What precheck checks
This page answers: what does each row of precheck’s table check, and what does a green table still not tell you?
Precheck is a Python program, precheck/precheck.py in Tiny Tapeout’s support tools, and the precheck job runs it on the GDS from the tt_submission artifact. Tiny Tapeout’s own guide says of it that “All the checks need to be green to submit to the chip”. The details on this page come from precheck.py at commit 9d13ff3, which is the commit the runner checked out in November 2026, and from runs of it on this course’s layouts and on public projects’ layouts. You’ll see most of these rows fail in Lab 02, and find two mistakes in Lab 03 that none of them catch.
How it runs
Precheck is given one file, tt_submission/<top_module>.gds. It finds info.yaml by looking in the folders above that file, reads top_module, tiles, analog_pins, uses_vapwr and the pinout from it, and picks the DEF template to compare your layout with: tt_analog_1x2.def for a 1x2 analog project, or tt_analog_1x2_3v3.def if uses_vapwr is true. (With analog_pins: 0 it would pick a digital template instead, so an analog layout has to use at least one analog pin.) The LEF and the Verilog are found next to the GDS, under the same name.
Then it runs every check for sky130A, in a fixed order, and doesn’t stop at the first failure. At the end it writes the table you see on the run’s summary page, with a ✅ or a ❌ and a message for each row, and the job fails if any row failed. The reports go into a folder that the job uploads as the precheck_reports artifact.
The fifteen checks
| # | Row | What it checks | Fails like this |
|---|---|---|---|
| 1 | Magic DRC | Magic’s full SKY130 design-rule deck, drc(full), on the top cell. Writes magic_drc.txt, each broken rule with its boxes |
Magic DRC failed; the report names the rule, such as Metal2 spacing < 0.14um (met2.2) |
| 2 | KLayout FEOL | KLayout’s SKY130 rule deck for the layers below metal: wells, diffusion, poly, contacts | Klayout feol failed with 1740 DRC violations (a public project, Nov 2026) |
| 3 | KLayout BEOL | The same deck for the metal and via layers | Klayout beol failed with 1 DRC violations |
| 4 | KLayout offgrid | Shapes whose corners aren’t on the manufacturing grid | |
| 5 | KLayout pin label overlapping drawing | Every pin label sits on drawn metal of its own layer | Klayout pin_label_purposes_overlapping_drawing failed |
| 6 | KLayout zero area | Shapes of zero area | |
| 7 | KLayout Checks | Three things: the GDS’s top cell is named top_module; no forbidden layer (metal 5) is used; the boundary layer exists |
Top macro name mismatch: expected …, got …, or Forbidden layer met5.drawing found in … |
| 8 | Pin check | Every pin of the DEF template is in the LEF and the GDS, on the right layer, with the right size and place; the die is the template’s size; power ports are on metal 4, at least 1.2 µm wide, and reach within 10 µm of the top and bottom edges | Some ports are missing or have wrong dimensions, see 53 LEF errors above, Wrong cell at GDS top-level, or Inconsistent die area between LEF and template DEF |
| 9 | Boundary check | Nothing outside the tile, and a boundary rectangle that covers exactly the tile | Shapes outside project area, Boundary layer doesn't cover project area |
| 10 | Power pin check | VGND and VDPWR are in the Verilog and the LEF, with USE GROUND and USE POWER; VAPWR is there if and only if uses_vapwr is true |
Verilog doesn't contain VAPWR, VGND does not have a corresponding 'USE GROUND ;' |
| 11 | Layer check | Every layer in the GDS is one Tiny Tapeout allows | Invalid layers in GDS: … |
| 12 | Cell name check | No cell name contains # or / |
|
| 13 | urpm/nwell check | A KLayout rule about the spacing of a resistor implant from n-well | Klayout nwell_urpm failed |
| 14 | Analog pin check | Each ua pin touches metal if and only if it’s below analog_pins, and has a pinout description if and only if it touches metal |
four messages, below |
| 15 | Verilog syntax check | <top_module>.v, your src/project.v, parses in Yosys |
Verilog syntax check failed |
Rows 2 to 6 and 13 are KLayout scripts; row 1 is Magic. Precheck builds its own Magic and KLayout for this, at versions fixed in its Nix file, and installs its own copy of the PDK, which is why the course installs the same versions for you (Your tools and the runner’s tools).
