Prototype Architecture — Overview
The locked prototype architecture for the lamp — two lamp boards plus a separate SWD adapter, a 24-LED warm-white array at 500mA, an AL8860 constant-current buck, an STM32G031 control MCU, and a stopped potentiometer mapped to brightness. Decisions, not options.
This section is the locked prototype architecture: the thing KiCad work starts from. Everything here is a decision with a recorded reason that traces back to the power and research sections. Where those pages enumerated options, this section picks one and says why — see Decisions for the chosen-vs-rejected record.
Nothing here is bench-confirmed
Board P's topology was never validated on working hardware (seeLessons Carried From zudo-pd), and Board L is a new design that has never existed as a PCB. This is a datasheet-grounded paper design. Treat the first build as the first real test, and follow the bring-up gates inNext Steps.
Headline numbers
| Parameter | Locked value |
|---|---|
| Lamp board count | 2 — Board P (USB-PD front end) + Board L (driver, LEDs, control); the SWD adapter is a separate service accessory |
| PD contract | 15 V / 3 A (45 W cap) |
| System draw | ~6.2 W, ~410 mA — 13.7% of the contract cap, 7.3x margin |
| LED array | 24x SMD2835 warm white, 8 parallel strings of 3 series |
| CCT / CRI | 3000 K (2800–3100 K bin), CRI Ra ≥ 80 |
| Drive current | 500 mA total, 62.5 mA per string |
| Light output | ~700 lm typical, ≥620 lm worst-case bin |
| Driver | AL8860MP-13 constant-current buck, hysteretic, ~440 kHz |
| Logic rail | 3.3 V from an AP63203 synchronous buck |
| MCU | STM32G031F8P6 (Cortex-M0+, 64 KB flash, TSSOP-20) |
| Knob | ALPS RK10J11E0034 stopped potentiometer → base brightness |
| Dimming | Analog (RC-filtered 20 kHz PWM into the driver's CTRL pin) — no visible flicker by construction |
| Board L dissipation | ~6.1 W spread over 24 LED sites + 8 ballast resistors — plain FR-4, no MCPCB |
System block diagram
Board split
Two boards, no LED daughterboard. Board P must be its own PCB (it is a reusable switched-15 V sink module, and it has to be bench-brought-up alone before anything is connected to it). Everything else lands on Board L.
The case for a third board — an LED module on aluminum MCPCB — was evaluated and rejected. The thermal budget page sets the plain-FR-4 ceiling at roughly 1 W per LED site; this design runs 0.19 W per site across 24 sites. That is squarely in that page's "many small sites, generous pour" regime, not the concentrated-COB regime that forces metal core. A third board would add a second interconnect carrying the driver's switching node over a cable — a real EMI cost — to solve a thermal problem this design does not have.
The split boundary is still designed in
The LED array is drawn as a self-contained block with a 4-net interface: LED_P(array anode rail), LED_N (common cathode = the driver's switching node), NTC_SENSE, and GND. If the later enclosure phase forces the emitters onto a separate plane from the knob, the array can be lifted onto a daughterboard across exactly those four nets without touching the rest of Board L. The reason it is not split now is the LED_Nswitching node: running a ~440 kHz, 15 V-swing node down a wire harness is an EMI problem that only pays for itself if the mechanics actually demand it.
Interfaces
| Interface | Physical | Nets |
|---|---|---|
| Charger → Board P | USB-C receptacle (J1) | CC1, CC2, VBUS, GND |
| Board P → Board L | Direct XFCN 1x6, 2.54 mm pair: PZ254V-11-06P male on Board P and PM254V-11-06-H85 female on Board L | 1,2 = switched 15 V (paired); 3 = ATT; 4 = PDOK; 5,6 = GND (paired). Unkeyed: verify physical pin-1 alignment before power |
| Board P programming | J2 pogo pads 1x4 (I2C NVM + reset), J3 pogo pads 1x8 (debug) | SCL, SDA, GND, RESET |
| Board L programming | J3 pad group 1x5, 2.54 mm (SWD) | SWDIO, SWCLK, NRST, 3V3, GND |
| Board L debug | J4 pad group 1x3, 2.54 mm (UART) | PA2 TX, PA3 RX, GND |
| Power control | C5446803 slide switch assembled on rear Board L | Fused positive feed / switched V15 return; Board P remains powered |
| User control | ALPS RK10J11E0034 stopped potentiometer, rear low-profile wheel | Endpoints and wiper |
The XH connector's per-contact current math is inherited unchanged from Board P's downstream interface: 3 A/contact nameplate, doubled contacts per rail. Against this design's actual 410 mA draw that is 14x derated margin per contact, not the 1.6x the connector was sized for against the 3 A contract cap. The doubling stays anyway — it costs nothing, it is already in Board P's spec, and it keeps the harness reusable if a brighter revision raises the current.
Lamp power
The C5446803 slide switch turns off the LED driver and MCU supply by opening Board L’s fused positive feed. Board P remains connected to USB-PD. The short slide switch is assembled on the rear PCB; ON restarts the MCU after sufficient rail decay; rapid toggles need verification.
What the knob does
The knob sets base brightness, and nothing else. Speed and depth of the fire/wave modulation are firmware constants. The full reasoning is in Decisions; the short version is that a lamp knob that does not change brightness is a surprise. RV1 has fixed end stops and absolute position; there is no push-button mode gesture. Firmware must start dark and ramp to the validated position, with endpoint calibration, averaging and installed-direction verification.
Where to go next
| Page | Contents |
|---|---|
| Board P | Locked front-end spec, full net table, interface pin assignment |
| Board L | Driver/control/LED design, full pin-level net table, all component values with worked math |
| Final BOM | Every part, C-number, tier, stock, and the Extended-fee arithmetic |
| Decisions | Chosen vs rejected per slot, with reasons |
| Next Steps | The KiCad phase — bring-up order, footprint pipeline, risks |