ARTIFICIAL INTELLIGENCE-BASED MODELING OF HEAT TREATMENT EFFECTS ON THE ELECTRICAL AND OPTICAL PROPERTIES OF SEMICONDUCTOR MATERIALS-A LITERATURE-CONSTRAINED COMPUTATIONAL STUDY

The semiconductor films’ defect populations, grain growth, phase stability, charge transport and photon–matter interactions are governed by heat treatment. Don’t use dashes or em dashes. Conventional optimization involves repeated cycles of deposition, annealing and characterization, which becomes costly when the temperature, holding time, atmosphere, film. This study proposes a physics-informed artificial-intelligence framework predicting three coupled responses: optical band gap, logarithmic electrical resistivity and visible-light transmittance. A dataset of 480 records was generated within literature-reported domains on ZnO, SnO2, In2O3 and CdO thin films. Material identity, annealing temperature and duration, oxygen fraction and thickness, heating rate. An 80:20 stratified split was used to train the random forest, gradient boosting, support vector regression, and multilayer perceptron models. The evaluation of performance was carried out through determination coefficient, mean absolute error, root mean square error, residual behaviour, cross validation. Gradient boosting resulted in the most balanced estimates with test R2 values of 0.985 for band gap, 0.923 for log-resistivity, and 0.880 for transmittance. The most important factor for predicting resistivity was temperature; the identity of the material and its thickness were important for controlling the band-gap and transmission responses. According to response-surface analysis, there exists an intermediate processing window around 430–510 °C for ZnO films when low resistivity and high transparency are optimized simultaneously under moderate oxygen content. The findings show that using interpretable machine learning can link conditions of processing to useful properties while retaining known physical trends. Since the data set is literature-constrained and synthetic, rather than stemming from a new experimental campaign, the performance numbers establish computational feasibility, not performance for all materials. The workflow suggested assists in planning an experiment, screening of process windows and uncertainty-aware optimization. It will be equipped to integrate future laboratory measurements.