Abstract: Surface profile measurement of silicon wafers after flattening and polishing can quantify wafer flatness and is a key part of quality control.However, the principle of high flatness silicon wafer surface inspection equipment is complex, with over 10, 000 inspection sites and time up to 1 minute, while the large size of the equipment makes it difficult to integrate in the polishing machine for in-situ inspection of silicon wafers, which limits the production efficiency.Machine learning techniques with excellent data analysis and prediction capabilities can improve production efficiency and accuracy when applied in industrial production.In this study, a neural network-based wafer profile prediction model is developed for 300 mm wafers based on the IC material genome platform.The model greatly reduces the input information required to characterize the wafer profile, and a prediction value with an absolute average error of 4.07 nm can be obtained from the thickness information of only 14 key points, resulting in a highly accurate prediction of the wafer profile in only about 10 s and reduce the hardware requirements of the measurement equipment.The prediction of the wafer shape can be performed with high accuracy in only 10s, providing an idea for fast measurement in wafer flattening process.