ROBOTECA FREE SAMPLE
Dashboard

A free lesson from Digital & Chip Design (HDL / VLSI): the whole module, nothing cut short.

LESSON · VLSI

The chip design flow (RTL → GDSII)

Turn 1 30 min LESSON

ALearning Material

Designing a chip is a pipeline that turns behaviour into a manufacturable layout. Knowing the stages (and their jargon) is essential VLSI literacy.

Designing a chip is a pipeline that turns behaviour into a manufacturable layout, and knowing the stages and their jargon is basic VLSI literacy. It runs from specification, to RTL design in HDL, to functional verification, to logic synthesis (RTL becomes a gate-level netlist of standard cells), to physical design (place and route), to sign-off checks, and finally to GDSII, the layout file sent to the foundry at tape-out.

‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍Two themes dominate. First, verification is often the majority of the work, because an ASIC bug can't be patched: finding it after fabrication means a costly, months-long re-spin, which is also the crux of the ASIC-versus-FPGA choice: an FPGA configures your RTL onto a reprogrammable chip (fast, cheap to iterate, lower performance), while an ASIC manufactures custom silicon (huge upfront cost, best performance at volume). Second, timing dominates sign-off: a synchronous chip only works if every signal meets setup and hold everywhere, which Static Timing Analysis checks exhaustively.

The flow, top to bottom:

  1. Specification: what the chip must do (function, speed, power, area).
  2. RTL design: describe behaviour in HDL (Verilog/VHDL) at the Register-Transfer Level: registers + the logic between them. (This is what you learned in the HDL topic.)
  3. Functional verification: simulate the RTL against testbenches; prove it's logically correct before spending effort on physical design. Verification is often the majority of the work on real chips.
  4. ‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍Logic synthesis: a tool converts RTL into a gate-level netlist of standard cells (predefined AND/OR/flip-flop layouts) from a technology library, optimising for timing/area/power.
  5. Physical design (place & route): arrange the cells on the die (placement) and connect them with metal wiring (routing), respecting design rules.
  6. Sign-off checks: - STA (Static Timing Analysis): verifies the design meets timing at the target clock without exhaustive simulation (checks setup/hold across all paths). - DRC (Design Rule Check): geometry obeys the foundry's manufacturing rules. - LVS (Layout vs Schematic): the layout matches the intended netlist.
  7. GDSII: the final layout file sent to the foundry ("tape-out") to make masks and fabricate the chip.

ASIC vs FPGA: two destinations for RTL: - FPGA: your RTL is configured onto a reprogrammable chip. Fast to iterate, cheaper upfront, lower performance/efficiency. Great for prototyping/low volume. - ‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍ASIC: your RTL is manufactured into custom silicon. Huge upfront cost (masks, months), but best performance/power/cost at high volume. A mistake means a costly re-spin, hence heavy verification.

Why timing dominates. A synchronous chip works only if every signal arrives at its flip-flop within one clock period (setup) and not too early (hold). STA checks this everywhere. Missing timing = the chip won't run at its rated speed.

Why it exists. Turning chip behaviour into a manufacturable layout is a long pipeline, and a single mistake taped out to a foundry costs months and a fortune. Knowing the flow (RTL, verification, synthesis, place-and-route, signoff, GDSII) and its jargon is essential VLSI literacy and the map behind the OpenLane capstone.

Mental model. The flow is like publishing a book: spec = outline, RTL = manuscript, verification = editing/proofreading, synthesis = typesetting into standard fonts/ blocks, place & route = laying out pages and the index, sign-off = final print check, GDSII = the printing plates sent to the press. Once printed (taped out), fixing a typo means a whole new print run, so you proofread obsessively.

‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍Common misunderstandings.

  • "Synthesis or layout is the bulk of the work." Verification often dominates. You prove the RTL correct before physical design, because a fabricated bug means a costly re-spin.
  • "FPGA and ASIC are the same target." FPGA configures reprogrammable hardware (fast, cheap to iterate); ASIC manufactures custom silicon (huge upfront cost, best performance at volume).
  • "If it is logically correct it will run at speed." Timing dominates. STA must confirm every path meets setup/hold at the target clock, or the chip will not hit its rated frequency.

Connections. RTL is exactly the HDL you wrote (HDL topic), and STA/setup-hold is the timing lesson (HDL Turn 2) at signoff. This is the flow the OpenLane RTL-to-GDSII capstone walks end to end, and verify-before-committing mirrors simulate-before-you-cut in CNC and predict-then-measure in electronics.

BImmediate Active Recall

‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍QUERY

Put these in order: place & route, RTL design, GDSII/tape-out, logic synthesis, verification.

REVEAL
ANSWER

RTL design → verification → logic synthesis → place & route → GDSII (tape-out).

Did you recall it?
QUERY

What does logic synthesis produce, and from what?

REVEAL
ANSWER

It converts RTL (HDL) into a gate-level netlist of standard cells from a technology library, optimised for timing/area/power.

Did you recall it?
QUERY

What do STA, DRC, and LVS each check at sign-off?

REVEAL
ANSWER

STA = timing (setup/hold met at target clock); DRC = layout obeys foundry geometry rules; LVS = layout matches the intended schematic/netlist.

Did you recall it?
QUERY

Key trade-off between FPGA and ASIC?

‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍REVEAL
ANSWER

FPGA = reprogrammable, fast/cheap to iterate, lower performance/efficiency (prototyping/low volume). ASIC = custom-manufactured, high upfront cost and long lead time, best performance/power/cost at high volume.

Did you recall it?

CConceptual Questions

Answer each in your own words in the box, then reveal the model answer to compare. These ask why, not how, and your answers are saved.

