrogate model of Bayesian optimization is based on Gaussian process regression, which can handle noisy offsets on the objective function evaluations inherent in ensemble Monte Carlo simulations. Here, we focus on the optimization of a matured mid-infrared QCD design detecting at 4.7  $$\upmu {\mathrm{m}}$$ . For optimization we choose as figure of merit the specific detectivity, which is a measure for the signal-to-noise ratio. As the trade-off between high extraction efficiency and low detector conductance is important for good detection performance, we search for the perfect layer composition and vary the thicknesses of different cascade layers. Due to the high-temperature requirements interesting for cost-effective and mobile on-chip sensing applications, a simulation temperature of 300 K is selected. Our optimization strategy yields an improvement of specific detectivity by a factor of $${\sim 2-3}$$ at room temperature using two different parameter sets. Furthermore, we investigate the sensitivity of our approach to fabrication tolerances, showing robustness of the optimized designs against growth fluctuations under fabrication conditions.