Skip to the content
4 layers100 × 80 mmCoupon, the printed circuit board in 3D

FireFlow — a modulation-first ambient groovebox: the modulation system is the instrument, not a layer on top of it. Five independent lanes per part driving a synth, granular, wavetable, resonator or bucket-brigade deck. Playable today as a VCV Rack plugin; a standalone Daisy hardware prototype is planned.

Size
100 × 80 mm
Copper layers
4
Thickness
1.6 mm
Parts
78
Nets
66
Pads
294
Tracks, vias
447, 167
Pours
4
Routed
Yes
KiCad
KiCad 10.0 (file 20260206)
Created
10 Jul 2026
Updated
2 Nov 2026

Overview

Coupon is an open-source printed circuit board by mcbronkowitch, from the mcbronkowitch/fireflow repository on GitHub, published under the MIT licence. It is a 4-layer board of 100 × 80 mm with 78 parts on 66 nets, routed with 447 tracks and 167 vias.

It is built around the 74HC595, 74HC165D and 74HC4051, with the CD74HC4067M among its other parts. Its parts include 8 diodes and 1 switch. It belongs with the audio and music boards of this library.

Main components

4

README

hardware/coupon/README.md

The FireFlow test coupon

A 100 x 80 mm, 4-layer board built around one Daisy Patch Submodule, whose sole job is answering the eight measurement points the M6 hardware envelope spec puts before panel layout can start — docs/superpowers/specs/2026-08-08-fireflow-hardware-envelope-design.md §5, addendum 2026-08-30 — chief among them the ADC mux settle time (docs/hardware/settle-budget.md's prediction, confirmed here rather than on a breadboard, because the model is linear in node capacitance and a breadboard's jumper-wire capacitance has no defined value). It is not a prototype of the instrument: both mux chip footprints, an unpopulated COM capacitor pad, a probe point on COM, two pot values side by side, hard-tied rail neighbours, fixed-divider reference channels, the real 74HC595/74HC165 chain, and the supply topology as a switchable 0R link are the whole circuit. The full layout act — placement, zone plan, routing — and its reasoning live in docs/superpowers/specs/2026-08-31-coupon-layout-design.md.

Building it

Two generators, run in order, each under a different interpreter:

text
"C:\Users\bernd\AppData\Local\Programs\KiCad\10.0\bin\python.exe" scripts\build.py
"C:\Users\bernd\AppData\Local\Programs\KiCad\10.0\bin\python.exe" scripts\build_pcb.py

Both need KiCad's own Python (pcbnew is not on the system interpreter's path) — build.py first, because build_pcb.py places parts from netlist.build(), the same intent build.py's own proof chain already checked against the exported schematic netlist. build.py writes coupon.kicad_sch, the ERC report and the schematic PDF; build_pcb.py writes coupon.kicad_pcb and runs ten proof steps in order, aborting on the first red one — net comparison, courtyard/shadow placement sanity, plane connectivity, stitch-copper hygiene, ratsnest, the five analog rules of the layout spec's Section 4, a final DRC pass (proof/drc.rpt), and render + Gerber/drill export. scripts/review.py (same interpreter, once build_pcb.py has written a board — Section 6 of its output reads coupon.kicad_pcb through check_layout.py) writes proof/review.md, the sheet meant for actually reading: requirements, channel plan, the Eurorack connector, BOM, known ERC violations, and the layout section — placement table, the four numeric analog-rule measurements, and the two families of measured-but-not-gated deviations (the long 0R ties and the moat-crossing address bus, both below).

Stuffing it

scripts/assembly_plan.py draws the sheet you actually solder from, because the fabricated board cannot be read: 0805 parts sit on a 2 mm pitch and their reference designators need about 5 mm, so the silkscreen overlaps itself and the analog corner is one grey smear. The board renders under proof/ have the same problem for the same reason.

text
python scripts\assembly_plan.py

This one runs under the system interpreter — it reads coupon.kicad_pcb and nothing else, so no pcbnew, no symbol libraries, no KiCad install. Every number in it is read rather than transcribed: placement, rotation, courtyard and pad positions come from the footprints, and the part values come from each footprint's own Value property — the same field the BOM in proof/review.md reports. Re-run it after any placement change and the drawing follows; there is no hand-kept table to drift.

