next up previous contents
Next: 9. Inverse Modeling with Up: 8.2 Optimization Results Previous: 8.2.1.2 Gradient Computation Phase

8.2.2 System Scalability

Since the time which is required to carry out an optimization can be quite long, scalability is an important issue. Given that the model must not be simplified any further and simulation tools do not offer any scalability (no inherent parallelization features), the single option is an improvement of the TCAD system's performance. The dominant part of optimization is the gradient computation. It can be accelerated as long as the overall number of jobs required during gradient computation exceeds the number of available CPUs. Therefore, in such a situation an optimization can easily be scaled by adding hardware to the system which is comparably simple and cheap.



Rudi Strasser
1999-05-27