Rabbit board

Body board rev C: netlist, placement, routing

Status 2026-10-11: netlist and two-sided placement done (floorplan of 2026-10-11 in gen/placement.json); routing on 8 layers in progress (autorouter plateau 27-29 unrouted connections, the rest is being finished by hand).

• The approved spec is docs/reports/2026-10-10-body-pcb-rev-c.md (Russian). Its §11a decisions and the "10-11 October" section override older data.
• The per-IC source tables are data/pcb-revc/ics/*.json (gitignored); reviews and the applied fixes are in data/pcb-revc/review/.
• The generator gen/design.py is the single source of truth for parts, nets and default positions; gen/placement.json (263 overrides, applied at the end of design.py) moves parts to the 2026-10-11 floorplan.
• History: docs/log/2026-10-10-body-pcb-rev-c.md.

python3 gen/footprints.py                       # custom footprints -> lib/rabbit_body.pretty
python3 gen/outline_revc.py                     # outline.json + the corner ToF poses in tof.json (from kit_outline.json)
PY=/Applications/KiCad/KiCad.app/Contents/Frameworks/Python.framework/Versions/3.9/bin/python3
$PY gen/build_board.py --no-route               # placement only (without --no-route: + planes and pours, gen/routing.py)
gen/make.sh                                     # full run: build, fill, fastroute/Freerouting, cleanup, DRC (SES=... reuses a route)
python3 gen/check_netlist.py [--board]          # netlist checks (--board needs $PY)
$PY gen/drc_summary.py                          # kicad-cli DRC -> fab/drc.json
$PY gen/power_check.py                          # copper width and vias of the high-current nets after routing
python3 gen/stock.py [boards]                   # JLC stock of every BOM line (default 2 boards) -> fab/stock.json + csv
PLOTPY=<python with matplotlib> gen/render.sh   # renders/top, bottom (seen from below), iso, *_labelled

Board

• Outline (gen/outline_revc.py -> gen/outline.json): X -70..70, Y -158..156.2 (board frame: origin at the deck centre, +Y = robot front, top view). The kit plate grown on 2026-10-11 (owner): the tail goes to Y -158 and the board is ±70 wide from the tail to Y 30 (9.8 mm to the rear wheels' inner faces); the front keeps the kit's waist for the steering sweep; the nose goes to Y 156.2 so the ZED 2i (43.1 mm deep, front face on the edge) clears the front deck-3 standoffs by 2 mm. The four corner chamfers are turned square to the ToF brackets: front ±50°, rear ±135°, rear chamfers 30 mm long.
• Holes: 4 × M4 deck standoffs (±38.09, -121.89 / 107.64), 2 × M3 (±55.26, 84.95), 2 × M3 for the ZED 2i on 5 mm standoffs (±18, 134.9; holes 36 mm apart), 6 ToF bracket footprints ToF_Bracket_M2.5 (two M2.5 holes, flange and screw-head keep-outs), the ZED USB cable cutout 11.1 × 18 mm.
• Stack (gen/routing.py, gen/autoroute.py): 8 layers. Signals on F.Cu, In2.Cu, In3.Cu and B.Cu; In1 and In6 are GND planes with a GND_MOT island under the motor block (joins GND only at the star tie NT1 next to the battery connector); In4 is one VBUS plane over the whole board; In5 is a split power plane: MOT_BUS over the drive, SERVO_RAW and SERVO_PWR over the servo/arm block, +3V3_MCU under the MCU, +5V everywhere else. The router treats In1/In4/In5/In6 as power layers (vias only). The battery chain (3 mm) and the 5-10 A nets (2 mm) run on the outer layers only; 2 oz outer copper is assumed. Local pours that would trap other nets' pads are filled after routing.
• 667 footprints: 411 on the top, 256 on the bottom. 613 assembled placements (361 top, 252 bottom; 589 SMT, 24 THT), 46 mechanical or copper-only, 1 module (the Pi), 7 DNP. 159 BOM lines (unique LCSC numbers), 47 ICs, 433 nets.
• DNP: the PH/EN MODE 0R R1106/R1206; the second ITRIP line Q1102/Q1202 and R1151; the IMU CLKIN link R423/R424. Everything else, including the MMC5983MA and the bottom THT XT30s, is assembled by JLC (owner, 2026-10-10).
• Stock (gen/stock.py, 2026-10-11, 2 boards): all 159 lines in JLC stock with at least 3× the need, so one JLC order with no Global Sourcing. Thinnest: Würth 2×20 C7433390 (89), PCA9685 (157), RT0603 549 Ω 0.1 % (224), STM32G474 (248), TPS259470A (331), EEHZK1E471P (404), DRV8243 (408).