It writes two sheets:

  • proof/coupon-assembly.svg — the working sheet. The whole board, the analog corner enlarged, and a stuffing list with one row per reel. All 78 parts named, coloured by value, with pad 1 marked at its measured position (the notch end on the SOIC chips) and a bar on each LED's _K pad.
  • proof/coupon-overview.svg — the same geometry at a width that survives a narrow column, with eight numbered zones instead of 78 labels. This is the one the journal prints, with the assembly sheet linked beside it.

Labels are placed by search, and the script checks its own work: a slot is rejected if it touches a part body, another label, a leader already drawn or the board outline, and a leader may not cross a part unless nothing legal is left. It prints how many labels needed that fallback (one, as of this writing) and exits non-zero if any part ends up unlabelled or any two labels overlap.

The website keeps its own copies under different names:

text
python scripts\assembly_plan.py --out-dir ..\..\..\FireFlow_Website\public\media\site --prefix fireflow-hw-coupon

coupon.kicad_pcb is generated fresh by every build_pcb.py run, but the file is byte-reproducible. pcbnew orders every UUID-keyed item — footprints, the individual track/via segments sharing a net, and the zone filler's own bookkeeping — by whatever random UUIDs that run happened to mint, so an unseeded build reorders on every run even with no source edit at all (measured, fix-round probe, 2026-09-02: two unseeded builds from the identical committed source diffed at 26226 lines — git diff --no-index, default context, | wc -l; the exact count varies run to run since it depends on which random UUIDs each pair of builds draws, but it is reliably five figures). hardware/gen/kipcb.py's new_board() seeds pcbnew's UUID generator with a fixed constant before creating anything, which makes that whole order reproduce identically — two from-scratch builds under the same seed produced a 0-diff coupon.kicad_pcb. Regenerating and committing the file is therefore safe and expected; a real content change is the only thing that should still show up in git diff.

Every other generated file still shows as modified after each run — that is expected, and it is timestamps only. All 32 of them (every gerber, coupon.drl, both board PNGs, drc.rpt and drc-placement.rpt) embed a CreationDate/wall-clock line KiCad regenerates on every export; the content underneath is unchanged (confirmed by diffing a gerber with its CreationDate/Created by ... date lines excluded: 0 lines differ). Discard this churn with git checkout -- hardware/coupon/fab and, under proof/, every file EXCEPT review.md (coupon-board-front.png, coupon-board-back.png, drc.rpt, drc-placement.rpt) — review.md is review.py's own real content, not timestamp noise, and committing it is the point of that step.

The two solder jumpers switch one way only

Envelope spec §5 point 8 asks for the supply and ground topology "as a switchable 0R link, not as a second board variant: the same copy with and without is the better measurement." On this board the "without" half of that does not exist. AGND and A+3V3 reach the rest of the world through JP_GND and JP_3V3 and through nothing else — the Eurorack header grounds pins 3..6 to GND (scripts/netlist.py:139) and the submodule's A10 feeds +3V3 (:97). Open both jumpers and the analog island is unpowered and unreferenced; both muxes lose VCC. Point 8 is not met as built, and no ordering of the measurements recovers it. A second board does not either: its open state is the same off state.

What the open state is still good for, and only before the jumpers are soldered: meter TP_AGND against TP_GND, and TP_A3V3 against TP_3V3. Both must read open. Continuity there means copper crosses the 1.0 mm moat somewhere it should not — a defect in exactly the separation point 8 is about, and one that is invisible for good once the jumpers are bridged. A continuity check is an unpowered measurement anyway, so this costs nothing but the order it is done in.

