Our own circuit boards, in the dock-door guards

One board we design for each steel guard at a dock door, with every chip soldered on it. Guard A carries the RK3576J brain, the door and dock camera chips and a certified mains supply on the door’s own circuit; guard B carries the ESP32-S3, the second door camera chip, the 60 GHz radar and the air chip. Each board is a main board over the steel bay plate, a strip behind the front face and a strip behind the nose, joined by the board’s own flexes. No plugs, ribbons or adapters inside; one latched lead joins the two guards.

Stage 1 of 6 done: shape and placementNot yet ready to manufactureUpdated 4 November 2026 · rev A-0

Design files Design review Quote pack Who can make it Our board platform

The board in 3D

Every part at its real size, in its real place. Turn it, zoom in, switch between the boards in the guards and laid flat, and click any part to see what it is. The chip bodies are simple blocks until each part's own 3D model is added.

Show the guards

Loading the 3D board…

Drag to turn · scroll to zoom · right-drag to move · click a partDrag to turn · tap a part

Where everything is in the guards

Door camerasboth noses

A Sony IMX415 camera chip in guard A and an OmniVision OV2640 in guard B, each on a small tab turned 35° out to the trailer and 15° down, the lens barrel through the steel nose to just inside the face: the trailer, its creep, the leveler lip and every pallet through the door, seen from both sides.

Dock cameraguard A front

A second IMX415 behind the front face, turned 30° to the door and 15° down over the dock floor: the leveler plate, forklifts coming to the door, people at the open edge, pallets staged in the lane.

Radar and airguard B front

A TI IWRL6432AOP 60 GHz radar chip 4 mm behind the 50 × 20 window measures the speed of forklifts and people coming to the door; a Sensirion SHT45 behind the vents reads the door air.

Call lights and soundersboth guards

Eight colour LEDs behind each nose slot (seen from the trailer and across the door) and eight behind each front slot (seen in the building), and a sounder behind the tone holes: green door free, amber trailer, red stop.

Brain, radio and securityguard A bay

The RK3576J with its memory and a heat block down to the steel bay plate, the Wi-Fi module behind a new antenna window, the security chips that sign every call, the latched microSD cards with our software, and a Hall switch under a magnet in each cap.

Mains and the leadguard A bay

A certified PCB-mount 5 V supply on the dock door’s own 120 V circuit, fused, with a push-in terminal; a latched, keyed lead in the door header conduit carries 5 V, data and the camera sync to guard B.

Both guard boards laid flat, both sides, every part at its real size

Laid flat: guard A on the left, guard B on the right; each main board at the bottom, its front strip above (flex drawn short) and its nose strip to the left, the camera tabs in their slots. Tap the drawing to open it full size.

Design review: checked before we order

Every board-level number DockOne's jobs depend on, checked against its target with the source of each number, using the rules every one of our boards shares (products/_shared/board-rules.mjs). Fixed means a lesson changed this board; to prove means it passes on paper and is measured on rev A before boards are ordered.

No check fails24 pass · 3 fixed by the review · 12 to prove before we order · 0 failDesign review of 4 November 2026