PROMPT

Why is functional verification often the majority of the work on a real chip?

‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍REVEAL MODEL ANSWER
MODEL ANSWER

Because an ASIC mistake can't be patched in software. Fixing a bug found after fabrication requires a new set of masks and a months-long, very expensive re-spin. So you must prove the RTL logically correct before committing to physical design and manufacture, and proving it exhaustively against many testbenches is a huge effort. The enormous cost of a post-silicon bug is what front-loads so much work into verification.

Compared to the model answer - did you get it?
PROMPT

For the same RTL, what's the core trade-off between targeting an ASIC and an FPGA?

‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍REVEAL MODEL ANSWER
MODEL ANSWER

An FPGA configures your RTL onto a reprogrammable chip: fast iteration, low upfront cost, and you can fix bugs by reloading it, but with lower performance and power efficiency, ideal for prototyping and low volume. An ASIC manufactures your RTL into custom silicon: huge upfront mask cost and months of lead time, but the best performance, power, and per-unit cost at high volume, with a bug meaning a costly re-spin, hence the heavy verification.

Compared to the model answer - did you get it?
PROMPT

What do the three sign-off checks STA, DRC, and LVS each verify?

‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍REVEAL MODEL ANSWER
MODEL ANSWER

STA (Static Timing Analysis) verifies the design meets setup and hold timing on every path at the target clock, mathematically rather than by exhaustive simulation. DRC (Design Rule Check) verifies the layout geometry obeys the foundry's manufacturing rules (spacing, widths, enclosures). LVS (Layout Versus Schematic) verifies the drawn layout actually matches the intended netlist. Together: timing, manufacturability, and correctness.

Compared to the model answer - did you get it?

DPractice Problems

P1 (easy). Which stage turns your Verilog into specific AND/OR/flip-flop cells?

P2 (medium). Why is functional verification done ‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍before synthesis and physical design, not after tape-out?

P3 (harder). A startup needs to iterate a design weekly during R&D, then ship millions of units. Which target for R&D, which for production, and why?

Solutionsclick to reveal

P1. Logic synthesis (maps RTL to a gate-level netlist of standard cells).

P1Compared to this solution - did you get it right?

P2. Because errors get exponentially more expensive downstream: catching a logic bug in RTL simulation costs minutes; catching it after tape-out costs a full re-spin (months and large sums). Verify early to avoid fabricating a flawed chip.

P2Compared to this solution - did you get it right?

P3. FPGA for R&D. Reprogrammable, so weekly iteration is fast and cheap with no fabrication. ASIC for production: the high upfront mask cost is amortised over millions of units, giving the best per-unit cost, power, and performance once the design is stable.

‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍P3Compared to this solution - did you get it right?

EFeynman Exercise

Explain the RTL-to-GDSII flow to a beginner using the book-publishing analogy. Match each stage (spec, RTL, verification, synthesis, place & route, sign-off, GDSII) to a publishing step. Then explain why "tape-out is like sending plates to the press". Why fixing a mistake afterward is so painful, and what that means for how much you verify.

REVEAL MODEL ANSWER
MODEL ANSWER

‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍Making a chip is like turning a blueprint into a printing plate for a photocopier that prints in silicon. You write what it should do, prove on paper it's right (because once the plate is etched you can't scribble corrections), then a tool translates it into a wiring diagram of pre-made parts, arranges and connects those parts on the die, runs a battery of checks, and finally hands the foundry the final artwork to make the masks. The 'prove it on paper first' step is huge precisely because re-etching the plate costs a fortune and months.

Compared to the model answer - did you get it?

FError Analysis Framework

  • Skipping/under-verifying. Why: eager to tape out. Recognise: bugs found in silicon. Avoid: invest heavily in verification before synthesis.
  • Confusing synthesis with place & route. ‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍Why: both "tooling." Recognise: mixing logical vs physical stages. Avoid: synthesis = RTL→gates; P&R = gates→layout.
  • Ignoring timing (STA). Why: focusing on function only. Recognise: chip fails at target clock. Avoid: meet setup/hold via STA sign-off.
  • Wrong target (ASIC for prototyping). Why: not weighing iteration cost. Recognise: expensive, slow respins. Avoid: FPGA to iterate, ASIC for volume.

GMini Challenge

For the same Verilog RTL, when would you target an FPGA versus an ASIC, and why is heavy verification especially critical for the ASIC path?

REVEAL MODEL ANSWER
MODEL ANSWER

‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍Target an FPGA for prototyping or low volume: fast iteration, low upfront cost, and you can reprogram it when you find a bug, accepting lower performance/power efficiency. Target an ASIC for high volume where best performance, power, and per-unit cost matter, accepting huge upfront mask cost and months of lead time. Verification is critical for the ASIC because a bug found after fabrication can't be patched (it forces a costly, months-long re-spin with new masks) so you must prove correctness exhaustively before tape-out.

Compared to the model answer - did you get it?

Quiz Check

A quick auto-graded check, separate from the recall cards above. Your score is pooled with the recall cards into this module's Mastery score, and completing this lesson requires the quiz submitted with pooled mastery at 80% or above.

‍​‌‌​​‌‌​​‌‌‌​​‌​​‌‌​​‌​‌​‌‌​​‌​‌​​‌​‌‌​‌​‌‌‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​‌​‌‍QUIZAuto-graded check · feeds your mastery score
  1. On real chips, the majority of effort is often:

  2. FPGA versus ASIC: an FPGA:

  3. Even a logically correct chip will not run at speed unless:

  4. The final layout file sent to the foundry (tape-out) is:

This is a free sample

Progress and the spaced-repetition reviews are part of the course. The full track continues from here.