RTL Design / Module 0: Welcome & How to Learn
Lesson 0.3

How real teams review RTL — what “good” looks like

Writing RTL that “works in sim” is only half the job. In companies, another engineer (or a tool) will ask: is this safe to synthesize, maintain, and tape out? This lesson teaches the review lens you will use for every later module — including your I2C capstone.

💡 Review analogy

A restaurant kitchen is not only judged by taste. Health inspectors check cleanliness, process, and labels. RTL review is the health inspection for chips.

1

What seniors scan in the first 60 seconds

  1. Can I read the intent? Clear module name, ports, and comments where non-obvious.
  2. Clock & reset story — which edge? sync/async reset? named consistently?
  3. Combo vs seq separation — messy mixed always blocks are a smell.
  4. Defaults in FSMs / cases — missing defaults → latches or X-propagation.
  5. Is there a testbench? Untested RTL is a draft, not a deliverable.
2

The quality checklist (pin this)

Use this on every lab — including the tiny passthrough from Lesson 0.2. Module 9 will deepen each row.

CategoryReviewer asks…Quick self-test
ReadableCan a teammate explain this in 10 min?Module header + sensible signal names
SynthesizableWill synthesis infer latches or reject constructs?No #delay in RTL; complete if/case in combo
Reset-cleanKnown state after reset?Every flop has reset path; control FSM has RESET state
Clock-cleanOne clock per flop domain?No gated clocks inside RTL without documented reason
Lint-cleanWidth mismatches, multi-drivers, unused logic?Run mental pass: port widths match, one driver per net
CDC-awareCross-clock signals handled?Single-bit sync or FIFO — never raw multi-bit bus
Timing-awareLong combo paths?Flag deep arithmetic; pipeline if needed (Module 8)
VerifiedTests that would fail if broken?At least reset test + one functional scenario
  • Readable — a teammate can explain the block after 10 minutes
  • Synthesizable — no accidental latches, no #delays in RTL, no initial for ASIC logic
  • Reset-clean — known state after reset; no mystery X on control
  • Clock-clean — one clock per flop domain unless CDC is intentional
  • Lint-clean — multi-driven nets, width mismatches, unused signals addressed
  • CDC-aware — multi-bit buses not “just sampled”; sync strategy named
  • Timing-aware — deep combo clouds flagged; ready for Module 9 thinking
  • Verified — directed tests (later: assertions) that would fail if you break it

You will not master every row today. You will start noticing them. That awareness is the difference between tutorial Verilog and job-ready RTL.

3

Bad vs good RTL (copy and compare)

Example A — reset

Bad — no reset
always @(posedge clk) begin
  q <= d;
end
Good — async reset
always @(posedge clk or negedge rst_n) begin
  if (!rst_n) q <= 1'b0;
  else q <= d;
end

Example B — latch

Bad — incomplete if
always @(*) begin
  if (enable) out = in;
end
Good — default
always @(*) begin
  out = 1'b0;
  if (enable) out = in;
end

Example C — FSM

Bad — no default
case (state)
  IDLE: next = START;
  START: next = DATA;
endcase
Good — explicit default
case (state)
  IDLE:  next = START;
  START: next = DATA;
  default: next = IDLE;
endcase
4

What reviewers comment on (by category)

TopicTypical commentYour fix
NamingRename cryptic signalsRole-based names; _n for active-low
ResetsUnknown state at power-upDocument reset strategy in header
FSMsMissing diagram / defaultlocalparam states; default to safe IDLE
CDCRaw bus across clocksFIFO or handshake — Module 9
LintWidth mismatchExplicit casts; match widths
TBHappy path onlyReset + stress tests; note in reply
5

How to write a review reply

  1. Acknowledge — what you changed and where.
  2. Evidence — sim log, wave screenshot, or lint clean.
  3. Defer — only with spec reference if you disagree.

Template: Comment N: [done/deferred] — fix summary — proof

Review posted→ Fix + re-sim→ Reply with proof→ Merge

I2C capstone uses the same sign-off loop.

6

Common “instant reject” bugs

🪤

Inferred latch

Incomplete if/case in combo logic. Looks fine until silicon or FPGA maps a latch.

⏱️

Blocking in sequential

Wrong assignment style → sim/synth mismatch. Module 3 drills this hard.

🔀

Casual CDC

Passing a multi-bit counter into another clock “because it worked once.”

🧪

No failing test

Only happy-path TB. Reviewers ask: what did you try to break?

Mindset: assume your first draft has at least one of these. Hunting it is normal — not failure.

Red flags that fail reviews (no merge until fixed):

  • Multi-bit signal sampled into another clock without documented sync
  • Blocking assignments (=) in clocked always blocks driving flops
  • #delay or initial blocks inside synthesizable RTL modules
  • FSM with unreachable states and no default — silent hang in silicon
  • Testbench that only checks happy path; no reset or back-to-back stress
  • Copy-pasted IP with wrong parameter widths — lint screams, review stops
7

How you will practice reviews here

  • Every major lesson ends with a key takeaway — treat it as review notes.
  • Module 9 is a full quality deep dive (lint / CDC / timing / area / QoR loop).
  • Module 11 I2C capstone requires a filled checklist + short design-review write-up.

🎯 What to remember

  • Good RTL is readable, synthesizable, reset/clock clean, lint/CDC aware, and verified.
  • Reviewers look for intent and risk — not clever one-liners.
  • Start using the checklist on every lab, even tiny ones.
  • Next: the full course map — where each module sits on the Spec → Silicon path.
Advertisement