05 esp32c3_min¶
Reference: Espressif "ESP32-C3 Series Hardware Design Guidelines" and the ESP32-C3-WROOM-02 datasheet. From memory of those documents, not re-read: 10 uF + 0.1 uF at the 3V3 pin, EN RC of 10k / 1 uF, IO9 low at reset for download boot, IO8 and IO2 pulled high, supply of at least 500 mA, no copper under or beside the antenna, USB Serial/JTAG on IO18 (D-) / IO19 (D+). Chosen here: AP2112K regulator, USBLC6 ESD part, header pinout, LED on IO7.
Stresses: an RF module with an antenna keep-out, a footprint that carries its own thermal vias, and USB pins on the side facing a header.
What broke
1. KEEPOUTS had never been used. It worked first time: no pour, track or via
in the antenna strip on either layer.
2. The module footprint's thermal vias are 0.2 mm; the board minimum was
0.3 mm.
3. KiCad warns that the module footprint differs from its library copy once
kicad-tools has written it. Pad, line, text and property counts are equal.
Not explained; the check is set to ignore for this board only.
4. Autorouted USB ran 45-50 mm round the outside of the right-hand header with
5.5 mm of skew.
5. My first hand-routed pair crossed itself on the bottom layer (KiCad:
"tracks crossing", "shorting"). I drew two parallel verticals whose
diagonals had to swap sides.
6. The EN capacitor was 14.8 mm from the EN pin.
7. Silkscreen labels collided with reference designators, whose final
positions are chosen by a tool after I place the text.
Tool changes
- RULES can lower a board minimum; SEVERITIES can set a KiCad DRC check's
severity per board (use with a written reason).
- copper.check_tracks: gen now refuses hand tracks of different nets that
cross on a layer or pass over a foreign hand-placed via.
- Segments returned twice by SES import are de-duplicated (they doubled every
measured length); vias and track stubs the autorouter abandoned are removed.
Method that worked: USB as a hand-routed pair on the bottom layer (vias at the module pins and at the ESD part), which leaves the top free for the header fan-out. EN capacitor on the bottom, under the EN pin. Labels on the back silkscreen.
Second pass¶
FR_BOTTOM_COST = 20 raised the bottom pour from 44% to 68% of the board in
one piece (the bare antenna strip is 15%). The connector end of the USB pair
then wandered like on 04, so the ESD part moved to sit centred over the
connector and that section is hand-routed. This board's 0.2 mm clearance does
not fit two 0.2 mm tracks in the 0.95 mm channel; they are 0.15 mm there.
Third pass: the vendor document, read¶
Espressif's schematic checklist asks for 22-33 ohm series resistors on the
USB lines close to the chip. The board had none: I had remembered "optional
placeholders". R8 and R9 (22 ohm) now sit on the bottom under the module's USB
vias, 1.05 and 2.24 mm from the pins, placed with pad_at. The four
connector-to-resistor paths are matched within 2.8 mm.
Left open - Item 3, the footprint mismatch warning. - The keep-out is the footprint courtyard's rectangle, not Espressif's drawing. - No RF measurement is possible here; antenna performance depends on the ground edge under the module and is unverified. - USB runs 40 mm over a bottom layer that is otherwise the ground pour. - Not built.
Fourth pass: the footprint warning explained¶
Item 3 was a real defect. kicad-tools 0.29 creates the board with each
footprint's own zones still in the library's coordinates, then shifts them by
each later move, so the module's antenna keep-out sat 136.5 mm left and 83 mm
up of where KiCad puts it. The board was saved only by its own KEEPOUTS.
Found by changing one thing at a time and asking KiCad's DRC (the
lib_footprint_mismatch check switched back on): restoring the keep-out's
layer list left the warning; moving only the zone to the library outline plus
the footprint position cleared it. On the back side the outline is mirrored top
to bottom; with that, KiCad accepts the module at 0, 90 and 180 degrees on top
and 0 and 90 on the bottom.
board.place_footprint_zones now places those zones after every placement,
the example's override is gone, and its full gate is 0 violations, 0
unconnected, 0 parity issues with the check on.