The checksum experiment compares the original and pipelined RTL with identical parameters, constraints, libraries, tool version, and physical seed. Only then can its timing gain be weighed against area, power, hold, and congestion movement.
Record | Why |
|---|---|
RTL/constraint revision | Identifies functional and timing intent |
Tool/library/RC versions | Prevents environment drift |
Top/defines/parameters | Confirms same elaborated design |
Seed and physical mode | Separates placement noise |
WNS/TNS/area/power/congestion | Shows multi-metric trade-off |
Path/cell deltas | Explains mechanism, not just headline numbers |
Small changes near placement thresholds may show noisy QoR. Repeat representative seeds or use physically guided comparison before claiming an architectural win.
These hypothetical seed-11 values teach the decision process; no implementation run produced them. WNS is the worst required setup view, hold is the worst required hold view, and power uses identical workload/activity assumptions.
Metric / planning limit | C0 | C1 | Delta / interpretation |
|---|---|---|---|
Setup WNS >= 0 ns | -0.18 ns | +0.06 ns | +0.24 ns; nominal setup improvement |
Setup TNS = 0 ns | -1.20 ns | 0 ns | +1.20 ns; check every view |
Hold WNS >= 0 ns | +0.03 ns | -0.02 ns | -0.05 ns; new release blocker |
Cell area <= 11000 um^2 | 10000 | 10800 | +8%; within planning budget |
Dynamic power <= 5.25 mW | 5.00 | 5.40 | +8%; exceeds budget |
Peak routing utilization <= 85% | 72% | 81% | +9 percentage points; investigate hotspot |
The proposed mechanism is a shorter add chain per stage; inspect mapped arithmetic and routed path endpoints to establish it. Seeds 29/47 and the full view matrix remain required. A favorable single seed does not establish robustness, and the adverse hold/power deltas prevent a keep decision even in this illustrative extract.
Nothing about the RTL change yet. Reproduce both versions with the same library, constraints, tool setup, and physical assumptions.
The seed-11 values are hypothetical, to teach the decision process. No implementation run produced them.