Chapter 5 Robust Performance Tailoring with Tuning - SSL - MIT

The requirement chosen here is somewhat arbitrary, but it is important to note
that as the requirement changes so does the level of uncertainty that can be tolerated
by each design. To illustrate this point consider the contours shown in Figure 3-12.
The performance requirement is shown on the y-axis, and the maximum level of uncer-
tainty that can be tolerated is plotted along the x-axis. The dark patch in the upper
left sector of the plot, Region 1, represents the design regime that can be accommo-
dated with PT design methods. As the value of the performance variance increases,
indicating more relaxed requirements, a higher level of parametric uncertainty can be
tolerated. For example, if σreq = 250µm, then just over 2.3% uncertainty variation
can be tolerated with PT design techniques. However, if the performance requirement
is tightened to 200µm then only ≈ 1% variation in the uncertainty parameters can
be tolerated.
The RPT design, Region 2 (light patch), covers a greater uncertainty range than
the PT design, significantly opening the design space. At a performance requirement
of 350µm the RPT design can accommodate over 16% variation in the uncertainty
parameters. This value is a significant improvement over the PT design. However,
as the performance requirement becomes more aggressive, the RPT design, like the
PT design, can tolerate much less uncertainty. At a requirement of 200µm, theRPT
design can tolerate ≈ 2.5% variation in the uncertainty parameters, only slightly
better than the PT design at the same level. Although the RPT design methods
do result in designs that are significantly more robust than simple PT, they are not
adequate for high-performance and high-uncertainty designs, as evidenced by the
large unfilled area in the lower right corner of the plot.
3.5 Summary
Robust performance tailoring is a design methodology that is used to produce designs
that are robust to parametric uncertainty. In this chapter the effects of uncertainty
on the PT design are explored motivating the need for robust design techniques.
Three different robust cost functions are reviewed and applied to the SCI development
105