上海科技大学知识管理系统

The interband cascade laser (ICL) represents a significant class of mid-infrared lasers, offering valuable applications across a range of scientific and technological domains. The conventional approach to designing ICL relies on the expertise of the designers and extensive simulation tests, which is time-consuming and restricts the flexibility of the design process. In this paper, we present an automated wavefunction identification program that rapidly and accurately identifies key wavefunctions in ICL band diagrams using neural networks (NNs), achieving an area under curve of greater than 90%. Based on the results of the automatic identification, a NN model is employed to predict the key performance metrics of ICL. The model focuses on transition energies and overlap of electron and hole wavefunctions in the W-type active region, as well as energy level differences D1 and D2 between electron and hole wave functions in the injector. By employing automated hyperparameter optimization, a mean square error of 10-4 was attained after 100 epochs, with high R-squared values of 0.994, 0.946, 0.947 and 0.982 for the transition energy, D1, D2 and overlap. Moreover, it takes only 12 s for the trained NN to obtain the results of 2000 structures, which is about 20 000 times faster than the traditional simulation method (240 000 s). On the basis of the predicted results, the optimal structure of the ICL was rapidly identified while simultaneously considering the W-type active and injected region wave functions. The predicted optimal structure for the 4.6 mu m wavelength emission achieves a high overlap (0.347) at a low theoretical average electric field (64 kV cm-1). The results obtained by our approach were found to be in close agreement with the real simulation results, with a maximum error of only 3.03%. This provides a valuable strategy and a convenient method for optimizing ICL designs in the future.