Semiconductor Process Simulation With RNN Step Profile Prediction

The patent segments the semiconductor manufacturing process into multiple discrete process steps, with each step represented by a dedicated process emulation cell in the RNN architecture. This segmentation allows the system to generate and analyze profiles for each intermediate step independently, rather than only producing the final profile. The segmented approach enables detailed inspection and fault tracing at each process stage.

The patent employs preliminary action by using the RNN model to predict and generate intermediate step profiles before actual manufacturing occurs. The system performs preliminary simulation of each process step to estimate profiles, allowing for pre-identification of potential faults and optimization opportunities before committing to the actual manufacturing process.

The patent replaces complex physical measurement and inspection systems with a computational RNN-based simulation system. Instead of using expensive and complex experimental methods to obtain profiles at each process step, the system uses machine learning models to substitute for physical measurement, significantly reducing both cost and technological complexity while maintaining the ability to generate multiple process step profiles.

The patent utilizes parameter changes by training the RNN model on historical process data with various parameters (process conditions, equipment settings, etc.) and then using the trained model to predict profiles under different parameter combinations. This allows for efficient process optimization by simulating how changes in parameters affect intermediate profiles without requiring actual experimental trials for each parameter combination.