Do not power the board with the jumpers open. It is tempting — the digital half needs neither of them — but the six address and enable nets run from U_SR1 straight into both muxes, whose VCC is A+3V3 and whose GND is AGND (scripts/netlist.py:149). With JP_3V3 open that VCC floats while U_SR1 drives those inputs at 3.3 V, which forward-biases the muxes' input clamp diodes and powers the chips parasitically through their own address pins — above the absolute maximum rating of VCC + 0.5 V, with latch-up as the failure mode. The current involved is microamps and it will very probably survive, but nothing is gained by finding out. Nor can it be avoided by not running the scan: ~OE is tied to GND and ~SRCLR to +3V3 (:258), so the 595 outputs are driven from the instant power is applied, with whatever the shift register powered up holding. Meter first, then bridge both jumpers, then apply power — in that order, once.

What does not work is a second join by wire. The only two places where both planes are probeable are TP_AGND at (10.00, 66.00) and TP_GND at (94.00, 3.00) — 105.0 mm apart, the board's full diagonal. A wire that long is a component and not a connection, the same trap the COM rule below names.

A substitute for the lost A/B, neither reviewed nor measured: change the link's impedance instead of removing it. A ferrite bead or a small resistor across JP_GND's pads answers "how good does the single join have to be" at the designed star point. It costs one rework cycle on a pad, and whether an 0603 lands cleanly on SolderJumper-2_P1.3mm_Open_Pad1.0x1.5mm's 1.3 mm pitch is unchecked.

Struck 2026-09-28: point 2's "capacitor at COM" case will not be measured. Bastian decided against the capacitor test; C_COM16/C_COM8 stay unpopulated on every board, and "fit none" (docs/hardware/settle-budget.md §4 finding 1) stays arithmetic. The paragraph below is kept as the record of what that board would have needed.

A second board for point 2's "capacitor at COM" case need not be fully stuffed. It needs the submodule and its sockets, J_PWR, the bulk caps, both jumpers, U_SR1/U_SR2, one mux, R_REFA1/R_REFA2, the decoupling caps and the capacitor under test. No pots, no button, no 165, no LEDs — LEDs only if the round is chasing noise rather than time, which is the load they are there for (scripts/design.py:115).

Bringing it up

First powered board, 2026-09-17. Everything populated except the seven pots, both solder jumpers bridged, and every measurement below taken with the Patch Submodule out of its sockets. That is the whole scope of what a module-less power-up can prove: +3V3 reaches the board only through the submodule's A10 (scripts/netlist.py:97), so with the module out no chip on either side of the moat has a supply. A dead board is the expected state here, not a fault.

Nothing has to be counted, and no reference designator has to be read off the silkscreen. Two signatures identify everything:

  • Which bulk-cap pad is ground. Probe between a pad of C_BP12 and a pad of C_BN12, all four combinations: exactly one reads 0 Ω, and those are the two ground pads. The opposite combination is +12 V against -12 V.
  • Which socket pin is which. With one probe on a ground pad, walk the other along bank A's ten pins. Exactly two read 0 Ω (A4, A7 — same column, one per row), exactly one reads ~1.6 kΩ (A10, +3V3), the rest are open. A1 and A5 being open is correct: with no module, ±12 V has no DC path to ground at all.

Bank A is the 2×5 group at the lower right of the module outline, beside J_PWR, long axis horizontal. Holding the board component side up with the pot/jack edge at the bottom and the header to the right: lower row, right to left, is A1..A5; upper row, right to left, is A10..A6, so A1 and A10 share the column nearest the header. (Pad coordinates: scripts/placement.py's module docstring, transformed by the 180° placement.)