The analog pin check has four messages, one for each way the metal, analog_pins and the pinout can disagree. Each says which pin and suggests two fixes:
Analog pinua[2]is connected to some metal butanalog_pinsis set to 2 ininfo.yaml.Analog pinua[0]is connected to some metal but the description ofua[0]in the pinout section ofinfo.yamlis empty.Analog pinua[1]is not connected to any adjacent metal butanalog_pinsis set to 2 ininfo.yaml.… is not connected to any adjacent metal but the description ofua[n]in the pinout section ofinfo.yamlis non-empty.
“Connected” here means metal 4 reaching into a band from 0.1 to 0.5 µm around the pin’s rectangle, or a via 3 on the pin itself. Precheck doesn’t follow the metal any further, so a stub that goes nowhere counts, which is why Lab 01’s placeholder passes.
Reading the table and the log
Because precheck runs every check, one mistake can turn several rows red. In Lab 02’s breakage 5, one line in info.yaml, uses_vapwr: true, makes four rows fail, because it changes the template every later check compares against. The table on the run’s summary page is the quickest place to start: read every ❌ row, and look for the one that explains the others.
The log isn’t always ordered the way the table is. Some checks print their errors as they go and some only fail at the end, so the first ERROR: line in the log can be a consequence of another check’s failure. In Lab 02’s breakage 3, a renamed top_module with an unchanged GDS, the log’s first 53 error lines are the pin check’s, one per pin:
ERROR: Pin clk not found in /foss/designs/projects/ttsky-ic102-a/tt_submission/tt_um_yourname_ic102_v2.lef
while the cause is in row 7 of the table, Top macro name mismatch: expected tt_um_yourname_ic102_v2, got tt_um_yourname_ic102.
A ✅ needs reading as well. In the same breakage, Magic DRC passed. It had been asked to check a cell called tt_um_yourname_ic102_v2, which the GDS doesn’t contain, and its log says what it did instead:
Cell tt_um_yourname_ic102_v2 couldn't be read
No such file or directory
Creating new cell
...
No errors found.
It made an empty cell, checked that, and found nothing wrong, so the table shows a ✅ for Magic DRC, and only the log tells you that the check ran on an empty cell. You’ll write a script in Lab 05 that has to tell a pass like this one from a real pass.
What precheck leaves to you
Precheck checks that a layout is legal and fits the tile. It doesn’t check that it’s the circuit you meant, and several things a reviewer cares about are outside every one of its rows:
- Whether the layout is your circuit. There’s no LVS (layout versus schematic) in precheck, so nothing compares your layout with a netlist. In Lab 03 you’ll make a short between an inverter’s input and output: Magic DRC finds 0 errors, all 15 rows are ✅, and only
make lvsobjects. The course runs LVS with Magic’s extractor and netgen, on your machine, against a reference netlist kept in your repository. - The tie-offs. Tiny Tapeout’s specs say “Connect any unused uo_out, uio_out and uio_oe pins to GND”. Nothing in precheck looks; in Lab 03, with the 24 tie-offs removed, precheck still passes and LVS lists the 24 pins.
- What the pinout says. Precheck checks only that a used analog pin has a description, never what it says, and doesn’t read the digital pins’ descriptions at all. A pinout that calls the input the output passes.
- Whether the circuit works. Precheck simulates nothing. For an analog design, simulation of the extracted layout, with its parasitics, is how you find out, and it happens in the workbench.
- Whether the GDS is the one your sources build. Precheck checks the file you committed. If you changed
wires.tcland forgot to runmake, it checks the old layout, and passes it.
Between them, the docs job and precheck check that your files are complete and legal. Checking that they describe the circuit you meant is left to you and your reviewer, and Pull requests and review is about how you do that.
Check yourself
A layout passes all 15 checks. Name two things that could still be wrong with it.
TipAnswerAny two of: a short or an open that LVS would find; missing tie-offs; a pinout that describes the pins wrongly; a circuit that doesn’t do what it should; a GDS that doesn’t match the committed sources.
Your table has ❌ on “KLayout Checks” and “Pin check”, and the first error in the log is
ERROR: Pin ua[0] not found in …lef. Where do you start?TipAnswerWith row 7, “KLayout Checks”, because it comes first in the table and its message names the cause. If it says
Top macro name mismatch, the pin errors follow from it: the pin check looks for pins on a macro with the expected name and can’t find any.