During analog integrated circuit design, a compromise between different competing circuit performances must be found. The so-called Pareto front shows all optimal compromises between the circuit performances. Novel deterministic methods for the efficient computation of the Pareto front are presented in this work. A new parallel optimization algorithm based on Sequential Quadratic Programming is described, which features an exchange of solutions between search runs for different points on the Pareto front and a special treatment of optimization constraints. The Pareto front is computed in a new iterative manner by increasing the number of considered performances step-wise. This leads to a better coverage of peripheral parts of high-dimensional Pareto fronts. Finally, the work presents a general problem formulation and efficient optimization approaches for the computation of a specification Pareto front that considers process variations and operating conditions.
«
During analog integrated circuit design, a compromise between different competing circuit performances must be found. The so-called Pareto front shows all optimal compromises between the circuit performances. Novel deterministic methods for the efficient computation of the Pareto front are presented in this work. A new parallel optimization algorithm based on Sequential Quadratic Programming is described, which features an exchange of solutions between search runs for different points on the Par...
»