Measured, in the order taken:

checkreadingwhat it proves
across C_BP12, across C_BN1216–17 MΩ, climbingno short on either ±12 V rail; nothing but the cap loads them
C_BP12 pad vs C_BN12 pad, 4 combinationsone at 0 Ω, opposite openrails separated; identifies both ground pads
across C_B3V3 (+3V3 vs GND)1.6 kΩsee below
same, with SW1 held down1.4 kΩR_BTN and the button
bank A sweep against groundA4, A7 at 0 Ω; A10 at 1.6 kΩ; rest openthe pin numbering, without counting
A1 vs J_PWR's north pins, A5 vs its south pins0 Ω (0.1 Ω probe contact)the ±12 V copper reaches the module
powered: A1, A5, A10 against A7-12 V, +12 V, 0 Vpolarity, and that nothing feeds +3V3

The 1.6 kΩ is the single most informative reading on the board, and it is not a fault. Three reference dividers sit across the analog rail — R_REFA1/A2 20k, R_REFC1/C2 20k, R_REFB1/B2 2k — and in parallel they predict 1.67 kΩ, with the 1k divider dominating. Pressing SW1 puts R_BTN's 10k in parallel and predicts 1.43 kΩ. Both landed. Because the dividers hang on A+3V3/AGND while the meter sat on +3V3/GND, that one reading also proves both jumpers conduct — the current goes out through JP_3V3 and back through JP_GND.

Two practical notes:

  • Park the black probe in A7, not A4. Both are ground. A7's neighbours A6 and A8 are unconnected, so a slipped probe there does nothing; A4's neighbour is A5 at +12 V.
  • The bulk caps hold ±12 V for minutes after power-off. Against their own measured 16 MΩ leakage and nothing else, the time constant is about 160 s (arithmetic from the measured value, not timed). Meter A1 and A5 down to ~0 V before the module goes in, or bleed them through a resistor.

What was not measured, and cannot be any more: the moat continuity check this README asks for above — TP_AGND open against TP_GND, TP_A3V3 open against TP_3V3. The jumpers were bridged first, and after that both pairs read 0 Ω whether the moat is clean or shorted. On this board that answer is gone for good.

The seven pots went in the same evening, 2026-09-17 — the RV09 vertical parts from order-bom.md, RV3/RV4/RV7 at 20 k and the other four at 10 k. The support lugs were slightly too large for the board's slots and were clipped off; each pot hangs on its three solder joints alone, with no glue under the body. RV4, the tight position between JP_GND and R_SP11, clears both with a visible gap, and its case reads open against GND (checked 2026-09-28, board unpowered). (This README said "still unpopulated" until 2026-09-28; the board had not been in that state since the 17th.)

With the pots in, the bring-up scan read all seven wipers at 63481…63485 at the clockwise stop, against 63484…63486 on the rail-tied reference channels (commit 084d702's message; no capture is committed). The counter-clockwise readings were taken but not recorded. That is a wiring check, not a characterisation: at either stop a pot's source impedance is near zero, so neither reading says anything about settle time or the wait effect. The hard case — mid travel, R/4, 2.5 kΩ at 10 k and 5 kΩ at 20 k — was measured on 2026-09-28 on RV2, RV4 and RV6: docs/hardware/pots-measured.md. For that round the pots were set to mid travel on the bring-up scan, to what REF_A reads there, and not touched until both probe images had run.

Layer and zone map

4 copper layers, y growing downward from the board's top-left corner:

layercarries
F.Cusignal routing, both plane nets' stitching vias/tracks
In1.Cuground -- two filled zones, GND under the digital half of the board and AGND under the analog half, joined only at JP_GND
In2.Cusupply -- two filled zones, +3V3 under digital and A+3V3 under analog, joined only at JP_3V3, sourced from the module's own A10 (+3V3 OUT)
B.Cusignal overflow where F.Cu is blocked, plus the address/enable bus (see below)

±12V is never a plane -- it runs as short, wide (0.8 mm) point-to-point track, IDC header -> SM supply pins -> bulk caps, per the layout spec's Section 3.