Blocks and placement (after placement.json)

Layer rules:

• Tall parts only on the top: inductors, polymers, THT headers.
• check_netlist.py checks bottom heights: at most 8 mm over the motors (Y -60..-130), at most 10 mm elsewhere; nothing on the top under the Pi.
• Floorplan rules (data/pcb-revc/review/floorplan-2026-10-11.md): ≥ 2 mm free ring around 0.5 mm QFNs, 3 mm around the LQFP-64, no logic-GND parts over GND_MOT.

Renders: renders/top_labelled.png, renders/bottom_labelled.png (seen from below, so mirrored), renders/top.png, renders/bottom.png, renders/iso.png; re-render after routing.

What changed against the first rev C netlist (10-11 October)

• Vertical connectors only (owner): USB-C J50/J53 GCT USB4115-03-C, CAN GH-4 JST BM04B-GHS-TBT, ToF GH-6 HCTL HC-GH-6PLT (C3011957, GH-compatible on the JST BM06B land; genuine JST had 1 in JLC stock), XT30 AMASS XT30UPB-F on the bottom, lidar JST B5B-XH-A.
• ToF: 6 ports, one per printed bracket under deck 2 (front, front ±50°, rear, rear ±135°; gen/tof.json, field of view checked in cad/robot/tof_fov.py), all on +3V3_TOF (the TPS62826 EN is the group switch); the TCA9548A is alone on the Pi's I2C6 (GPIO22/23, J3 pins 15/16, 2.2 k pull-ups); mux channels 6/7 and TCA6416A P06/P07/P16/P17 are tied off.
• Simplified (owner, 2026-10-11): no per-port ToF eFuses and no ADS7128 0x10; one eFuse U78 for both encoders; INA238 only on battery 0x40, motor bus 0x41, servo 0x43, 5 V total 0x47. 763 parts before, 667 now.
• xArm 1S bus on the board: TPS259470L U88 (active current limit, R_ILM 750 Ω = 4.45 A) from SERVO_RAW, OVLO on the steering eFuse's divider (61.9k/10k = 8.63 V); STM32 PB9 USART3_TX AF7 single-wire half-duplex -> BSS138 -> 5 V bus (390 Ω 1206, TPD1E10B09), 2 × PH-3; ARM_PWR bleed 4.7 kΩ. The round-1 TLV6700 cut and the TPS259824L (0 at JLC) are gone.
• MCU: MCU_ALIVE PC3 (pin 11), PB9 = arm UART, PA6 = USB VBUS detect, PA4 reads DRVOFF through 10 kΩ, nSLEEP pull-downs 4.7 kΩ, IMU CLKIN held low by 100 kΩ.
• Pi power through USB-C J53: U87 is the auto-retry TPS259470A (R_ILM 887 Ω = 3.76 A), EN 100k/100k from +5V (2.6 V), not on the expander; Rp 10 kΩ (3.0 A); ribbon pins 2/4 open.
• Polarity: XT60 J1 and XT30 J20/J21 pad 1 = GND at the chamfered end ('-'), pad 2 = '+' (round 1 F-1/F-2).
• MOT_BUS_OFF (0x22 P2_4) pulled up to +3V3_MCU, not +3V3_PI: the motor bus stays off when the Pi is dead or unplugged (round 1 P-4).
• C26 (button debounce) to BAT_F: plugging the battery in no longer latches the board on (round 1 P-2).
• Series resistors to the Pi (E-stop, UART0 RX, lidar UARTs) 100 Ω -> 1 kΩ (back-powering).
• Board grown and re-floorplanned for routing; ZED 2i on two M3 standoffs on the board.

