What Does Vivado Do? Synthesis to Bitstream, Step by Step

Vivado reads your Verilog and your constraints (an XDC file), synthesizes the RTL into a netlist of LUTs, flip-flops and other primitives, optimizes it, places each primitive on a site in the chip, routes the wires between them, reports timing and resource use, and writes a bitstream that configures the FPGA. You can drive it through the project GUI or a short Tcl script in batch mode; the script below built every Basys 3 and Arty A7 project on this site.

What you'll learn

  • Each step from RTL to bitstream
  • Project mode and non-project (Tcl) mode
  • Which reports to read, and what WNS means
  • How to rebuild a design from a script
The Vivado flowSeven steps in order: read the Verilog and XDC, synthesize, optimize, place, route, write reports, and write the bitstream.Verilog and XDCread_verilog, read_xdcsynth_designRTL to netlistopt_designclean up logicplace_designchoose locationsroute_designconnect the wiresReportstiming, utilizationwrite_bitstreamruns DRC too
The non-project flow: each box is one Tcl command.

The script

build.tcl
# Vivado non-project build of one design: synth, place, route, timing and
# utilization reports, and a bitstream (which also runs the DRCs).
#   vivado -mode batch -source build.tcl -tclargs <top> <part> <xdc> <out_dir> <verilog files...>
set top  [lindex $argv 0]
set part [lindex $argv 1]
set xdc  [lindex $argv 2]
set out  [lindex $argv 3]
file mkdir $out
foreach f [lrange $argv 4 end] { read_verilog $f }
read_xdc $xdc
synth_design -top $top -part $part
opt_design
place_design
route_design
report_timing_summary -file $out/timing.rpt
report_utilization -file $out/util.rpt
write_bitstream -force $out/$top.bit

Run it as vivado -mode batch -source build.tcl -tclargs top part xdc out_dir files. The project's build_all.sh calls it once per design and pulls the worst slack and the resource counts from the reports.

What each step does

StepDoesWatch for
synth_designMaps RTL to LUTs, flip-flops, block RAM and DSP; infers memories and arithmeticLatch warnings, removed logic, unconnected ports
opt_designRemoves redundant logic and merges constantsLogic you needed being removed because nothing reads it
place_designChooses a site for every primitiveCongestion on large designs
route_designConnects every net through the routingUnroutable nets, timing getting worse
report_timing_summaryChecks every path against the clock constraintsNegative WNS or TNS
write_bitstreamRuns design rule checks, then writes the .bit fileUnconstrained I/O, missing IOSTANDARD

Results for the beginner projects

DesignBoardWNS at 100 MHzLUTsFlip-flops
BlinkBasys 35.949 ns727
Seven-segment counterBasys 35.496 ns4262
UART echoBasys 35.753 ns7380
VGA color barsBasys 36.232 ns3646
BlinkArty A76.056 ns727
Counter on LEDsArty A75.202 ns2339
UART echoArty A76.058 ns7376

Every design met timing with more than 5 ns to spare out of the 10 ns period. The blink design's 27 flip-flops are its 26-bit counter, which counts to 49,999,999, plus the LED output.

Project mode vs non-project mode

Project mode (the GUI) keeps runs, reports and the block design in a project directory and is easiest for IP and block designs such as a Zynq system. Non-project mode keeps everything in memory and in your script, so builds are reproducible and fit version control. Many teams use the GUI to explore and a script to build.

Common mistakes

  • Ignoring synthesis warnings; most bugs show up there first.
  • Reading the timing report after synthesis only. Routing adds delay, so check it after route_design.
  • Leaving a port without a pin or IOSTANDARD; write_bitstream stops with a DRC error.
  • Building only in the GUI, so nobody can reproduce the result.

Check your understanding

  • Q1. Which step chooses where each LUT goes? Answer: place_design.
  • Q2. What does a positive WNS mean? Answer: the slowest path still meets the clock with that much time to spare.
  • Q3. Why does the blink design need 26 counter bits at 100 MHz? Answer: it counts to 49,999,999, and 2^26 = 67,108,864 is the smallest power of two above 50 million.

The FPGA series

Related

Sources

Last verified

All figures are from Vivado 2025.2 runs of scripts/fpga/build_all.sh on 17 November 2026.