The board splits into two domains, full board width each, with a 1.0 mm moat between them (narrow enough that both solder jumpers' fixed 1.3 mm pad pitch still straddles it): digital, y 1..47.6 mm (the shift-register chain, LEDs, Eurorack header, bulk caps), and analog, y 48.6..79 mm (both muxes, the pots, reference dividers, the audio jack). This is a horizontal split, not the left/right one the layout spec's Section 3 describes on paper -- the Patch Submodule's own pad layout decided it: rotated 180 deg so its sense pins (A2/A3) and audio pins (B1/B2) face the board's bottom edge, both analog anchors end up on the module's south side, so analog is everything below it and digital is everything else. placement.py's module docstring has the full measured reasoning (probed pad coordinates, why 90/270 deg were never an option).

Two known deviations are measured, printed by check_layout.py and review.py's Section 6, and deliberately not gated -- fixing either is a placement-level rework, not something Task 6 (or any task before an order) attempts:

  • Six of the eight 0R neighbour ties sit well away from the mux pin they carry, against "directly at the mux pin" of the layout spec's Section 4 rule 3. The measured spread is a continuum, not a few outliers — only R_LO3 (5.99 mm) and R_HI1 (8.88 mm) are genuinely at their pin, and the three longest (R_HI3 20.56 mm, R_HI4 22.53 mm, R_LO4 22.77 mm) are barely further out than R_HI2 (18.97 mm) and R_LO2 (20.05 mm). Read the table in proof/review.md, not a count. The longest three sit where they do because the only F.Cu approach within reach also carries the 8:1 mux's own enable pin, which cannot lose its route.

    What this is expected to cost — reasoned, not measured. These ties hold the neighbour channels at a rail while the measured channel settles (docs/hardware/settle-budget.md §6). A neighbour tied through 0 R is a static node: 22 mm of copper leaves it essentially at the rail, and a node held at constant DC is a poor aggressor. Charge injection at the address switch decays through the extra trace inductance in nanoseconds, against a 1.6 µs window. So this is expected to be irrelevant to the experiment the coupon exists to run — rule 3 is good hygiene here, not a hard requirement. The coupon itself is the instrument that would settle it; settle-budget.md §5 names crosstalk as what its model is blindest to.

  • Five of the address/enable bus's six nets cross the moat on the board's west edge, 70-74 mm from the star point JP_GND, against Section 3's "beside the star point." All six originate at U_SR1, which sits in the far-west digital column (beside the LEDs) rather than beside JP_GND on the far east -- the crossing follows the source pin, not a routing choice that could have moved it without re-placing U_SR1.

The same fix (re-pitching the 8:1's column and its west approach) would touch both. Nobody has done that work yet.

Before this board is ordered

Populate copies, never rework one board -- the fab ships a handful of identical boards either way, and any question a stuffed board could answer by soldering something differently is answered by populating a second copy, not by reworking the first. Jumper the digital side freely, but never the COM node or the audio path -- everywhere else a wire is a connection; on COM or on AUDIO_OUT_L/AUDIO_OUT_R a wire is a component, and the whole point of measuring on a board instead of a breadboard is that neither of those paths carries an unmodelled jumper-wire capacitance.

One thing still needs hands on a real board, and one that used to is now closed (docs/roadmap.md's M6 section carries the up-to-date status):

  1. The Eurorack bus board -- closed 2026-09-02, metered. This was the one connection on the whole board with no part datasheet behind it: an IDC 2x5 box header's own drawing is mechanical only, and the pinout comes from Doepfer's A-100 standard instead.

    The standard was read first, and it does answer pins 7/8 -- the 16-pin ribbon runs in pairs (-12V, GND, GND, GND, +12V, +5V, CV, Gate) and the 10-pin connector is its first ten conductors, so 3..8 are all GND and +5 V starts no earlier than 11/12. review.py's Section 3 carries the sources. 7/8 stay open regardless, as insurance against a non-standard board rather than against a +5 V that the standard does not put there; it costs two ground pins.

