Akademik Veri Yönetim Sistemi
Solar energy, cilt.307, sa.114339, ss.1-15, 2026 (Scopus)
In this study, a Deep Q-Learning-based, artificial intelligence (AI)-assisted multi-objective optimization framework was developed for the design of semi-transparent cadmium telluride (ST-CdTe) solar cells. The proposed approach integrates optical analyses based on the Transfer Matrix Method with photovoltaic simulations using SCAPS-1D to simultaneously optimize all layers in a multilayer CdTe-based solar cell with a dielectric/metal/dielectric (DMD) transparent contact. The Deep Q-Learning agent autonomously learned to balance the trade-off between maximizing photocurrent density and achieving the target average visible transmittance (AVT = 20%, 25%, 30%) through interactions within the simulation environment. Optical analyses revealed that strong electric field confinement and wave-guided photon trapping were preserved even in ultra-thin CdTe absorber layers (<150 nm). The integration of the MoO3/Au/WO3-based DMD transparent contact architecture not only maintained semi-transparency in the visible spectrum but also enhanced photon harvesting via internal optical reflection. SCAPS-1D results confirmed that the optimized structures achieved photovoltaic performance of Voc ≈ 0.95 V, Jsc ≈ 20–21 mA cm−2, and PCE ≈ 15.7%. These findings demonstrate that Deep Q-Learning-based AI systems can simultaneously optimize optical and electrical parameters, offering a physics-informed redefinition of the transparency–efficiency trade-off in ST-CdTe solar cells. The proposed method introduces a broadly applicable paradigm for AI-driven design in photonic and optoelectronic device engineering.