Multi-objective optimization and analysis for the design space exploration of analog circuits and solar cells

This paper introduces PAREDA (ParetoDesignAutomation), a composite automated methodology for the optimization of analog circuits and solar cell devices. The PAREDA framework combines randomized algorithms, domain and constraints sensitivity analysis, epsilon-dominance and global robustness analysis in order to perform simulation-based, multi-scenario and multi-objective optimization. PAREDA is evaluated on the problems of designing a three-stage operational amplifier, a yield-aware optimization of a folded-cascode operational amplifier (requiring multiple operating conditions) and a model for selective emitter solar cells. Comparisons with a selection of state-of-the-art techniques (such as NSGA-II and YdIRCO) highlight the effectiveness of PAREDA both in terms of Pareto optimality of the solutions found and time-to-converge. The solutions obtained by PAREDA dominate those of comparative techniques, in particular, the proposed technique shows a significant average performance improvement (ranging from 35% to 49%) with respect to such techniques. Moreover, the CPU time required by PAREDA to converge is smaller of at least 75% if compared with the other methodologies here analyzed (e.g. significantly improved designs for folded-cascode operational amplifier are found in just 320 s). Finally, the PAREDA algorithm can also benefit from parallelization, which leads to a significant speed-up with respect to the nonparallel version.