    Then the bench answered the part no document could: Bastian metered the bus board and it follows the standard -- red stripe on -12V, GND on 3..8, +12V on 9/10. Polarity was the only failure mode that costs the board, and it is confirmed the right way round, so this stops being an order blocker. Pins 7/8 stop mattering altogether rather than merely reading as expected: they are open here, so whatever the bus carries there cannot arrive.

    The shroud deviates from Doepfer's own advice on purpose. Doepfer recommends against keyed headers on bus boards, because mis-keyed boards and cables are common. This board keeps one anyway: a rotated or row-shifted plug is +/-12 V into the Patch SM, and the notch is what makes that impossible. Its one cost cannot be tested until the board exists -- a cable keyed the other way will not seat at all -- so that is an assembly-time check, not an order blocker. If the plug does not go in, the fault is the key and the answer is a different cable, never a cut-off nose.

    Where the notch actually points, measured from the committed coupon.kicad_pcb on 2026-09-02 -- so nobody re-measures it: the header sits at (88.00, 22.00), rotation 0 deg, its pins running south to (90.54, 32.16), so the long axis is north-south. The shroud's outer wall is x 84.82..93.72, y 16.90..37.26; its west inner wall (x 86.02) breaks between y 25.03 and y 29.13, with both break ends running out to x 84.82. That gap is the polarising notch: 4.10 mm wide, centred on y 27.08, which is the pin-5/6 row and the header's own centre. Pin 1 is in the x 88.00 column, the notch side. So the notch faces west, into the board. That was never a free choice -- the footprint fixes the notch against the pin numbering and the numbering comes from the netlist; the only free parameter was the rotation, and it is 0 deg.

    Mechanically this is clear on both sides. Nothing is placed west of the header in the y 17..37 band (the nearest neighbour that way is SW1 at y 8, well north). East of it, C_BN12/C_B3V3/C_BP12/TP_3V3 sit on x 96.5 between the shroud and the board's east edge at x 100, but they are flat 0805/testpoint parts passing under a plug that rests on the shroud. The ribbon can therefore fold either way: 6.3 mm to the east edge, or out over open board to the west.

  2. The Patch Submodule's own land-pattern dimensions -- vendored from Electrosmith's KiCad library (hardware/lib/README.md), unproven until a module actually seats in a fabricated board.

Parts

78
QtyValueFootprintReferences
160RR_0805_2012Metric_Pad1.20x1.40mm_HandSolderR_HI1, R_HI2, R_HI3, R_HI4, R_LO1, R_LO2, R_LO3, R_LO4, R_S8, R_S16, R_SP10, R_SP11, R_SP12, R_SP13, R_SP14, R_SP15
101kR_0805_2012Metric_Pad1.20x1.40mm_HandSolderR_LED1, R_LED2, R_LED3, R_LED4, R_LED5, R_LED6, R_LED7, R_LED8, R_REFB1, R_REFB2
8greenLED_0805_2012Metric_Pad1.15x1.40mm_HandSolderD1, D2, D3, D4, D5, D6, D7, D8
8probeTestPoint_Pad_D1.5mmTP_3V3, TP_A3V3, TP_ADC11, TP_ADC12, TP_AGND, TP_AUDIO_L, TP_CLK, TP_GND
5100nC_0805_2012Metric_Pad1.18x1.45mm_HandSolderC_IN1, C_M8, C_M16, C_SR1, C_SR2
510kR_0805_2012Metric_Pad1.20x1.40mm_HandSolderR_BTN, R_REFA1, R_REFA2, R_REFC1, R_REFC2
410kPotentiometer_Alpha_RD901F-40-00D_Single_VerticalRV1, RV2, RV5, RV6
410u 25VC_0805_2012Metric_Pad1.18x1.45mm_HandSolderC_B3V3, C_BA3V3, C_BN12, C_BP12
320kPotentiometer_Alpha_RD901F-40-00D_Single_VerticalRV3, RV4, RV7
20R linkSolderJumper-2_P1.3mm_Open_Pad1.0x1.5mmJP_3V3, JP_GND
274HC595SOIC-16_3.9x9.9mm_P1.27mmU_SR1, U_SR2
2COM probeTestPoint_Pad_D1.5mmTP_COM8, TP_COM16
2DNPC_0805_2012Metric_Pad1.18x1.45mm_HandSolderC_COM8, C_COM16
13.5 mm outJack_3.5mm_CUI_SJ1-3513N_HorizontalJ_AUDIO
174HC165SOIC-16_3.9x9.9mm_P1.27mmU_IN1
174HC4051SOIC-16_3.9x9.9mm_P1.27mmU_MUX8
1CD74HC4067MSOIC-24W_7.5x15.4mm_P1.27mmU_MUX16
1Daisy Patch SMDAISY_PATCH_SMU_SM
1Eurorack powerIDC-Header_2x05_P2.54mm_VerticalJ_PWR
1tactileSW_PUSH_6mmSW1

