by adom
A 12,960-LED addressable wall for the Adom workcell: 9 identical 160x160mm tiles (1,440 Dialight 587-1024-147F 2020 LEDs each, RP2040 + USB-C per tile) on a LrgLrg 576x576 scaffold. Remote Adom users each own a pixel.
1482e7a
LED Grid Wall — design decisions (2026-09-28)
Goal
A huge contiguous grid of individually addressable mini RGB LEDs in an Adom workcell. Remote Adom users each "own" one LED and set its color anytime; a live camera watches the grid so users can find their pixel. Emergent collaborative art.
Locked decisions (with the user)
- LED: Dialight 587-1024-147F (Mouser 645-587-1024-147F)
- 2020 package (2.0 x 2.0 x 0.9 mm) — the smallest addressable class, per user
- Single-wire NRZ 800 kbps, 24-bit GRB — WS2812-protocol compatible (datasheet p.4)
- No external capacitor required (datasheet p.2: self-contained IC + dies)
- VDD 3.3–5.5 V, 5 mA/channel (dimmer than WS2812B-2020's 12 mA/ch — accepted tradeoff)
- Mouser stock 20,947; $0.37 @500, $0.29 @9000
- Pinout: 1=DOUT, 2=VDD, 3=DIN, 4=GND — NOT WS2812B-2020 footprint-compatible (VDD/GND swapped)
- Sourcing note: user directed Mouser (3rd party fab build); Mouser does not distribute Worldsemi. WS2812B-2020 (existing Adom basic part) is 0 stock at LCSC/Adom inventory, so the Dialight 2020 is the only Mouser-stocked 2020 single-wire addressable LED.
- Tile: ~160 x 160 mm, 9 tiles (3x3) on a LrgLrg 576x576 scaffold
- 3rd party fab panel verified 300 x 380 mm (proforma 2026-08-21); with the user's 1" margin the max single board is ~249 x 329 mm — a 160 mm tile yields 2 per panel.
- LED pitch 3.5 mm -> ~44 x 40 = ~1,760 LEDs per tile, ~15.8k total
- MCU strip on the bottom edge of each tile so the LED field stays contiguous.
- MCU: RP2040 + USB-C per tile
- ray/rp2040 wiki component page exists; PIO drives 4 parallel WS2812 chains/tile (~440 LEDs/chain -> ~75 fps full-tile refresh)
- Workcell host relays remote color commands over USB CDC; no RF on the tile.
Support parts (all have wiki component pages)
- RP2040 — ray/rp2040
- Flash W25Q128 (SOIC-8) — adom/w25q128fvsig or jvsniq
- 12 MHz crystal SMD3225 — adom/x322512msb4si (or adom/abm8-272-t3)
- 3.3 V LDO AP2112K-3.3 (SOT-23-5, 600 mA) — adom/ap2112k-3.3
- Level shifter SN74AHCT125PWR (TSSOP-14, quad, 5 V Vcc, VIH=2V) — adom/sn74ahct125pwr
- USB-C receptacle 16-pin — adom/3d-xkb-u262-16xn
- Power entry XT60PW-M — adom/xt60pw-m-208a (20.8 A; firmware caps tile draw to <=20 A)
- Passives — Adom Basic Parts (100 nF/1 uF/10 uF MLCCs, 27R/1k/10k resistors)
Power budget (per tile, worst case)
- 1,760 LEDs x 15.8 mA (3 ch x 5 mA + 0.8 mA idle) = 27.8 A peak @ 5 V (full white, uncapped)
- Firmware global power cap: <= 20 A/tile (100 W); typical art patterns ~10-20%.
- 9 tiles = 180 W capped, ~1.4 kW theoretical uncapped -> cap is mandatory.
- 4-layer stackup F.SIG / GND / V5 / B.SIG so inner planes carry LED current.
Known issues found during selection
- adom/ws2812b-2020 page: STEP file header and geometry are a 5.0x5.0 mm 5050 model (bounds 6.45 x 6.25 x 1.6 mm), GLB is ~4 x 3.6 x 2.4 mm — neither matches the 2.0x2.0 footprint. Filed for fixing; not used by this design (Dialight part instead).
Amendment 2026-09-29a — Dialight footprint corrected by measurement
The component-page build measured the 600dpi datasheet renders: p.1/p.2 callouts are
size+gap pairs, not size+pitch. Real geometry: terminals 0.6x0.6 on 1.4mm pitch (centers
±0.70); recommended pads 0.9x0.9 on 1.7mm pitch (centers ±0.85). Published footprint
Dialight_587-1024-147F uses pads at (±0.85,±0.85), courtyard ±1.55mm, IPC-7351 zero
orientation, pad 1 (DOUT) top-left. The earlier ±0.45/0.8mm numbers in this file and
GENERATOR-SPEC.md are superseded. Page: adom/587-1024-147f v0.1.1 (published, verified).
