"We had 12 signed-off scenarios and added a low-voltage mode; the overnight run failed 2,400 setup endpoints in exactly one scenario, low-voltage slow cold. The other low-voltage scenario, hot, and the functional cold scenario were clean, so I ruled out the mode's constraints and a general cold problem. I compared the same endpoints cold against hot: every one lost 0.12 ns per nanosecond of path delay, a proportional loss, so I looked at the libraries: at 0.72 volts the cold arcs are 12% slower, temperature inversion. Our corner list assumed slow means hot. The fix was to make the cold scenario permanent, fix the shared segments of the failing paths first, which cover about three quarters of the endpoints, and re-run hold at low voltage hot because the speed-ups erode it. The trade-off is power from threshold swaps; the evidence is the full-matrix summary, per scenario, with endpoint counts before and after."
Step | Sentence |
|---|---|
Assumption | Corner list assumed slow means hot; constraints and tools assumed correct |
Mechanism | Temperature inversion at low voltage: cold arcs 12% slower, loss proportional to path delay |
Trade-off | Threshold swaps and sizing cost leakage and input capacitance; hold must be rechecked |
Evidence | Per-scenario summary, endpoint counts before and after, hold in the fast hot scenario |
words = 65 * 140 // 60 # about 150 words at a 140-words-per-minute pace
script = 152 # words in the narration above (counted with len(text.split()))
assert abs(script - words) < 25
steps_present = {"assumption": True, "mechanism": True, "trade_off": True, "evidence": True}
assert all(steps_present.values())
print("PASS: about", words, "words fit in about a minute; script is", script, "words; all four framework steps present")Whether you narrow before you dig: scenario attributes first, endpoints second.
Whether you can name the discriminating test that separates library effects from constraint effects.
Whether you know what not to do (waive, loosen, drop) and say why.
'The other low-voltage scenario, hot, and the functional cold scenario were clean': it rules out the constraints and a general cold problem.