When two paths with the same data delay have different slack, the difference is in the clock sections. Reading them tells you whether skew, uncertainty or a pessimism credit is responsible, and whether a fix belongs to the clock side.
Line | Question to ask |
|---|---|
Launch clock network delay | Is it close to the capture side? A large difference is skew |
Capture clock network delay | Is the capture flop in a region with systematically later clocks? |
Clock uncertainty | Does the value match the documented budget for this clock and check? |
Clock reconvergence pessimism (credit) | Is the credit present and plausible? A missing credit on close flops is suspicious |
Source latency lines | Present when the clock enters the block with delay |
# the same path with the pessimism credit shown (illustrative)
clock network delay (propagated) 0.35 2.35
clock reconvergence pessimism 0.04 2.39
clock uncertainty -0.10 2.29
library setup time -0.10 2.19
data required time 2.19arrival = 1.20
required_no_credit = 2.00 + 0.35 - 0.10 - 0.10
required_with_credit = required_no_credit + 0.04
assert abs(required_no_credit - 2.15) < 1e-9 and abs(required_with_credit - 2.19) < 1e-9
slack_no, slack_with = required_no_credit - arrival, required_with_credit - arrival
assert abs(slack_no - 0.95) < 1e-9 and abs(slack_with - 0.99) < 1e-9
print("PASS: the pessimism credit adds", round(slack_with - slack_no, 2), "ns of slack")The credit line adds 0.04 ns to the required time, changing the slack from 0.95 to 0.99 ns. On a violating path, that credit can be the difference between passing and failing, which is why comparisons must be made with the same setting. For a path near zero slack, checking the credit is part of the standard read.
Capture and launch latencies in the report match the clock tree plan; a surprising value points to a clock tree problem.
Uncertainty matches the constraint; a different number means a different constraint was applied.
Credit lines are present in signoff reports and absent only where launch and capture share no path.
At the clock sections: skew between launch and capture, uncertainty and the pessimism credit. The 0.10 ns shortfall may be a clock-side effect rather than data delay.