Area	Check	Our board	Target	Status
Camera	Door camera A: a 20 mm trailer creep step, trailer floor 2 m in at 3.13 m M12 lens, 60° across (chip on its side): about 36 px per degree across 2160 px. Creep is tracked on the trailer's floor edge; 4 px on one step is enough to see it move between frames (est.).	13.2 px	≥ 4 px	passes
Camera	Door camera A: frame rate the calls need A standard mode of the IMX415.	5 fps at 2160 × 3840	chip: 30 fps at 2160 × 3840	passes
Camera	Door camera A: data rate vs its 4-lane link 88 % spare on the link.	0.41 Gbps (10-bit)	≤ 80 % of 3.564 Gbps	passes
Camera	Door camera A (gap): a 10 mm gap opening at the leveler lip at 1.82 m M12 lens, 60° across (chip on its side): about 36 px per degree across 2160 px. The open-edge call starts when the lip lifts off the trailer bed (est. 4 px on a 10 mm gap).	11.3 px	≥ 4 px	passes
Camera	Door camera A (gap): frame rate the calls need A standard mode of the IMX415.	5 fps at 2160 × 3840	chip: 30 fps at 2160 × 3840	passes
Camera	Door camera A (gap): data rate vs its 4-lane link 88 % spare on the link.	0.41 Gbps (10-bit)	≤ 80 % of 3.564 Gbps	passes
Camera	Dock camera A: a 50 mm hi-vis stripe on a person at the door at 2.22 m M12 lens, 70° across: about 54.9 px per degree across 3840 px. People are found by the vest's stripes; 8 px on one stripe is the smallest the model is trained on (est.).	70.8 px	≥ 8 px	passes
Camera	Dock camera A: frame rate the calls need A standard mode of the IMX415.	2 fps at 3840 × 2160	chip: 30 fps at 3840 × 2160	passes
Camera	Dock camera A: data rate vs its 4-lane link 95 % spare on the link.	0.17 Gbps (10-bit)	≤ 80 % of 3.564 Gbps	passes
Camera	Door camera B: a 20 mm trailer creep step, trailer floor 2 m in at 3.13 m M7 lens, 60° across (chip on its side): about 20 px per degree across 1200 px. The second view for stereo: 4 px on the step, the same as guard A (est.).	7.3 px	≥ 4 px	passes
Camera	Door camera B: frame rate the calls need A standard mode of the OV2640.	5 fps at 1200 × 1600	chip: 15 fps at 1200 × 1600	passes
Camera	Door camera B: data rate vs its 1-lane link 60 % spare on the link.	0.08 Gbps (8-bit)	≤ 80 % of 0.192 Gbps	passes
Camera	Image processor load, all cameras 4 cameras into one image processor.	109 Mpx/s	≤ 480 Mpx/s (RK3576J ISP)	passes
Camera	Cameras in step for timed calls Trailer creep, pallet height and the far rear post are measured from both door cameras at once; a trailer creeping 70 mm moves only millimetres between frames. A real-time (PREEMPT_RT) kernel is asked for in the quote (lesson: sync).	one frame-sync line from the RK3576J to both IMX415 chips, and down the lead to guard B, which time-stamps every OV2640 frame to it	the two door views within 1 ms of each other (est.)	passes
Camera	AI engine time per frame Trailer, pallet, person and forklift models on one NPU; the radar is read on guard B. Measured on the dev kit.	6 TOPS	every door and dock call from 5 fps door and 2 fps dock frames in under 0.5 s, the stereo creep track at the same time	to prove on the dev kit
Power	Supply About 5.5 W working (RK3576J at 3.5 W, two cameras 0.6 W, Wi-Fi 0.5 W, guard B 1.0 W) and 11 W at a call (16 call lights at full 3.8 W, two sounders 1 W), est.: 55 % of the supply.	Mean Well IRM-20-5 PCB-mount 5 V 20 W supply on the dock door's own 120 V circuit (class II, fused), guard B fed 5 V over the lead	≤ 70 % of the supply (14 W)	passes
Power	Power cut The supply's hold-up and the log write are measured on rev A with the power pulled mid-call.	no cells: when the door circuit is off the dock door does not work either; the trailer log is written to eMMC every second	no lost calls or log in a power cut	to prove on the dev kit
Heat	RK3576J temperature, a summer dock with the doors open, 45 °C in the guard top (est.) The steel guard and its bay plate sit in the dock air; counted at 45 °C. 3.5 W through a 24 × 24 aluminium block and gap pads to the 5 mm steel bay plate welded into the guard (5 K/W, est.).	63 °C	≤ 85 °C (rated)	passes
Heat	Camera chips away from the processor The camera chips are on their own tabs on the strips, well away from the brain.	door camera 228 mm, dock camera 186 mm from the RK3576J	≥ 30 mm	passes