Amendment 2026-09-29b — I2C tile mesh, one USB-C uplink (user-directed)
Per the user: tiles interconnect by I2C between their RP2040s; only ONE tile (the master) uplinks to the control panel over USB-C.
- ONE board design stands: every tile gets the same MCU strip (USB-C populated on all; only the master's cable is plugged).
- Each tile gains 2x MachineContactMedium for I2C SDA/SCL in the MCU strip; the scaffold harness buses them to all 9 tiles (GND returns through the power contacts).
- 4 address-strap solder jumpers (GPIO4..7, pull-ups, strap to GND) set the tile's I2C address 0..15; address 0 = master (USB bridge). Slave tiles run I2C peripheral mode.
- Bus: I2C 400kHz (FM+ 1MHz possible after bring-up), 22k pull-ups on every tile (2.4k combined with 9 tiles — inside the 3mA drive budget).
- Firmware: one UF2 for all tiles; role from straps. Master: USB CDC <-> I2C bridge. Bulk frames carry the tile address; sparse per-LED updates are the common case (full-tile frame ~110ms at 400kHz — art-rate, fine).
Amendment 2026-09-29c — 6 layers, entire MCU strip on the BACK face
After the strip routing analysis proved the 14mm front strip could not carry the MCU section at 0.2mm clearance, the whole MCU strip (RP2040, flash, crystal, LDO, shifter, USB-C, straps, caps, I2C contacts) moved to the back face, and the board went 4L -> 6L (F.Cu / In1 GND / In2 V5 / In3 GND / In4 GND / B.Cu). Wins:
- The front is a pure LED canvas — no strip band on camera; the grid reads fully contiguous across the 3x3 wall (the user's actual requirement).
- V3V3 and V1V1 become B.Cu zones that connect their pads directly (no stub vias).
- Strip signals escape the QFN with staggered fanout + an F.Cu river (front strip empty).
- Docking contacts MP1-4 move to the back too, which is how they mate scaffold pins (pins insert from below).
- 4 GND layers for the 20A LED return. LED count is unchanged: 1424/tile (40x36 grid minus 16 docking cells), 12,816 for the wall.
# LED Grid Wall — design decisions (2026-09-28)
## Goal
A huge contiguous grid of individually addressable mini RGB LEDs in an Adom workcell.
Remote Adom users each "own" one LED and set its color anytime; a live camera watches
the grid so users can find their pixel. Emergent collaborative art.
## Locked decisions (with the user)
1. **LED: Dialight 587-1024-147F** (Mouser 645-587-1024-147F)
- 2020 package (2.0 x 2.0 x 0.9 mm) — the smallest addressable class, per user
- Single-wire NRZ 800 kbps, 24-bit GRB — WS2812-protocol compatible (datasheet p.4)
- No external capacitor required (datasheet p.2: self-contained IC + dies)
- VDD 3.3–5.5 V, 5 mA/channel (dimmer than WS2812B-2020's 12 mA/ch — accepted tradeoff)
- Mouser stock 20,947; $0.37 @500, $0.29 @9000
- Pinout: 1=DOUT, 2=VDD, 3=DIN, 4=GND — NOT WS2812B-2020 footprint-compatible (VDD/GND swapped)
- Sourcing note: user directed Mouser (3rd party fab build); Mouser does not distribute Worldsemi.
WS2812B-2020 (existing Adom basic part) is 0 stock at LCSC/Adom inventory, so the
Dialight 2020 is the only Mouser-stocked 2020 single-wire addressable LED.
2. **Tile: ~160 x 160 mm, 9 tiles (3x3) on a LrgLrg 576x576 scaffold**
- 3rd party fab panel verified 300 x 380 mm (proforma 2026-08-21); with the user's 1" margin
the max single board is ~249 x 329 mm — a 160 mm tile yields 2 per panel.
- LED pitch 3.5 mm -> ~44 x 40 = ~1,760 LEDs per tile, ~15.8k total
- MCU strip on the bottom edge of each tile so the LED field stays contiguous.
3. **MCU: RP2040 + USB-C per tile**
- ray/rp2040 wiki component page exists; PIO drives 4 parallel WS2812 chains/tile
(~440 LEDs/chain -> ~75 fps full-tile refresh)
- Workcell host relays remote color commands over USB CDC; no RF on the tile.
