A dashboard with forty metrics is wallpaper: nobody reads it, so nobody notices when one turns. The integrator needs one headline metric per discipline that is hard to make green while the chip is broken, plus the rule that keeps each honest. On NanoSoC the numbers are small and checkable, which makes this lesson the reference for what each discipline owes the dashboard.
Eight disciplines report. Timing reports worst slack across the corner and mode matrix with all exceptions reviewed: green means no negative slack and no unreviewed exception, because an unreviewed false path is a wrong timing exception waiting to happen. Power reports the intent checks clean: all 114 isolation cells present and enabled in the right order, retention lists complete, in a power-aware simulation that actually switched the domain off. Physical verification reports zero open DRC and LVS violations outside waived cells. Crossings report zero open CDC and RDC violations, structural, with the synchroniser inventory matching the ten crossing groups of NanoSoC. Verification reports functional coverage of the top-level plan with the boot, interrupt, DMA and power-state tests passing on the frozen baseline. Connectivity reports the master-slave reachability matrix and the tie-off audit clean, including the check that the deliberate DMA alias is documented and that no header addresses an alias image. Equivalence reports every netlist revision proven equivalent to the RTL, the last ECO included. Gate-level simulation reports the boot, the DMA self-test and the power-state walk passing on the current netlist.
No metric here is ungameable: the last column of the table shows the usual way each is made to look green when it is not, and the rule in the third column is what keeps it honest.
Discipline | The metric | Green means | Dishonest green looks like |
|---|---|---|---|
Timing | Worst slack, exceptions reviewed | No negative slack, every exception evidenced | A false path covering a real path |
Power | Intent checks plus power-aware sim | 114 clamps present, order proven in sim | Static checks only, domain never switched in sim |
Physical | Open DRC and LVS count | Zero outside waived cells | Waived cells nobody re-checked |
Crossings | Open CDC and RDC count | Zero, inventory matches the ten groups | Synchronisers the checker cannot see, unlisted |
Verification | Top-level coverage plus must-pass tests | Boot, IRQ, DMA, power walk green on baseline | Coverage on last month's RTL |
Connectivity | Reachability plus tie-offs | Reachability matrix and tie-off audit clean; the 16-image DMA alias documented and no driver or header uses an image other than the first | Tests that only touch the alias images, or a matrix that skips the DMA window |
Equivalence | Netlist revisions proven against RTL | Every netlist revision proven, last ECO included | Equivalence run on the previous netlist |
Gate-level sim | Boot, DMA self-test, power walk on the netlist | All three pass on the current netlist | Run on an older netlist, or with the power-down step skipped |
A signoff dashboard compares measured results with budgets. The NanoSoC teaching budgets are 82.74 mW in RUN, a 0.655 mm2 core, interrupt latency of 135.4 to 182.3 ns for the irq_dma path, hardware reset in 53.948 us and a 12.581 ms boot with the SPI load owning 86.9 percent of it. These are teaching estimates, not measurements, and the area is a floorplan input rather than a signoff outcome. A signoff report that exceeds a budget goes red until a waiver or an approved budget change moves it.
Each discipline shows its one metric with the evidence link: the result row in the manifest, on the frozen baseline.
Every amber names its owner and its date to green; every green built on waivers names the waiver count.
No metric is green on a stale baseline: the dashboard carries the baseline tuple on its face.
Pick one green per week and reproduce it from the baseline: rerun the fastest check in that discipline and compare. Reproduced greens build trust in the rest; a green that cannot be reproduced demotes the whole dashboard to amber until the process is fixed. Start with connectivity, because the map checker runs in milliseconds.
Red, or amber at best. The metric requires the switching simulation; static intent checks without it are the dishonest green in the table.