Novel approach for KrF chemically amplified resist optimization assisted by deep learning

The development of chemically amplified resists requires many experiments to optimize the chemical composition, which includes the type of monomer molecules and their component ratios, initiator concentration, and process conditions. In addition, the optimization process requires extensive knowledge and experience. In this paper, we apply deep learning to predict the exposure properties, such as sensitivity and contrast, of KrF chemically amplified resists and to optimize the ratio of monomer components. The experimental data are used to predict photoresist development properties by deep learning using in-house code. To achieve this goal, we synthesized several photoresist resins with different proportions. Each resin was then used to prepare photoresist formulations, which were subsequently subjected to exposure and development testing under various energy conditions. Using the film thickness data obtained, we trained our deep learning system to more comprehensively predict the exposure and development curves of photoresists under different resin component conditions. The results of validation experiments showed that the predicted results were consistent with the experimental results, and the predictions for the exposure and development characteristics of different monomer component ratios were quite accurate, confirming that the deep learning outcomes possess high credibility and feasibility.