## Support parts (all have wiki component pages)
- RP2040 — ray/rp2040
- Flash W25Q128 (SOIC-8) — adom/w25q128fvsig or jvsniq
- 12 MHz crystal SMD3225 — adom/x322512msb4si (or adom/abm8-272-t3)
- 3.3 V LDO AP2112K-3.3 (SOT-23-5, 600 mA) — adom/ap2112k-3.3
- Level shifter SN74AHCT125PWR (TSSOP-14, quad, 5 V Vcc, VIH=2V) — adom/sn74ahct125pwr
- USB-C receptacle 16-pin — adom/3d-xkb-u262-16xn
- Power entry XT60PW-M — adom/xt60pw-m-208a (20.8 A; firmware caps tile draw to <=20 A)
- Passives — Adom Basic Parts (100 nF/1 uF/10 uF MLCCs, 27R/1k/10k resistors)
## Power budget (per tile, worst case)
- 1,760 LEDs x 15.8 mA (3 ch x 5 mA + 0.8 mA idle) = 27.8 A peak @ 5 V (full white, uncapped)
- Firmware global power cap: <= 20 A/tile (100 W); typical art patterns ~10-20%.
- 9 tiles = 180 W capped, ~1.4 kW theoretical uncapped -> cap is mandatory.
- 4-layer stackup F.SIG / GND / V5 / B.SIG so inner planes carry LED current.
## Known issues found during selection
- adom/ws2812b-2020 page: STEP file header and geometry are a 5.0x5.0 mm 5050 model
(bounds 6.45 x 6.25 x 1.6 mm), GLB is ~4 x 3.6 x 2.4 mm — neither matches the 2.0x2.0
footprint. Filed for fixing; not used by this design (Dialight part instead).
## Amendment 2026-09-29a — Dialight footprint corrected by measurement
The component-page build measured the 600dpi datasheet renders: p.1/p.2 callouts are
size+gap pairs, not size+pitch. Real geometry: terminals 0.6x0.6 on 1.4mm pitch (centers
±0.70); recommended pads 0.9x0.9 on 1.7mm pitch (centers ±0.85). Published footprint
`Dialight_587-1024-147F` uses pads at (±0.85,±0.85), courtyard ±1.55mm, IPC-7351 zero
orientation, pad 1 (DOUT) top-left. The earlier ±0.45/0.8mm numbers in this file and
GENERATOR-SPEC.md are superseded. Page: adom/587-1024-147f v0.1.1 (published, verified).
## Amendment 2026-09-29b — I2C tile mesh, one USB-C uplink (user-directed)
Per the user: tiles interconnect by I2C between their RP2040s; only ONE tile (the master)
uplinks to the control panel over USB-C.
- ONE board design stands: every tile gets the same MCU strip (USB-C populated on all;
only the master's cable is plugged).
- Each tile gains 2x MachineContactMedium for I2C SDA/SCL in the MCU strip; the scaffold
harness buses them to all 9 tiles (GND returns through the power contacts).
- 4 address-strap solder jumpers (GPIO4..7, pull-ups, strap to GND) set the tile's I2C
address 0..15; address 0 = master (USB bridge). Slave tiles run I2C peripheral mode.
- Bus: I2C 400kHz (FM+ 1MHz possible after bring-up), 22k pull-ups on every tile
(2.4k combined with 9 tiles — inside the 3mA drive budget).
- Firmware: one UF2 for all tiles; role from straps. Master: USB CDC <-> I2C bridge.
Bulk frames carry the tile address; sparse per-LED updates are the common case
(full-tile frame ~110ms at 400kHz — art-rate, fine).
## Amendment 2026-09-29c — 6 layers, entire MCU strip on the BACK face
After the strip routing analysis proved the 14mm front strip could not carry the MCU
section at 0.2mm clearance, the whole MCU strip (RP2040, flash, crystal, LDO, shifter,
USB-C, straps, caps, I2C contacts) moved to the back face, and the board went 4L -> 6L
(F.Cu / In1 GND / In2 V5 / In3 GND / In4 GND / B.Cu). Wins:
- The front is a pure LED canvas — no strip band on camera; the grid reads fully
contiguous across the 3x3 wall (the user's actual requirement).
- V3V3 and V1V1 become B.Cu zones that connect their pads directly (no stub vias).
- Strip signals escape the QFN with staggered fanout + an F.Cu river (front strip empty).
- Docking contacts MP1-4 move to the back too, which is how they mate scaffold pins
(pins insert from below).
- 4 GND layers for the 20A LED return.
LED count is unchanged: 1424/tile (40x36 grid minus 16 docking cells), 12,816 for the wall.