Shock	Shock and vibration A forklift strike on the guard (the PAS 13 test energies) shakes the top bay; about 500 g peak at the boards (est., the strike is low on the guard, the bay is 1 m up). MEMS oscillators are fitted anyway, and both card sockets are latched.	500 g (est.)	standard parts	passes
Software	Starts our software from the microSD card, no setup Owner requirement: power it from its own supply and our software runs from the card. Rev A and sale boards boot from the card; guard A's eMMC holds no system copy (it ships empty, so the card always starts first: Rockchip tries the eMMC before the card) and keeps the trailer log. Updates write to the card with two system copies, only when the door is shut and never with a trailer at the door. The quote asks for U-Boot and a device tree from Rockchip's RK3576 SDK, and our ESP32-S3 image for guard B.	microSD socket with a hinged, latched lid (Amphenol 101-00660 class), fitted on every board: our software starts from the card; microSD socket with a hinged, latched lid (Amphenol 101-00660 class), fitted on every board: guard B's software and clip store on the card	card socket fitted on every board, latched	passes
Software	Card power switch and protection A card that hangs must be restarted by cutting its power, not the whole board; the card slot is the one place a person touches the electronics (lesson: sdstart). Change: U90, U92, U91, U93 added (4 Nov 2026, CricketOne card-start review).	2 power switches and 2 ESD arrays for 2 card sockets	every socket: switched card supply, ESD on all 6 lines	fixed by the review
Software	The card wins the start Rockchip's start order is SPI flash, eMMC, card, USB: a written eMMC starts before the card. Proved on the dev kit: power on with a card in and the eMMC empty (lesson: sdstart). Change: eMMC ships empty and the loader reads the card first (4 Nov 2026; it used to say the eMMC kept a copy, which would have started instead of the card).	eMMC shipped empty; our loader reads the card first	the processor starts from the card every time	fixed by the review
Software	Card grade Consumer cards wear out and fail hot; the image is read-only with a separate data partition. Proved: 1,000 power cuts during writes on rev A (lesson: sdstart).	industrial pSLC microSD (Kingston Industrial or Swissbit S-56u), −40 to 85 °C	rated for the heat, cold and writes	to prove on the dev kit
Software	Clip buffer for an unsure call The rest of the memory holds Linux, the program and the AI model (lesson: clip).	39.5 s of 8-bit 5 fps video in 40 % of the 4 GB memory	≥ 3 s	passes
Radio	Wi-Fi from inside a steel guard The guard is 4.8 mm steel on every side; only the PC windows let radio out, so the antenna sits right behind its own window (CAD rev T). Antenna outside the metal: the 60 × 10 PC antenna window in the guard A front face (lesson: enclosure). Measured on rev A (lesson: antenna).	guard A's Wi-Fi 6 module behind a new 60 × 10 PC antenna window in the front face, into the building	certified Wi-Fi module, site Wi-Fi at the door	to prove on the dev kit
Radio	Guard B's radio (BLE back-up link) If the lead is cut, guard B still reaches guard A over BLE across the door. Antenna outside the metal: the 60 × 10 PC antenna window in the guard B front face (lesson: enclosure). Measured on rev A (lesson: antenna).	ESP32-S3 module's PCB antenna behind the same 60 × 10 window in guard B	certified module; the lead is the main link	to prove on the dev kit
Layout	Fast lines on a solid ground plane The layout rules go to the maker in the quote; the routed board's design rule check (0 copper errors, every keep-out kept) and the maker's impedance coupon prove them (lesson: planes).	RK3576J memory; 3 camera MIPI links (100 Ω pairs); USB (90 Ω); antenna feed (50 Ω, keep-out on every layer); card lines (length-matched, no vias under the socket)	ground on inner layer 1, supply on inner layer 2, no vias in pads, impedance lines routed first	to prove on the dev kit
Moisture	Conformal coating Rain blows in at an open dock door, and the guards are hosed down and salted in winter. Asked for in the quote (lesson: moisture).	coated	coated, sensors, sockets and antennas masked	passes