More from mcbronkowitch/fireflow

1
  • 4 layers300.8 × 110 mmFireflow, the printed circuit board in 3D

    Fireflow

    mcbronkowitch/fireflow

    FireFlow — a modulation-first ambient groovebox: the modulation system is the instrument, not a layer on top of it. Five independent lanes per part driving a synth, granular, wavetable, resonator or bucket-brigade deck.

    208 parts2MIT

Similar boards

  • 4 layers120 × 108 mmDk2 06, the printed circuit board in 3D

    Dk2 06

    mattybrad/drumkid2

    189 parts12GPL-3.0
  • 2 layers80.2 × 94.6 mmMANUEL ARGIBAYMOLINA ELECTRONICS, the printed circuit board in 3D

    MANUEL ARGIBAYMOLINA ELECTRONICS

    ManuelArgibay/Industrial-I-O-Controller

    Industrial 24V I/O controller featuring protected high-side and low-side outputs, multiplexed analog and digital inputs, and MCU-based control.

    89 parts
  • 2 layers335.1 × 620 mmWEASEL KiCad, the printed circuit board in 3D

    WEASEL KiCad

    joebmz98/Weasel-West-Coast-Synthesiser

    Digital east coast style synth heavily inspired by the Buchla Easel based on the Daisy Seed.

    314 parts3MIT
  • 4 layers100 × 100 mmPCBA OCTV, the printed circuit board in 3D

    PCBA OCTV

    intechstudio/grid-hardware

    403 parts59GPL-3.0
  • 2 layers105 × 195 mmStemsMIDI, the printed circuit board in 3D

    StemsMIDI

    kinoshita-lab/StemsMIDI

    MIDI Controller for Traktor Stems

    191 parts5MIT
  • 4 layers32.7 × 99.7 mmGrid devkit, the printed circuit board in 3D

    Grid devkit

    RustyCircuit/grid-devkit

    KiCad PCB designed to interoperate with the Intech Studio Grid ecosystem. This devkit is a fork of Intech's VSNX hardware design, adapted to fill a gap between fully-developed hardware and breadboard/DIY controller…

    167 parts1GPL-3.0

Questions

How big is the Coupon PCB?

It measures 100 × 80 mm and has 4 copper layers, 1.6 mm thick.

What parts does the Coupon use?

78 parts, built around the 74HC595, 74HC165D, 74HC4051 and CD74HC4067M. The full list of its parts is on this page, with links to each part's symbol and footprint where PCBKit's library has them.

Where can I download the Coupon KiCad files?

From the mcbronkowitch/fireflow repository on GitHub, at the commit this page was made from: the board is hardware/coupon/coupon.kicad_pcb and its schematic hardware/coupon/coupon.kicad_sch. PCBKit opens the board in the browser, with nothing to install.

Can I use the Coupon design in my own project?

Its repository publishes it under the MIT licence: you may reuse it on that licence's terms (keep the author's notice, and for a copyleft licence share your changes alike).