Chapter 5 Robust Performance Tailoring with Tuning - SSL - MIT
Chapter 5 Robust Performance Tailoring with Tuning - SSL - MIT
Chapter 5 Robust Performance Tailoring with Tuning - SSL - MIT
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
have a very small nominal performance range (341.43µm to 369.1µm) indicating that<br />
the design is quite robust to uncertainty. Unfortunately, the robustness comes at<br />
the price of nominal performance and none of the hardware simulations meet the<br />
requirement. Upon tuning these hardware simulations the performance range shrinks<br />
further (331µm to 349.63µm) due to a lack of tuning authority in the robust design,<br />
and all of the tuned designs also fail to meet the requirement.<br />
The two RPTT designs are differentiated in the plot by the value of α placed<br />
parenthetically in the label. As seen in the previous simulation, both RPTT designs<br />
show a blend of the PT and RPT design characteristics. At α = 0, the nominal<br />
performance range (232.52µm to 565.5µm) is greater than that of the RPT design<br />
but much smaller than that of the PT design. This result indicates that even <strong>with</strong> no<br />
weight on the robustness cost the RPTT formulation produces a design that is dra-<br />
matically less sensitive to uncertainty than the PT optimization. The RPTT design<br />
also has greater tuning authority than the RPT design, and the tuned performances<br />
range from 237.67µm to 312.45µm. The maximum value of this range is well below<br />
the requriement of 330µm, so that all of the tuned RPTT designs are successful.<br />
The second RPTT design, acheived through an RPTT optimization <strong>with</strong> α =0.1,<br />
shows increased robustness to uncertainty <strong>with</strong> a nominal performance ranging from<br />
261.10µm to 448.9µm. The range of tuned performances is also smaller than that<br />
in the first RPTT design starting at 274.30µm and continuing to just under the re-<br />
quirement at 320.15µm. All of the design simulations are again tunable to below the<br />
requirement, but the RPTT design <strong>with</strong> α =0.1 issuperiortothatforα =0in<br />
that it is just tunable enough and therefore more robust to uncertainty. The tuning<br />
authority and robustness is balanced such that the design is perfectly tailored to meet<br />
the requirement.<br />
The accompanying bar chart, Figure 5-10(b), presents the percent of success-<br />
ful simulations for each design. As seen in the previous hardware simulation (Fig-<br />
ure 5-8(b)) only the RPTT designs are successful for all of the simulations, and the<br />
RPT design never meets requirements. At the higher uncertainty level just over 70%<br />
of the PT simulations succeed, <strong>with</strong> 51% requiring hardware tuning. Over a quarter<br />
175