Quality of Results (QoR) is not a one-shot gate — it is a loop. You write RTL, run checks, read reports, fix issues, and repeat until the block meets the team's bar for function, timing, area, CDC, and power. This lesson ties Module 9 together and points to Module 10 verification.
The iterative workflow every RTL engineer runs: implement → lint → simulate → synthesize → analyze reports → rewrite RTL → repeat until "done enough" for the milestone.
Early loops are fast (lint/sim). Later loops add synthesis. Don't skip straight to PNR — fix RTL issues cheaply first.
| Stage | What you learn | Typical failure → fix |
|---|---|---|
| Lint | Coding violations, latch inference, CDC struct | Complete case; fix async reset; add synchronizer |
| Sim | Functional correctness | FSM bug, off-by-one, handshake stall |
| Synthesis | Timing slack, area, inferred memories | Pipeline combo; ROM vs case; register enable |
| CDC/RDC | Cross-domain protocol | Gray ptr; FIFO; remove bus sync |
| Power | Clock gating coverage | Add CE; operand isolation |
#!/bin/bash
# Minimal QoR loop for a block — adapt to your flow
BLOCK=uart_tx
RTL=rtl/${BLOCK}.v
TB=tb/${BLOCK}_tb.v
echo "=== 1. Lint ==="
verilator --lint-only -Wall $RTL 2>&1 | tee reports/lint.log
grep -i error reports/lint.log && exit 1
echo "=== 2. Simulate ==="
iverilog -g2012 -o sim.vvp $RTL $TB && vvp sim.vvp | tee reports/sim.log
grep -i fail reports/sim.log && exit 1
echo "=== 3. Synthesize (example) ==="
# yosys / dc / genus — read timing + area reports
# grep "slack" reports/timing.rpt
echo "=== 4. Review & rewrite ==="
# Map worst path → RTL line → pipeline or restructure
# Re-run loop
// Lint waiver — must include justification (not "ignore") // Rule: CDC_MULTI_BIT // Signal: status_bus_cross // Owner: jsmith // Reason: Replaced with toggle handshake in rev 2.1 — waiver obsolete, remove. // Expiry: 2026-Q3
Chasing timing in place-and-route when the RTL has a 3-level priority mux in one cycle — you will iterate for weeks. Read the synthesis critical path first; fix RTL structure, then re-synthesize.
# QoR iteration log — template per block milestone | Iter | Lint ERR | Sim | WNS (ns) | Area (kGE) | CDC open | Action taken | |------|----------|--------|----------|------------|----------|---------------------| | 1 | 12 | fail | — | — | 3 | fix latch, add sync | | 2 | 0 | pass | -0.45 | 42.1 | 0 | pipeline arbiter | | 3 | 0 | pass | +0.08 | 43.0 | 0 | ACCEPT — RTL freeze |
Perfection is not the bar at every checkpoint. Know what "done" means for each milestone:
| Milestone | Done enough means |
|---|---|
| First RTL drop | Lint-clean (or waivers filed); sim passes directed tests; CDC protocol named per crossing |
| Feature complete | Self-checking TB; coverage on main modes; synthesis runs without error |
| RTL freeze candidate | Timing met or known exceptions with fix plan; area within budget; CDC signoff started |
| RTL freeze | All waivers approved; no open functional bugs; timing/area signed by owner |
You have covered the full quality stack an RTL author owns before handoff to backend and verification teams:
Catch latches, incomplete case, and coding violations before sim waste.
Gray pointers, FIFOs, handshakes — never tear a bus.
Critical path intuition; pipeline deep combo; read slack reports.
Mux cost, resource sharing, clock enable — not combo gating.
# Module 9 self-check — answer before Module 10 1. Name three lint rule families and one example violation each. 2. Why is 2-FF sync insufficient for a 32-bit binary counter? 3. What RTL structure kills timing? Name two fixes. 4. When does resource sharing save area? When does it fail? 5. Write clock-enable RTL — never combo-gate clk. Why? 6. What does negative setup slack mean? What do you do first?
Module 9 made your RTL worthy of silicon. Module 10 makes you prove it:
The full professional loop: Write RTL (Modules 1–7) → Pipeline for performance (Module 8) → Close QoR (Module 9) → Verify exhaustively (Module 10) → Integrate and tape out. You are now equipped for the verification half.
// QoR loop step 3 evolves into Module 10 rigor
integer errors;
initial errors = 0;
task check_eq(input [31:0] got, expected, input string msg);
if (got !== expected) begin
$display("FAIL %s: got %0d exp %0d", msg, got, expected);
errors++;
end
endtask
// At end of sim:
initial begin
// ... run tests ...
if (errors == 0) $display("=== QoR SIM PASS ===");
else $display("=== %0d FAILURES — fix RTL, re-loop ===", errors);
end
Regression in the loop: After RTL freeze candidate, every QoR iteration runs the full regression suite — not just the one test you fixed. A timing fix that breaks a corner-case test is not progress.