Fit	Board inside the product CAD, every part checked in 3D The board is checked on its own; the rev T CAD run places it in the product and checks every part in 3D (lesson: fit3d).	not yet in the product CAD	nothing touches	to prove on the dev kit
Camera	Every door and dock call in its camera's view Every inView target from the CAD (far rear post, trailer floor 2 m in, leveler lip, a pallet in the door, a pallet face in the trailer; the leveler plate, a forklift, a person, a staged pallet) from the lens front's real place in the wall.	Door camera, guard A: worst 26.7° / 26.3°; Dock camera, guard A: worst 24.1° / 20°; Door camera, guard B: worst 26.7° / 26.3°	inside each lens's view (door 60 × 91.5°, dock 70 × 43°, guard B 60 × 75°)	passes
Camera	Camera holes in the steel faces The bench cameras sat behind Ø20 holes turned 35° and 30° from the face, so most of their view was the inside of the steel. Our lens barrel now passes through the hole to just inside the face, the lens is chosen from the calls, and the door chips are turned on their side (lesson: window). Change: CAD rev T: nose holes Ø20 → Ø44 (A) and Ø28 (B), dock hole Ø20 → Ø40, clear PC covers (4 Nov 2026).	door A needs Ø43, dock A Ø39, door B Ø28; steepest line of sight 67° from the face	rev T holes Ø44, Ø40 and Ø28 with flush clear PC covers; ≤ 72°	fixed by the review
Fit	Camera tabs on their brackets Each camera tab is cut from its strip and held on the camera's axis by a printed bracket on the wall; the brackets are drawn in the rev T CAD and checked in 3D there (lesson: fit3d).	Door camera, guard A: holder 0.3 mm off the wall, tab 6.53-12.03 mm off the strip, hinge 22.2 mm; Dock camera, guard A: holder 0.3 mm off the wall, tab 7.04-12.11 mm off the strip, hinge 22.3 mm; Door camera, guard B: holder 0.3 mm off the wall, tab 3.83-7.17 mm off the strip, hinge 19.4 mm	holder ≥ 0.3 mm off the wall, bracket ≥ 0.5 mm, hinge within the 25 mm flex	to prove on the dev kit
Sensors	Approach radar through its window The radar chip sits 4 mm behind the face, so the 50 × 20 PC window passes its whole view. TI quotes the view from antenna plots; the edge is measured at a real door on rev A.	views 33.9° / 3.1° and 52.8° / 1.2°; the ±70° × ±40° view needs 47.2 × 18.6 mm at the face	inside the chip's view; the 50 × 20 window	passes
Power	Mains on the board The dock door circuit is 120 V; only the certified supply and its terminal are on the mains side. The test lab checks creepage, the earth to the guard and the hi-pot on rev A.	certified PCB-mount supply (IEC/UL 62368-1, class II), T2A fuse and surge part, 6.4 mm creepage with a routed slot, push-in terminal behind the conduit entry	UL/IEC 62368-1 at the test lab	to prove on the dev kit
Fit	Lead between the guards The only cable in the product, because a pair of guards stands either side of the door; RS-485 and the sync wire ride in it beside 5 V.	Molex Micro-Fit 3.0 8-pin, latched and keyed at both ends; about 9 m of 18 AWG in the header conduit drops 0.11 V at guard B's 0.6 A peak	≤ 0.25 V drop; latched, keyed, no adapters	passes
Security	Cap tamper sensor The bench used a plunger switch at the cap; on our board the top edge of the strip is the closest point to the cap. The magnet pocket is added in the rev T CAD and the switch point measured on rev A.	DRV5032 Hall switch at each front strip's top edge, 7 mm under a Ø6 × 3 magnet bonded in the cap; the strike sensor also logs the cap coming off	cap opened = logged and signed	to prove on the dev kit
Shock	Forklift strike on the guard The guard is the part the trucks hit; the PAS 13 test is run on rev A with the boards logging the strike and calling it.	boards on stand-offs, cards latched, MEMS clocks, both strips held by brackets	PAS 13:2017 impact test with the boards running and logging	to prove on the dev kit
Fit	Flex lengths F2 carries guard A's door camera MIPI lanes from the nose to the brain; the flex maker confirms impedance over that length.	F1 55 mm (A) and 79 mm (B); F2 86 mm (A) and 165 mm (B)	routed and held in the rev T CAD	to prove on the dev kit

