Classification and Prediction of Wafer Yield Using Gradient Boosting Algorithms: A Focus on Lithography Inline Parameters

Yield is one of the most critical driving factors in the semiconductor industry. Much time and resources are spent improving yield by identifying and solving problems. A yield prediction method will reduce the time and cost spent on yield improvement measures by indicating early process problems. Our investigation predicted the wafer test yield using the lithography process inline parameters. We consider different variants of the gradient boosting algorithm to compare the prediction results. Two years of real production data was used to train, validate, and evaluate the different algorithms. Following the proposed algorithm, the Classification algorithm predicts yield better than the numerical algorithm. Among the classification algorithms, CatBoost shows the highest accuracy of 98.54%, whereas the best Mean Absolute error (MAE) of 0.88 is indicated by the gradient boosting regressor among numerical algorithms.