Why it matters

Wires have resistance and capacitance that vary with manufacturing, and they vary in opposite directions: a wider wire has lower resistance and higher capacitance to neighbours. Extraction therefore provides several interconnect corners, and the worst one depends on the net.

The four usual corners

1
2
3
4
5
6
7
8
9
10
11

corners = {"cworst": (0.90, 1.20), "cbest": (1.10, 0.80), "rcworst": (1.25, 1.05), "rcbest": (0.80, 0.95)}   # (R scale, C scale)
R0, C0 = 800.0, 120.0              # ohm, fF (illustrative route)
def delay_ps(driver_ohm, name):
    kr, kc = corners[name]
    return 0.69 * (driver_ohm + R0 * kr) * (C0 * kc) / 1000.0    # lumped estimate, fs -> ps
def extremes(driver_ohm):
    d = {n: delay_ps(driver_ohm, n) for n in corners}
    return max(d, key=d.get), min(d, key=d.get)
assert extremes(200.0) == ("rcworst", "rcbest")     # strong driver: wire resistance matters
assert extremes(3000.0) == ("cworst", "cbest")      # weak driver: load capacitance matters
print("PASS: strong driver worst/best", extremes(200.0), "; weak driver worst/best", extremes(3000.0))

With a 200 Ω driver the wire's own resistance exceeds the driver's, so the RC-product corners are the extremes. With a 3000 Ω driver the delay is dominated by charging the capacitance, so the capacitance corners are. The lumped formula is a simplification used only to show the ordering; the extraction tool's results are the real numbers.

What the review looks for

• Setup uses the worst of the capacitance and RC-product corners; hold uses the best of both.

• Neither pair is dropped without evidence that one dominates every net of interest.

• The same extraction version and the same corner definitions feed implementation and signoff.

Related topics

• Operating Conditions

• Next: Temperature Inversion