Before we order the first boards, we prove these on the RV1126B development kit or in the rev A tests:

AI engine time per frame
Power cut
Card grade
Wi-Fi from inside a steel guard
Guard B's radio (BLE back-up link)
Fast lines on a solid ground plane
Board inside the product CAD, every part checked in 3D
Camera tabs on their brackets
Mains on the board
Cap tamper sensor
Forklift strike on the guard
Flex lengths

Where the design stands

1Shape and placementdone
Main boards, front and nose strips, camera tabs and flexes in both guards’ shape, all 91 parts placed at their real size, every camera checked in 3D against the steel and its calls, and KiCad design rule check (0 violations).
2Schematicnext
Every wire between the chips, from the V1 core and C1 camera circuits.
3Footprintsnext
The exact solder pads under each chip.
4Routinglater
Copper tracks on all layers, the camera lanes through the flexes.
5Files for the makerlater
Gerbers with copper, drill, mask, paste, the rigid-flex drawing and the assembly parts list. These go to the board maker.
6Rev A boardsMar 2027
Made, assembled and tested on our fixture.

Design files (rev A-0)

These are the stage 1 files: the board's shape and where every part goes. A board maker needs the stage 5 files to build it.

All files (zip)Everything below in one download, with the quote and review KiCad boarddockone-board.kicad_pcb: open with KiCad 10 (free)3D model (STEP)The bare board for mechanical CAD Side A drawing (PDF)Down to the bay plate, out to the steel faces Side B drawing (PDF)Into the bay Parts and placement (CSV)Every part, its size, side and position

Who can make it

Each guard board is a main board and two strips joined by flexes, with camera tabs on short flex hinges; guard A is 8 layers with a fine-pitch ball-grid processor and a mains supply on the board. It needs a maker who builds rigid-flex boards and fine-pitch ball-grid chips, and a test lab for mains safety. We plan three quotes once the stage 5 files are ready: two from China for price, and one from the US for speed.

Maker	Where	Rigid-flex	HDI / layers	Assembly	Fit for us
PCBWayfirst quote	Shenzhen, China	yes, 6 to 16 layers	HDI up to 32 layers, 0.35 mm BGA pitch	yes, turnkey, incl. BGA and X-ray	Best fit for rev A: the rigid-flex board and its assembly in one order
NextPCBsecond quote	Shenzhen, China	yes	HDI, 6 to 32 layers	yes (PCBA)	Second quote, to compare price and lead time
ALLPCB	China (a network of partner factories)	yes, through partner factories	1 to 30 layers	yes	Third quote; check which factory builds it
JLCPCB	Shenzhen, China	no (flex only: 1, 2 and 4 layers, up to 490 mm)	HDI added Sep 2026	yes, incl. BGA	Not for the one-piece board today. Good for the test fixture, simple boards, and the fallback of separate rigid boards joined by soldered flex
Sierra Circuits (ProtoExpress)US quote	Sunnyvale, California, USA	yes	yes	yes, in-house, incl. BGA; turnkey in about 5 days	US option: fastest turn and engineering help, at a higher price
Cirexx	Santa Clara, California, USA	yes	yes	yes (turnkey or consigned)	Second US option
Advanced Circuits	Aurora, Colorado, USA	yes	check	yes	US quick-turn; check its rigid-flex assembly
Epec	New Bedford, Massachusetts, USA	yes	check	yes; also builds battery packs	US option that can also make a certified LiFePO4 pack

From each maker's own capability pages, checked 3 November 2026. Limits change, so each quote confirms them for our exact stack-up.

