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.
If the hardware meets the requirement then no tuning is necessary and it is considered<br />
nominally successful. However, if σHW is not <strong>with</strong>in the requirement, then the hard-<br />
ware simulation is tuned assuming that the exact uncertainty parameter values, �pMC,<br />
are known, a procedure referred to as baseline tuning in <strong>Chapter</strong> 4. In reality, the<br />
tuning would take place through a model updating technique such as isoperformance<br />
tuning (see <strong>Chapter</strong> 4), however for the purposes of this study it is only necessary<br />
to determine if it is possible to tune the hardware below requirement, so baseline<br />
tuning is adequate. The results of the simulations include the nominal hardware<br />
performances and the tuned hardware performances for each of the designs.<br />
Data: PT, RPT, RPTT designs, uncertainty model, nsims<br />
Result: nominal HW performance, tuned HW performance<br />
begin<br />
for i = to nsims do<br />
Randomly choose �pMC from uncertainty model<br />
for PT, RPT and RPTT designs do<br />
Generate HW simulation by applying �pMC to model<br />
Evaluate nominal HW performance, σHW<br />
if σHW >σreq then<br />
Tune HW simulation <strong>with</strong> knowledge of �pMC → baseline tuning<br />
Evaluate tuned performance, σtune, andstore<br />
end<br />
end<br />
end<br />
end<br />
Figure 5-6: Hardware simulation algorithm.<br />
The PT, RPT (AO) and RPTT designs generated based on an uncertainty model<br />
<strong>with</strong> ∆ = 10%, α = 0.0 andσreq = 220µm are used to generate 200 hardware<br />
simulations. A map of the uncertainty space explored is plotted in Figure 5-7. The<br />
nominal uncertainty values are marked <strong>with</strong> the large dot in the middle of the grid<br />
and the bounds are denoted by the box around the grid. Each of the realizations<br />
used to generate the hardware simulations is marked <strong>with</strong> a dot. Note that although<br />
the 200 simulations aren’t quite enough to fill the space, they do an adequate job of<br />
sampling it.<br />
The results of the hardware simulation are presented in Figure 5-8. The upper plot,<br />
170