VLSI Design Flow: 13 Steps from Specification to Tapeout, Explained

The VLSI design flow turns a chip idea into a GDSII layout in 13 steps, from specification and RTL to clock tree synthesis, routing, timing signoff, DRC/LVS and tapeout. Each step explained, with a setup and hold slack example worked with real numbers and the job each step leads to.

VLSI Design Flow: 13 Steps from Specification to Tapeout, Explained

In short

  • A digital chip goes through 13 steps: 6 front-end steps (specification, RTL design, functional verification, logic synthesis, DFT insertion, equivalence check), 6 back-end steps (floorplan, placement, clock tree synthesis, routing, extraction with static timing analysis, physical verification) and finally signoff and tapeout.
  • The front end decides what the chip does and ends with a gate-level netlist. The back end decides how it is built and ends with a GDSII or OASIS layout file that goes to the foundry.
  • Timing is checked again and again along the way. In Skill-Lync's worked example below, a path at 1 GHz has +230 ps of setup slack and +30 ps of hold slack. Adding 60 ps more clock skew turns the hold slack into a −30 ps violation.
  • Each step has its own job title: RTL design, design verification, DFT, physical design, STA and physical verification engineer.

The VLSI design flow is the sequence of steps that turns a chip idea into a layout file a foundry can manufacture. It starts with a written specification. Engineers describe the logic in a hardware description language, verify it, synthesise it into logic gates and add test logic. They then place and connect millions of those gates on silicon, close timing, check the layout against the foundry's rules and hand over a GDSII file. Handing over that file is called tapeout.

This guide walks through all 13 steps in order and explains each one in plain words. It includes a setup and hold timing example worked with real numbers, which shows why clock tree synthesis matters. It also shows which job each step leads to. Skill-Lync computed the numbers with a short Python script and drew the figures from that same script.

What are the steps in the VLSI design flow?

The VLSI design flow has two halves. The front end turns a specification into a verified gate-level netlist: specification, RTL design, functional verification, logic synthesis, DFT insertion and equivalence checking. The back end, also called physical design, turns that netlist into a manufacturable layout: floorplanning, placement, clock tree synthesis, routing, timing signoff and physical verification. The flow ends with tapeout.

Block diagram of the VLSI design flow: six front-end steps from specification to equivalence check in blue, six back-end steps from floorplan to physical verification in amber, ending in signoff and tapeout
The 13 steps of a digital VLSI flow. The front end hands the back end a netlist plus timing constraints. Real projects loop back whenever a later check fails.

The diagram is drawn as a straight line, but real projects loop. If timing fails after routing, the team may go back to placement, change synthesis settings or even rewrite part of the RTL. The flow below is for a digital chip (an ASIC or the digital part of a system-on-chip). Analog blocks such as data converters and PLLs follow a different, more manual flow of schematic, simulation and hand-drawn layout. They join the digital flow as fixed blocks called macros.

Front end: from specification to netlist

1. Specification and architecture

The specification says what the chip must do and within what limits: its functions, interfaces, clock frequency, power budget, die area and cost. Architects then split the chip into blocks such as a processor core, memories, a DMA engine and I/O controllers, and decide how they talk to each other. Mistakes here are the most expensive kind, because every later step builds on them.

2. RTL design

Designers describe each block at register transfer level (RTL): what is stored in registers and how it changes on every clock edge. RTL is usually written in Verilog, SystemVerilog or VHDL. The building blocks are the ones you meet in a digital electronics course: finite state machines, counters, multiplexers and adders. Our free Digital Electronics book has a Verilog chapter with synthesisable examples.

3. Functional verification

Verification engineers prove the RTL does what the specification says before anything is built. They write testbenches, usually in SystemVerilog with the UVM methodology, and run millions of simulation cycles while checkers compare outputs against a reference model. Coverage reports show which features have not been exercised yet. Formal tools add mathematical proofs for properties that simulation cannot cover exhaustively. For how simulation works at the transistor level too, see our guide to circuit simulation in VLSI design.

