Building Your Own Signoff Report for the UART
In the last two articles we built the checklist and the decision rule. In this article we put them into one document: the signoff report. Its test is simple. Someone with no context on the project should be able to read it and reach your conclusion without asking you a single question.
The report for our UART
Write this for your own run, using the status lines you collected two articles ago. Here it is filled in with the results we got in Modules 7 and 8:
UART (uart_top): Physical Design Signoff Report
Run tag: uart_full
PDK: SkyWater SKY130A, sky130_fd_sc_hd
Clock: clk, 4.00 ns (250 MHz), uncertainty 0.15 ns, derate late 1.05 / early 0.95
1. Setup timing (Module 7.5), derated
Corner Worst slack TNS Status
ss_100C_1v60 +0.07 ns 0.00 PASS
tt_025C_1v80 +1.19 ns 0.00 PASS
ff_n40C_1v95 +1.91 ns 0.00 PASS
2. Hold timing (Module 7.5), derated
Worst slack at ff_n40C_1v95: +0.11 ns PASS
3. DRC (Module 8.1), Magic full deck: 0 violations PASS
4. LVS (Module 8.2), Netgen vs final netlist: match uniquely PASS
5. Antenna (Module 8.3), OpenROAD: 0 violating nets PASS
6. IR drop (Module 8.4): worst 4.1 mV on VPWR (budget 90 mV);
EM: all grid currents within limits PASS
7. Rows and density (Module 8.5): rows complete; no window above
maximum; met2-met5 below minimum in open areas, metal fill
to be generated at tapeout by the shuttle PASS (condition)
8. Independent DRC (Module 9.2), KLayout: 0 violations PASS
9. GDS sanity (Module 9.1): top cell uart_top, cells complete,
Magic/KLayout streamout XOR: no differences PASS
Waivers: none.
Conditions: item 7, metal fill by the shuttle; density to be
re-checked on the filled GDS if the shuttle provides it.
FINAL VERDICT: GO
Signed off by: <your name>, <date>
The hold figure is the worst hold slack at the fast corner after the ECOs; check your own run for the real value. Every row names its source, its tool and a clear status, and the verdict is one line nobody has to interpret.
Item 7 shows something real reports often contain: a pass with a condition. It is not a failure and it is not a waiver. It records that one piece of work, metal fill, happens after your handoff and who is responsible for it. Writing it down means nobody later assumes it was done.
In industry this document, often called a tapeout readiness review or signoff package, is usually checked and countersigned by someone other than the engineer who did the work. The person closest to a project is the one least likely to see its blind spots.
The report is not paperwork for its own sake. If a reader with no context cannot reach your verdict from the document alone, it is not finished, however much real work lies behind it.
Practice Exercise
While finishing the report, you realise you never re-ran DRC after the antenna fix in Module 8.3, which added a diode cell and re-routed a net. Every other item is current. What should the report say about DRC right now, and what should you do before finishing it?
Solution
The report must not say “DRC: PASS” on the strength of a result from before the diode was added. Any change to the geometry after a check ran makes that check’s result stale, however small the change looks. A new cell and a re-routed net are exactly the kind of edit that can create a new spacing or enclosure violation right next to the change.
Re-run Magic’s DRC (and KLayout’s, since item 8 has the same problem) on the current GDS before writing anything about DRC. LVS is stale for the same reason: the layout has a new device, so it must be compared again against a netlist that includes the diode. Only when those runs come back clean can the report say PASS. The report is written last, describing the design as it actually is, never first as a placeholder for results you expect to still be true.
With a signed-off GDS and a report in hand, the next section looks at how open shuttle programs turned files like this into real chips, and what they expected from a submission.