RTL Design / Module 9: Lint · CDC · Timing · Area · QoR
Lesson 9.1

What lint is — and why reviews fail without it

Static lint analyzes RTL without running simulation. It catches structural mistakes — latches, multi-drivers, clock misuse — that tests often miss. Companies gate merges on lint-clean because silicon does not forgive what your testbench never exercised.

Static lint

A rule-based scan of synthesizable RTL that flags patterns known to cause simulation/synthesis mismatch, timing failure, or silicon bugs — before you spend hours in waves.

1

Lint vs simulation — complementary, not competing

Simulation answers "does my testbench see the behavior I expect?" Lint answers "does this RTL structure violate known-good patterns?" A block can pass 100% functional coverage and still ship with an inferred latch, an async reset glitch, or an unsynchronized clock-domain crossing.

Lint vs simulation coverage

Simulation proves behavior on exercised paths. Lint audits structure on every line, every module, every clock domain.

QuestionSimulationStatic lint
Is my FSM reachable?Yes, if TB drives itFlags unreachable states in some tools
Is there a combo loop?May oscillate or X-propReports loop before sim runs
Is reset synchronized?Only if you check release timingFlags async reset on sequential logic
CDC crossing safe?Needs metastability injectionFlags missing synchronizers
2

Why companies gate merges on lint

At scale, lint is a quality firewall. One engineer's "minor waiver" becomes ten thousand violations across a SoC. Merge gates enforce a zero-tolerance policy on error-severity rules and cap waivers with documented justification.

🚧

Early bug cost

A latch found in lint costs minutes. The same latch found post-silicon costs millions.

📋

Review bandwidth

Human reviewers cannot grep 500k lines for CDC. Lint automates the boring, critical checks.

🔗

Downstream tools

Synthesis, STA, and formal all assume lint-clean clock/reset topology.

RTL commit→Lint CI→Sim CI→Synthesis→Signoff
Trap

"We'll clean lint before tapeout" never works. Violations compound, waivers become tribal knowledge, and nobody owns the 4,000 warnings filed under "legacy IP."

3

Common lint rule families

Commercial tools ship thousands of rules. Teams enable subsets per block type (combo datapath vs clocked control). These families appear in almost every signoff deck:

FamilyWhat it catchesExample rule ID
Latch / incomplete assignmentCombo logic without full assignmentINFERRED_LATCH
Multi-driven netsTwo drivers on same wireMULTIPLE_DRIVER
Clock rulesGated clocks, clock as data, mixed edgesCLK_AS_DATA
Reset rulesAsync reset misuse, reset as clockRST_SYNC_RELEASE
CDC / RDCCrossings without synchronizersCDC_NO_SYNC
Combo loopsZero-delay feedback pathsCOMBO_LOOP
Coding styleBlocking in seq, implicit netsBLOCKING_IN_FF
lint_waiver_bad.v
// BAD: pragma silences tool — hardware unchanged
// spyglass disable_block INFERRED_LATCH
always @(*) begin
  if (sel) y = a;
  // y holds when !sel → latch inferred
end
// spyglass enable_block INFERRED_LATCH

// GOOD: fix the RTL, no waiver needed
always @(*) begin
  y = 1'b0;          // default covers all inputs
  if (sel) y = a;
end
4

Where lint runs in the flow

Modern flows run lint at three checkpoints: per-commit CI (fast rules), nightly full-chip (CDC/RDC), and pre-tapeout signoff (waived-only-with-VP-approval).

  • Block lint: designer runs locally before opening PR — 5-minute feedback loop
  • SoC lint: integration team runs on stitched netlist — catches cross-module CDC
  • Equivalence: lint on post-synth netlist confirms no clock-gating surprises

Module 9 scope: Lessons 9.1–9.4 cover lint families and reading reports. Lessons 9.5–9.6 introduce CDC mindset and synchronizer structures — with a deep dive in the CDC course.

5

Starter lint mindset for designers

lint_clean_combo.v
// Patterns that pass most lint decks on first run
module lint_clean_combo (
  input  wire       clk,
  input  wire       rst_n,
  input  wire [1:0] sel,
  input  wire [7:0] a, b,
  output reg  [7:0] y
);
  // Rule: one driver per signal — use reg + always, not assign + always
  always @(*) begin
    y = 8'h00;                    // default → no latch
    case (sel)
      2'b00: y = a + b;
      2'b01: y = a - b;
      2'b10: y = a & b;
      default: y = a | b;
    endcase
  end

  // Rule: sequential logic — single clock, sync reset
  reg [7:0] hold;
  always @(posedge clk) begin
    if (!rst_n) hold <= 8'h00;
    else        hold <= y;
  end
endmodule
  • Run lint before requesting code review — not after reviewer comments
  • Treat ERROR severity as a compile failure
  • Never waive without understanding root cause
  • Keep a team waiver database with owner, expiry, and fix plan
  • Match lint rules to synthesis constraints — same clock definitions
▶ Practice lint-safe RTL in simulator
6

Tool landscape — same rules, different UIs

Whether your team uses Synopsys SpyGlass, Cadence JasperGold CDC, Siemens Questa Lint, or open-source Verilator with -Wall, the underlying checks overlap heavily. Learn the rule families (lesson 9.1 table) and you can navigate any report.

ToolTypical useDesigner entry point
SpyGlass / VC LintBlock + SoC signoffsg_shell or CI dashboard
Verilator lintFast pre-commit, open sourceverilator --lint-only -Wall
Questa RDCReset-domain crossingPaired with CDC deck
Formal CDCProve sync structuresIntegration / signoff team
🏃

Fast loop

Run lightweight lint locally before every push — catch latches in seconds.

🌙

Nightly full

SoC-level CDC/RDC runs on stitched RTL — catches cross-module issues.

✅

Signoff

Zero ERROR + audited waivers = tapeout gate.

7

What reviews miss without lint

Human review excels at architecture and protocol correctness. It is poor at exhaustively checking every always block for latch inference across 200 files. Typical post-silicon surprises that lint would have caught:

  1. Incomplete case in a decode block used only in test mode — never stimulated in sim
  2. Clock used as enable on a status flop — passes fast sim, fails STA
  3. Multi-driver bus from overlapping assign and always blocks
  4. CDC pulse stretched wrong — works at room temp, fails at cold
Trap

Arguing "my simulation passed" in a lint review is a red flag. Simulation coverage and structural correctness are orthogonal axes.

verilator_lint_quick.sh
# Fast open-source lint before every commit
verilator --lint-only -Wall --timing \
  -Wno-DECLFILENAME \
  rtl/my_block.v 2>&1 | tee lint.log

grep -E "Error|%Error" lint.log && exit 1
echo "Lint clean — safe to open PR"
review_checklist.md
# Pre-review lint checklist (designer)
[ ] Block lint ERROR count = 0
[ ] No new waivers without justification ticket
[ ] Clock names match SDC / block diagram
[ ] All combo blocks use always @(*) with defaults
[ ] CDC outputs documented (even if "no crossings")
[ ] Lint report attached to PR description

🎯 What to remember

  • Static lint scans RTL structure; simulation exercises behavior — you need both.
  • Merge gates exist because lint violations scale badly and silicon is unforgiving.
  • Rule families: latches, multi-drive, clocks, resets, CDC, combo loops, coding style.
  • Fix RTL first; waivers are audited exceptions, not a workflow.
  • Next: the three structural bugs lint catches most often — latches, loops, and multi-drivers.
Advertisement