4. Logic synthesis

A synthesis tool translates the RTL into a gate-level netlist. That is a list of standard cells, such as NAND gates, flip-flops and multiplexers from the foundry's cell library, plus the wires between them. The designer also supplies timing constraints in a constraints file (SDC): the clock period, input and output delays, and false paths. The tool picks cell types and sizes that meet those constraints at the smallest area and power.

5. DFT insertion

Design for test (DFT) adds logic that lets a factory tester check every manufactured chip quickly. The main technique is the scan chain. The flip-flops are stitched into long shift registers so test patterns can be shifted in, one clock applied and the results shifted out. ATPG (automatic test pattern generation) tools then create patterns that cover as many possible faults as they can. Memories get built-in self-test (BIST) circuits. DFT is the link between the design team and the test floor of an assembly and test (OSAT) plant.

6. Equivalence check

A logic equivalence check (LEC) proves mathematically that the synthesised, scan-inserted netlist still behaves exactly like the verified RTL. It is repeated after later netlist changes, because every optimisation step could, in principle, change the logic.

Back end: physical design from netlist to layout

Physical design takes the netlist and the constraints and produces the geometry: where every cell sits and where every wire runs, layer by layer. Open-source flows show the same sequence. OpenROAD-flow-scripts names its stages synth, floorplan, place, cts, grt (global route), route and final, and generates GDS automatically from RTL.

7. Floorplanning, partitioning and power planning

Floorplanning fixes the die size, places the large blocks (memories, analog macros, I/O pads) and decides where each block's pins go. On big designs the chip is first partitioned into blocks with as few connections between them as possible, so separate teams can work on each block. Power planning is part of this step. It lays out rings and straps of metal for the supply (VDD) and ground (VSS) so every cell gets a stable voltage without a large IR drop.

8. Placement

The placement tool puts every standard cell into the rows of the floorplan. It keeps connected cells close together to shorten wires, while leaving enough room for routing so no area becomes congested. Placement also resizes cells and adds buffers on long nets to fix early timing problems.

9. Clock tree synthesis (CTS)

One clock source has to reach thousands or millions of flip-flops. Clock tree synthesis builds a network of buffers and wires for this. Its job is not to make every wire the same length. Its job is to make the clock arrive at every flip-flop at nearly the same time. The difference in arrival time between two flip-flops is the clock skew, and the time from the clock source to a flip-flop is its insertion delay. CTS keeps skew low while controlling insertion delay, using balanced trees (such as the H-tree drawn on this post's cover), buffer sizing and sometimes deliberate useful skew. The worked example below shows why skew matters.

10. Routing

Routing draws the actual metal wires on a stack of metal layers connected by vias. Global routing first assigns each net to coarse regions to avoid congestion. Detailed routing then draws exact tracks that obey the foundry's spacing and width rules. Critical nets, such as clocks, are routed first, sometimes with extra spacing or shielding.

11. Parasitic extraction and static timing analysis

Once wires exist, an extraction tool calculates their resistance and capacitance (the parasitics). Static timing analysis (STA) then checks every timing path in the design without running a simulation. It does this at several timing corners: combinations of process, voltage and temperature at their extremes. A path passes when its slack is zero or positive. Slack is the difference between when a signal is required and when it actually arrives. The chip has to pass at every corner, not just the typical one.

12. Physical verification: DRC, LVS, antenna and ERC

Design rule checking (DRC) checks that the layout obeys the foundry's manufacturing rules, such as minimum widths, spacings, enclosures and metal densities. The foundry, not the designer, writes these rules. Layout versus schematic (LVS) extracts the circuit from the drawn layout and compares it with the netlist, to catch missing or extra devices, shorts and opens. Antenna checks catch long wires that could damage transistor gates by collecting charge during manufacture. Electrical rule checks (ERC) catch problems such as floating inputs.

13. Signoff and tapeout

When timing, power, DRC and LVS are all clean at every corner, the design is signed off. The layout is exported as a GDSII file, the industry's standard layout format since Calma introduced it in 1978, or its newer, more compact successor OASIS. That file goes to the foundry, which makes the photomasks. The hand-over is still called tapeout because GDSII files were once written to magnetic tape.