Full design notes

All 91 parts, what is on each board, every connection, heat, power, the factory test, cost and timeline.

DockOne: one circuit board of our own in each guard

DockOne is the pair of steel door-track guards at a dock door. Each guard gets one rigid-flex board of our own in its sealed top bay, with every chip soldered on it: no computer boards, camera cables, plugs or adapters inside. The only cable is the lead between the two guards, which a pair either side of a 2.7 m door cannot avoid; it is latched and keyed at both ends.

Board code: products/dockone/hardware/pcb/dockone-board.mjs (shape, every part, every check), board-3d.mjs (the 3D model on the page), board-review.mjs (the design review with the shared rules). Numbers marked est. are estimates until rev A.

1. The shape: three pieces in each guard

Each guard’s top bay has two spaces (cad/dockone-cad.mjs): the main bay over the 5 mm steel bay plate (inside x 1,466 to 1,640, y 5 to 198, z 995 to 1,226 under the cap) and the nose cavity beside the door-track notch (x 1,376 to 1,471, y 135 to 198). Each board is one piece:

Piece	Guard A	Guard B
Main board, flat on stand-offs over the bay plate	8 layers, 166 × 142 mm, 12 mm up: RK3576J (side A, heat block down to the bay plate), memory, eMMC; side B: the mains supply, its terminal and fuse, the lead header, security and ID chips, strike sensor, microSD	4 layers, 100 × 80 mm, 8 mm up: microSD, lead header, 5 V eFuse, RS-485, 3.3 V buck, security and ID chips, strike sensor
Front strip, standing behind the front face (into the building)	166 × 190 mm, 24 mm behind the face: 8 call-light LEDs and a light bar behind the front slot, sounder behind the tone holes, Wi-Fi 6/BT module behind the new antenna window, the dock camera on its tab, Hall tamper switch at the top edge	166 × 190 mm, 4 mm behind the face: the ESP32-S3 module behind the antenna window, the IWRL6432AOP radar chip behind the 50 × 20 window, SHT45 behind the vents, 8 call-light LEDs, sounder, Hall tamper switch
Nose strip, standing behind the nose face (into the doorway)	62 × 200 mm, 24 mm behind the face: 8 call-light LEDs behind the nose slot, the door camera on its tab	62 × 200 mm, 15 mm behind the face: 8 call-light LEDs, the OV2640 door camera on its tab
Flexes	F1 folds the front strip up from the main board’s front edge (55 mm); F2 runs from the main board’s left edge across and up to the nose strip (86 mm)	F1 79 mm, F2 165 mm

Camera tabs. Each camera chip sits on a 16 × 16 mm tab cut from its strip in a U-slot, on a 25 mm flex hinge, held on the camera’s axis by a printed bracket on the guard wall. The lens barrel passes through the hole in the steel to just inside the face, behind a flush clear cover.

2. What is on the boards

Brain (guard A): Rockchip RK3576J, industrial grade (−40 to 85 °C), 6 TOPS NPU, 16 MP image processor; 4 GB LPDDR4X; 32 GB eMMC; RK806S-5 power chip; MEMS clocks.
Cameras: door camera A and dock camera A are Sony IMX415 8 MP chips (C1) on 4-lane MIPI; door camera B is an OmniVision OV2640 2 MP chip on the ESP32-S3’s camera port, the second view for stereo.
Guard B core (S1): Espressif ESP32-S3-WROOM-1-N16R8, soldered by its edge pads; TI IWRL6432AOP 60 GHz radar with its rails, MEMS clock and flash; Sensirion SHT45.
Both guards: 16 call-light LEDs (8 in the nose, 8 in the front face), a magnetic sounder, an LSM6DSOX strike sensor, an ATECC608B security chip, an ID chip, a latched microSD card with our software, a DRV5032 Hall tamper switch under a magnet in the cap.

