The shorter process time is always better for turnaround time (TAT) and cost reduction in semiconductor manufacturing. In this paper, two machine learning (ML) models, e.g., Neural Network (NN) and Support Vector Regression (SVR), are adopted for the enhancement of ash rate in a plasma ash process. By the use of ash rate data recorded from design of experiments (DoE), the model of NN and SVR affords the fitting with R 2 value of 0.999 and 0.782, respectively, after training and >97% accuracy from the test validations. Two new ash recipes predicted by NN and SVR models are proved to show 5.4% and 4.2% faster ash rate than the data of Process of Record (POR). Moreover, they give 1.0% and 3.6% error between the actual and predicted ash rates. As a result, the NN-based ash recipe became the new POR in terms of faster TAT.