After tapeout the work leaves the design team. The foundry fabricates wafers through the masks; each die is probed on the wafer (wafer sort), the good dies are cut out and packaged, usually at an OSAT (outsourced assembly and test) plant, and every packaged chip runs a final test on automatic test equipment using the scan patterns added at the DFT step. The first silicon is then brought up on a board and checked against the specification before volume production.

Worked example: setup and hold slack on one timing path

Static timing analysis comes down to two checks on every path between two flip-flops. The setup check asks whether data launched on one clock edge reaches the next flip-flop early enough before the next edge. The hold check asks whether the new data arrives too soon and corrupts the value the capture flip-flop is still holding. If flip-flop setup and hold times are new to you, see the flip-flop timing section of our Digital Electronics book.

Take a path clocked at 1 GHz, so the period is ps. The launch flip-flop's clock-to-Q delay is ps. The logic between the flip-flops takes at most ps and at least ps. The capture flip-flop needs a setup time ps and a hold time ps. The clock uncertainty (jitter plus margin) is 30 ps for setup and 20 ps for hold. After CTS, the capture clock arrives ps later than the launch clock.

Setup check. Data arrives at ps. It is required by the next capture edge, minus the uncertainty and the setup time:

Timing bar chart: data arrives at 730 ps after the launch edge, the required time is 960 ps, so setup slack is plus 230 ps
Setup check for the example path. The data has 230 ps to spare before the 960 ps deadline.

The path passes with 230 ps to spare. Setting the slack to zero gives the shortest period this path allows: ps, or about 1.30 GHz.

Hold check. On the same edge, the fastest new data arrives at ps. It must not arrive before the capture flip-flop's hold window ends, which is ps:

What skew does. Suppose the clock tree is unbalanced and the capture clock arrives 100 ps late instead of 40 ps. Setup slack improves to +290 ps, because the deadline moves out by 60 ps. Hold slack falls to ps, a violation. A hold violation cannot be fixed by slowing the clock, because the clock period does not appear in the hold check. It makes the chip fail at every frequency. The fix is to add at least 30 ps of delay buffers on the short path, or to rebalance the clock tree.

Line chart of setup slack rising from 190 to 310 ps and hold slack falling from 70 to minus 50 ps as clock skew goes from 0 to 120 ps, with hold failing above 70 ps of skew
For this path, each extra picosecond of skew adds one to setup slack and takes one from hold slack. Hold fails beyond 70 ps.

This is why clock tree synthesis aims for controlled skew, not just short wires. It is also why STA is run at a slow corner for setup and a fast corner for hold. GATE also asks this kind of timing question in simpler form, as in this solved GATE question on the maximum clock frequency of a ripple counter.

Which job does each step of the VLSI flow lead to?

Each block of the flow is a career track in its own right, and Indian chip-design centres hire for all of them. Companies hire on the skills of one track. The table below matches each step to its role. For pay, the latest AmbitionBox ranges we checked (read 14 November 2026) are summarised in our India Semiconductor Mission 2.0 and VLSI jobs guide.

StepRoleWhat the engineer actually does
1Architect / micro-architectSplits the chip into blocks, sets performance, power and area targets
2RTL design engineerWrites and lints the Verilog/SystemVerilog for a block
3Design verification engineerBuilds UVM testbenches, closes functional coverage
4, 11Synthesis and STA engineerWrites constraints, runs synthesis, closes timing at every corner
5DFT engineerInserts scan and BIST, generates ATPG patterns, debugs test failures
7-10Physical design engineerFloorplan, placement, CTS and routing for a block until it meets timing and power
12Physical verification engineerRuns and fixes DRC, LVS, antenna and ERC
Analog macrosAnalog design and layout engineerDesigns and draws data converters, PLLs, regulators by hand

On AmbitionBox, read 14 November 2026, the average range for physical design engineers with 1-3 years' experience was ₹6.8-7.6 lakh a year. For RTL design engineers it was ₹7.6-8.4 lakh, and for DFT engineers ₹10-11.5 lakh. These are ranges of self-reported averages, not offers, and pay at semiconductor product companies runs well above IT services firms for the same title. For a broader picture, see our VLSI engineer career and salary guide, the VLSI job roles explainer and our list of companies hiring embedded and VLSI engineers.

