5.0 Conclusions

, we can get a measure of the relative importance of certain parameters with respect to the output function. For instance, it is clear that the length of transistor M6 has a bigger effect on the sensitivity of speed to Ld than other parameters. Table 2 summarizes the best choices of each parameter for every performance. The last column is a selection of parameter settings that represent a few optimal design choices. Different choice criteria could be applied to select an optimal design. Although it is not necessary to execute many confirmation runs, it is encouraged to explore some of the options around the chosen optimal design point. We decided to chose the combination [3,1,1,3,3,1] as an optimal design point and we have verified that the discrepancy between predicted values and simulation results for each output performance is much less than the accepted error, defined as three times the root mean squared error at the experimental points. Table 3 highlights the improvements in robustness of the comparator circuit as well as improvement in the power performance, for an unchanged switching speed. In spite of the fact that our original comparator design was hand optimized before it was used with DORIC, this methodology has successfully identified an equivalent circuit with significantly better manufacturability and improved performance. The Robust Design Method provides circuit designers with an efficient, simple and systematic way of improving their circuit performance, and increase robustness of their circuits to manufacturing variations. DORIC was developed as a user-friendly CAD tool that supports this methodology and summarizes the results in a powerful graphical representation. We showed how this experimental design method and the use of DORIC helps improve several manufacturability and performance characteristics of a basic clocked compara-tor.