A multi-objective clonal selection algorithm for analog circuit and solar cell design

We present PareDA (ParetoDesignAutomation), a composite automated methodology for the simulation-based multi-scenario multi-objective optimization of analog circuits and thin-film cell devices, relying on randomized algorithms, both domain and constraints sensitivity analysis, epsilon-dominance and global robustness analysis. We test PareDA algorithm on the designing problem of a three stage operational amplifier, a yield-aware optimization of a folded-cascode operational amplifier (requiring multiple operating conditions) and an optical model for tandem thin-film silicon solar cells. In these scenarios, comparisons with state-of-the-art techniques (as NSGA-II and YdIRCO) undoubtedly demonstrate PareDA effectiveness, in terms of Pareto optimality of the design found and convergence time. The latter obtains, in fact, a significant average performance improvement (from 35% to 49%), finding Pareto-optimal designs dominating the ones found by state-of-the-art algorithms. Moreover CPU time required by PareDA to converge is reduced of at least 75% compared to the other methodologies here analysed (e.g. optimal folded- cascode operational amplifier are found in just 320 s). Finally, PareDA algorithm thanks to parallel computations gains a 5.62x speed-up with 70% efficiency, compared to the non-parallel version.