3. Every connection, checked

Connection	What	Check
Mains in	WAGO 2604-3103 push-in terminal (L, N, earth to the guard) on guard A, wires clamped, no plug, behind a conduit entry in the outer side face (rev T)	450 V rated; the earth goes to the steel guard; fuse and surge part before the supply
Mains supply	Mean Well IRM-20-5, PCB-mount, encapsulated, class II	soldered on the main board; 6.4 mm creepage with a routed slot round its mains pins
Lead A to B	Molex Micro-Fit 3.0 43045-0802 8-pin vertical header on each main board, positive latch, keyed; 5 V ×2, GND ×3, RS-485 A/B, frame sync	same pin-out both ends, checked in the board code; mating housing 43025-0800 with crimp terminals; 18 AWG for power
Cameras	the camera chips on their own tabs, joined by the board’s own flex	no camera cable; MIPI lanes impedance-controlled through F1/F2
Card	Amphenol 101-00660 class hinged, latched microSD lid	a push-push socket would eject on a forklift strike

4. Heat

The RK3576J (about 3.5 W working, est.) sits on side A of guard A’s main board, 12 mm over the 5 mm steel bay plate, with a 24 × 24 aluminium block and gap pads between: the steel guard is the heat sink. At 45 °C in the guard top (a summer dock with the doors open, est.) the chip runs at about 63 °C (5 K/W, est.). The camera chips are on their own tabs, 186 to 228 mm from the brain.

5. Power budget (est.)

State	Guard A	Guard B	Total
Door shut, after-hours watch	2.0 W	0.6 W	2.6 W
Working: 5 fps door, 2 fps dock, radar 10 Hz	4.5 W	1.0 W	5.5 W
A call: all call lights at full and both sounders	8.5 W	2.5 W	11 W

The IRM-20-5 gives 20 W, so the peak is 55 % of it. Guard B’s 0.6 A peak drops 0.11 V over about 9 m of 18 AWG in the header conduit. No cells: when the door circuit is off, the dock door is not working either.

6. Factory test (every board, every time)

Spring-pin fixture on the factory pads: current-limited first power-on (a short is found in about 1 s), rails, the RK3576J or ESP32-S3 boots our test image from the card, both cameras stream a test chart, the radar sees a corner reflector, the strike sensor and SHT45 read, every call light and sounder runs, the Hall switch sees a test magnet, the lead loops RS-485 and sync between a guard A and a guard B board, the security chip gets its key, the ID chip its serial. Mains is tested on guard A with an isolating transformer and a hi-pot test.

7. Who makes what

The boards, their assembly and test: a rigid-flex maker with fine-pitch ball-grid assembly (the same makers as our other boards). The guards, cap and brackets: our CAD, steel by a fabricator. The supply, lead and terminal: bought and fitted by the board maker.

8. Cost per door, electronics (est.)

Item	qty 10	qty 1,000
Guard A board (V2 core, two C1 cameras, mains supply)	$190	$95
Guard B board (S1 core, OV2640, radar)	$70	$32
Lead, terminal, covers, magnets	$25	$12
Per door	$285	$139

9. Timeline

Shape and placement now (rev A-0). Schematic and footprints next, from Rockchip’s RK3576 and Espressif’s ESP32-S3 reference designs; the rev T CAD draws the bigger camera holes, the antenna windows, the brackets and the cap magnet pocket; rev A boards with our other boards in early 2027.

