Two recorded power reports of one multiplier disagree by a factor of five. Find out why, and decide which one goes into the budget.
Hardware context: Activity recorder on a bus, and recorded power reports of a multiplier. Turn waveforms into SAIF totals and judge how much of an estimate rests on recorded activity.
An 8 × 8 multiplier block (1,017 cell pins) gets new operands in every cycle. Its product is used in about one cycle in ten. One simulation of that workload was dumped twice: once from the RTL, once from the synthesized netlist.
Report A read the RTL VCD and says 1.265 mW. Report B read the netlist VCD and says 0.241 mW. Both used the same netlist, corner and clock. Each report came with its annotation summary.
Name the cause of the gap. Give the share of pins that report A annotated from its file, in per cent. Then say which report you would put in the power budget.
Requirements
| Requirement | Limit | Unit |
|---|---|---|
| The cause you named explains the gap | = 1 | count |
| Error of your annotated share for report A | ≤ 0.5 | % |
| The report you would use for the power budget | = 1 | count |
Recorded experiments: Recorded build-time OpenSTA 3.1.0 runs on the opiso0 netlist: sky130_fd_sc_hd, tt, 10 ns clock (probe tools/probes/lowpower/activity-data.mjs, 2026-11-12). Both read a VCD of the same Icarus run of the same workload. No tool runs during grading.
Full specification
Report A (RTL VCD):
| Group | Internal (W) | Switching (W) | Leakage (W) | Total (W) |
|---|---|---|---|---|
| Sequential | 7.832e-5 | 0.000e+0 | 1.312e-10 | 7.832e-5 |
| Combinational | 6.737e-4 | 5.127e-4 | 8.486e-10 | 1.186e-3 |
| Clock | 0.000e+0 | 0.000e+0 | 0.000e+0 | 0.000e+0 |
| Total | 7.520e-4 | 5.127e-4 | 9.798e-10 | 1.265e-3 |
vcd 50
unannotated 967
Report B (netlist VCD):
| Group | Internal (W) | Switching (W) | Leakage (W) | Total (W) |
|---|---|---|---|---|
| Sequential | 6.854e-5 | 0.000e+0 | 1.305e-10 | 6.854e-5 |
| Combinational | 9.199e-5 | 8.000e-5 | 8.443e-10 | 1.720e-4 |
| Clock | 0.000e+0 | 0.000e+0 | 0.000e+0 | 0.000e+0 |
| Total | 1.605e-4 | 8.000e-5 | 9.748e-10 | 2.405e-4 |
vcd 1017
unannotated 0
Recorded build-time OpenSTA 3.1.0 runs on the opiso0 netlist: sky130_fd_sc_hd, tt, 10 ns clock (probe tools/probes/lowpower/activity-data.mjs, 2026-11-12). Both read a VCD of the same Icarus run of the same workload. No tool runs during grading.
The share check compares your number with report A's summary, within 0.5 percentage points.
Examples
Example 1
Blame the Leakage- Input
{ "cause": "leakage", "annotated_percent": 4.9, "use": "B" }- Output
The cause you named explains the gap: 0 Error of your annotated share for report A: 0 % The report you would use for the power budget: 1
Compare the columns that differ, then read both annotation summaries.
Example 2
Blame the Clock- Input
{ "cause": "clock", "annotated_percent": 4.9, "use": "B" }- Output
The cause you named explains the gap: 0 Error of your annotated share for report A: 0 % The report you would use for the power budget: 1
Compare the columns that differ, then read both annotation summaries.
Example 3
Trust the Bigger Number- Input
{ "cause": "workload", "annotated_percent": 100, "use": "A" }- Output
The cause you named explains the gap: 0 Error of your annotated share for report A: 95.1 % The report you would use for the power budget: 0
Compare the columns that differ, then read both annotation summaries.
Submission Rules
- Submit one JSON object with every documented field.
- Run checks your answers against the recorded reports. Submit applies the same requirements and records the result.
- The numbers are estimates for one corner and one workload.
Hints
Hint 1
Compare the reports column by column. Which columns hold the gap, and which are tiny in both?
Hint 2
The Clock row is the same in both. Is there a clock tree in this netlist at all?
Hint 3
Divide the annotated pins by all pins. Then ask where the activity of the other pins came from.
What is checked: Recorded build-time OpenSTA 3.1.0 runs on the opiso0 netlist: sky130_fd_sc_hd, tt, 10 ns clock (probe tools/probes/lowpower/activity-data.mjs, 2026-11-12). Both read a VCD of the same Icarus run of the same workload. No tool runs during grading.
Run and Submit use the same measured requirements. Run is a preview; Submit records an attempt. No physical tool runs during this exercise; the records come from the pinned build-time experiments.
More problems on this topic