|
 |
|
|
| Step 4 - Nominal Optimization |
Input
- Feasible Circuit
- Operation Conditions
|
- Perform sensitivity analysis
- Trace of design parameters
- Circuit sizing using different optimization
algorithms with worstcase operating conditions
|
Output
- Nominal Optimized Circuit
- Optimized Performances
|
WiCkeDTM DFM Optimization provides three algorithms for automatic performance optimization. The designer enters performance specifications and selects parameters to be modified. Available algorithms are:
- Least-squares. Performance specifications are fulfilled, but the optimizer does not try to over-fulfill them
- Parameter distances. The optimizer tries to over-fulfill each specification as far as possible. Weight factors can be changed by the designer.
- Stochastic. Parameters are modified in a random manner over a sequence of generations with decreasing variance.
The first two optimizers are gradient-based. They converge quickly to the next local optimum of their goal function. Since the design space is restricted to the small feasible region, this is usually the only optimum. In rare cases with ill-defined, non-continuous performance functions over a large design space with few constraints, the later two optimizers provide a global approach to optimization. For most practical circuits, gradient-based optimization outperforms global optimization regarding both algorithmic effort and quality of results.
|
|
|