10. Still to check

The RK3576J package outline, ball map and thermal figures from Rockchip’s datasheet (sizes are est.).
The IMX415 lenses: a 60° portrait M12 lens for the door, a 70° M12 for the dock, a 60° M7 for guard B, each with a barrel long enough to reach the face.
Radio range from behind the 60 × 10 antenna windows in steel guards, at a real dock (rev A).
The PAS 13 strike test with the boards running and logging.
Mains safety at a test lab (IEC/UL 62368-1), with the earth to the guard.
The window heater for freezer docks (on the bench design, about 2 W): its driver and pads go on the strips in rev T.

Sources

Rockchip RK3576 product brief: https://www.rock-chips.com/a/en/products/RK35_Series/2024/0328/1975.html
Mean Well IRM-20 datasheet: https://www.meanwell.com/Upload/PDF/IRM-20/IRM-20-SPEC.PDF
Molex Micro-Fit 3.0 connectors: https://www.molex.com/en-us/products/connectors/wire-to-board-connectors/micro-fit-connectors
WAGO 2604 PCB terminal blocks: https://www.wago.com/global/pcb-terminal-blocks-and-pluggable-connectors
Sony IMX415: https://www.sony-semicon.com/en/products/is/security/security.html
TI IWRL6432AOP: https://www.ti.com/product/IWRL6432AOP
TI DRV5032: https://www.ti.com/product/DRV5032
Espressif ESP32-S3-WROOM-1 datasheet: https://www.espressif.com/sites/default/files/documentation/esp32-s3-wroom-1_wroom-1u_datasheet_en.pdf
PAS 13:2017 barrier impact tests (BSI): https://knowledge.bsigroup.com/products/code-of-practice-for-safety-barriers-used-in-traffic-management-within-workplace-environments-with-test-methods-for-safety-barrier-impact-resilience
IEC 62368-1: https://webstore.iec.ch/en/publication/27412

Design review (4 Nov 2026)

The review uses the rules every one of our boards shares (products/_shared/board-rules.mjs) and is shown on the board page. Result: 0 fail; 1 fixed; the rest pass or are proved on rev A.

Fixed (new lesson “window”): the bench cameras sat behind Ø20 holes turned 35° and 30° from the steel face, so most of their view was the inside of the guard. A first try with a 100° lens needed a hole wider than the guard. Now each lens is chosen from its calls (door 60° portrait, dock 70°), the door chips are turned on their side, and the lens barrel passes through the wall to just inside the face: every call’s line of sight is within 67° of the face, and the holes become Ø44 (guard A nose), Ø28 (guard B nose) and Ø40 (dock), with flush clear covers. The rev T CAD draws them.
New in the shared tools: the RK3576 (V2 core) in board-rules.mjs, and board-quote.mjs now names the board’s own Rockchip core in the scope, so V2 boards get the right quote.
Fast lines on a solid ground plane (to prove, lesson: planes, from the CricketOne test-board layout of 4 Nov 2026): RK3576J memory; 3 camera MIPI links (100 Ω pairs); USB (90 Ω); antenna feed (50 Ω, keep-out on every layer); card lines (length-matched, no vias under the socket). Ground on the layer next to the parts (inner layer 1, supply on inner layer 2; the whole bottom of a 2-layer section), one via beside each ground and supply pad and none in pads, no copper or vias in the antenna keep-out on any layer or under the card socket, the impedance lines routed first with a test coupon. The quote asks for it; the routed board’s design rule check proves it.
To prove on rev A: AI time on the NPU, power-cut log, radio range from behind the antenna windows, the camera brackets in the rev T CAD, mains safety, the cap tamper switch point, the PAS 13 strike, the flex lengths.

New shared rule (4 Nov 2026, from PumpOne): antennas outside metal boxes. A radio inside a closed metal enclosure now fails the review unless its antenna is outside the metal. This board’s antennas already use the 60 × 10 PC antenna window in each guard’s front face, so the check passes and the review names the window (lesson: enclosure).