Benchmarking Quantum Kernels on Real-World Quantum Cascade Laser Datasets

We benchmark quantum kernel approaches against classical kernels for predicting premature failure in quantum cascade lasers (QCLs), an electrically-injected semiconductor laser. Our analysis uses a complex, noisy data set comprising the electrical and optical properties of approximately 100 QCLs, collected during their first 400 hours of operation. We implement quantum-assisted support vector machines (QSVMs) to predict premature failure in QCLs, utilizing various quantum kernels, and comparing their performance to support vector machines (SVMs) using radial basis kernels. The simulations are performed on both ideal and noisy quantum backends to quantify the impact on model generalization. We further explore kernel constructions tailored to the time-dependent nature of the data. This study provides quantitative benchmarks for evaluating classical versus quantum kernel performance under realistic noise conditions with real-world datasets, with implications for predictive modeling and reliability analysis of semiconductor laser systems.