A path delay is the sum of its stage delays, each split into a cell part and a net part. The cell part comes from the library model; the net part comes from the wire estimate.
Take an EXAMPLE three-stage path with cell delays 0.17, 0.22 and 0.15 ns and net delays 0.05, 0.04 and 0.06 ns. Cells sum to 0.54 ns, nets to 0.15 ns, and the path to 0.69 ns.
Stage | Cell (ns) | Net (ns) | Stage total (ns) |
|---|---|---|---|
1 | 0.17 | 0.05 | 0.22 |
2 | 0.22 | 0.04 | 0.26 |
3 | 0.15 | 0.06 | 0.21 |
Path | 0.54 | 0.15 | 0.69 |
A table model looks up delay by input transition (slew) and output load. Points between grid entries interpolate, usually bilinearly in both axes.
Take an EXAMPLE table with slew points 0.05 and 0.15 ns and load points 0.01 and 0.05 pF. The four entries are 0.10, 0.18, 0.14 and 0.26 ns.
EXAMPLE table (not real library values), rows slew, columns load
slew 0.05 ns: load 0.01 pF -> 0.10 ns, load 0.05 pF -> 0.18 ns
slew 0.15 ns: load 0.01 pF -> 0.14 ns, load 0.05 pF -> 0.26 ns
query: slew 0.10 ns, load 0.03 pF
slew fraction = (0.10 - 0.05) / (0.15 - 0.05) = 0.05 / 0.10 = 1/2
load fraction = (0.03 - 0.01) / (0.05 - 0.01) = 0.02 / 0.04 = 1/2
delay = (0.10 + 0.18 + 0.14 + 0.26) / 4 = 0.68 / 4 = 0.17 nsA linear model fits delay as a base plus a load coefficient times the load. It is one multiply-add per stage, with no table storage.
It misses the input-slew effect: a slow input edge delays the output even at fixed load. Take EXAMPLE base 0.08 ns with coefficient 2.00 ns per pF at 0.05 pF load.
EXAMPLE parameters (not real library values): base 0.08 ns, coefficient 2.00 ns/pF, load 0.05 pF
linear delay = 0.08 + 2.00 x 0.05 = 0.08 + 0.10 = 0.18 ns
the table query above gives 0.17 ns at the same load with a stated slew
the 0.01 ns gap is the slew the linear form cannot seeBilinear weights are products of 1/2 and 1/2, so each corner weighs 1/4. Equal quarters give the mean, 0.17 ns.
The input slew. The linear form has no slew term, so a sharper or slower edge than assumed moves the true delay.