Where is VLSI design used?

Every modern electronic product runs on VLSI chips: phone processors and modems, memory, car controllers such as anti-lock braking and battery management, medical monitors, Wi-Fi routers and AI accelerators in data centres. India's role is mostly in design for now. Government releases citing industry estimates say India employs nearly 20% of the global chip design workforce. The first Indian wafer fab, at Dholera, is scheduled by the government to be commissioned in 2028. Our guide to VLSI design techniques covers the low-power and high-speed methods those chips use.

How do you start learning VLSI design?

Start with digital logic: Boolean algebra, combinational and sequential circuits, flip-flop timing and CMOS logic. Every VLSI track assumes these. Then learn Verilog, and pick one track, such as verification, physical design or DFT. Learn Linux and Tcl or Python scripting along the way, because every EDA tool is driven by scripts. GATE aspirants can practise the same fundamentals with our solved GATE digital electronics questions.

Skill-Lync's Basics of Digital Electronics course is an 11-week online course with 22 hours of content and 2 projects. It covers number systems, logic gates, Boolean minimisation, combinational and sequential circuits, finite state machines and programmable logic. That is the foundation layer, not a full physical design course. If you are deciding between chip design and embedded software, the embedded route has its own numbers. In Skill-Lync's placement records (snapshot 16 November 2026), embedded engineer freshers were offered 4.2 LPA on average (n = 231), mostly 2021-23 starting offers. You can also see which companies hired them, compare embedded systems courses in India or talk to a Skill-Lync counsellor.

Frequently asked questions

What is the difference between front-end and back-end VLSI design?

Front-end design decides what the chip does. It covers specification, RTL coding, functional verification, synthesis and DFT, and it ends with a verified gate-level netlist. Back-end design, or physical design, decides how the chip is built. It covers floorplanning, placement, clock tree synthesis, routing, timing closure and physical verification, and it ends with a GDSII layout for the foundry.

What is RTL to GDSII?

RTL to GDSII is the name for the whole digital implementation flow. It starts from register transfer level code written in Verilog, SystemVerilog or VHDL and ends with a GDSII layout file. In between come synthesis, DFT insertion, floorplanning, placement, clock tree synthesis, routing, timing signoff and physical verification. Open-source flows such as OpenROAD run the same stages automatically.

What is tapeout in VLSI?

Tapeout is the moment the final, signed-off layout is sent to the foundry for mask making and fabrication. The name dates from the time GDSII files were written to magnetic tape. A tapeout happens only after timing, power, DRC and LVS checks are all clean, because a mistake found after the masks are made costs a full respin.

Why is clock tree synthesis needed?

A single clock source must reach thousands or millions of flip-flops, and unequal arrival times create clock skew. Too much skew in one direction causes setup failures; in the other direction it causes hold failures, which no clock frequency can fix. Clock tree synthesis builds a buffered, balanced network that keeps skew and insertion delay under control.

Is VLSI a good career for ECE students in India?

For ECE and EEE students who enjoy digital logic and are willing to specialise, yes. Many global chip companies run large design centres in Bengaluru, Hyderabad and Noida, and India's semiconductor mission is adding assembly, test and fab jobs. Entry is competitive. Employers expect strong digital fundamentals, one hardware language and hands-on tool or project experience.

Which software is used for VLSI design?

Industry flows mostly use commercial EDA suites from Synopsys, Cadence and Siemens for simulation, synthesis, place and route, STA and physical verification. To learn, students can use free tools: Icarus Verilog or Verilator for RTL simulation, Yosys for synthesis, OpenROAD for place and route, and KLayout or Magic for layout viewing and checks.

Sources and further reading

Want this on your résumé?

Talk to a counsellor.

Talk to a career counsellor

Free 20-minute call. We'll map the right course to where you want to work.

By submitting you agree to our Terms and Privacy Policy.

Talk to a career counsellor

Talk to a career counsellor

Free 20-minute call. We'll map the right course to where you want to work.

By submitting you agree to our Terms and Privacy Policy.