Integrated statistical process and device simulation system with automatic calibration using single-step feedback and backpropagation neural network

An integrated simulation system that combines process and device simulation is developed in Submicron Development Center of AMD to address the following issues: 1. The manufacturing data are fundamentally statistical. It is difficult to measure exact values for each parameter in every process step, and single data point is simply insufficient to justify any decision. 2. A high order of data accuracy is required for fine-tuning an established fabrication process. 3. A high degree of flexibility and out-of-spec prediction ability is required for process development. 4. Fast ramp-up time for process control demands minimum engineering calibration effort. 5. Development engineers need full flexibility to modify process flow, process parameter, as well as underlying device physics models. These require a combination of statistical approach, device physics modeling and a series of data processing techniques.