Kept from rev B

• The LM74800 switch and latch, F1, the battery path, NT1 (star ground) and the brake chopper.
• The LM61495 buck block.
• The STM32 core: crystal, decoupling, BOOT0/NRST from the Pi, LDO.
• The E-stop NAND/ALIVE logic driving DRVOFF.
• RTC, button and bumpers.

Removed against rev B

• DRV8316, the BLDC phase, Hall and encoder connectors; the RoboClaw fallback (JP1, JP2, J48, J49).
• TPS73733, the rev B ToF wiring and the Jetson UART; the INA226s; the DRV8316 footprint.

Custom footprints (gen/footprints.py)

PCAL6524 uses the KiCad QFN-32-1EP_5x5mm_P0.5mm_EP3.1x3.1mm, close to but not the NXP HUQFN32 land.

Deviations from the JSON files and substitutions

1. Battery shunt. Vishay WSLP25122L000FEA (2512, 2 mΩ, 3 W, C2090981) instead of the WSLP2726: same value, fits the rev B Kelvin land.
2. ITRIP default 2.41 A. Series ladder 8.25k / 13.7k / 25.5k; ITRIP_SEL is pulled up (R1150 to +3V3_MCU), so 2.41 A applies by default and in MCU reset; the STM32 drives PB11 low for 4.05 A; with R1111/R1211 removed and PB11 low, 4.73 A. drv8243.json synced on 2026-10-11.
3. Motor bus off by default. MOT_BUS_OFF (0x22 P2_4) has its pull-up (RN array) to +3V3_MCU; Q31 holds the LM5069 UVLO low until the Pi drives the line low.
4. Expander reset. EXP_RESET_N (GPIO6, 10 kΩ pull-down) resets both PCAL6524s and the TCA6416A: every switched load stays off until the Pi drives it high. The TCA9548A has its own reset (GPIO19, 10 kΩ pull-up).
5. STM32 PA6 is USB_VBUS_DET (22k/33k). TPS2121 ST goes to a test point (no MCU pin left).
6. ILM readback through 10 kΩ at the eFuse, 1 nF only at the ADS7128 input (§11a).
7. Pi USB-C eFuse U87 (Q47 -> C32): TPS259470A auto-retry, R_ILM 887 Ω; 0x23 is full, so its FLT and current go to ADS7128 0x17 AIN4 (as GPIO, pulled up to ADC_AVDD) and AIN3. The arm eFuse U88 uses AIN5 (current) and AIN6 (FLT) the same way.
8. ToF rail enable from 0x22 P2_5 with a pull-down; PG to 0x23 P2_6.
9. TCAN1044AV VCC from +5V_MCU, so CAN works on USB power alone. CAN connector pin 1 = +5V_AUX (spare 5 V eFuse).
10. SK9822 supply straight from +5V.
11. Servo bulk 2 × Panasonic EEHZK1E471P (470 µF 25 V, C242138).
12. TPS16630. Jetson: MODE = GND (auto-retry), 5.1k = 3.53 A, SHDN through Q7 from GPIO25. Spare output: latch-off, 4.7k = 3.83 A, SHDN 10 kΩ pull-down. Their currents come from the TPS16630 IMON pins (ADS7128 0x17 AIN1/AIN2); the separate INA238 0x44/0x45 and shunts are gone.
13. Fan PWM through a 2N7002K (fans pull PWM up to 5 V): inverted in firmware.
14. LM5069 freewheel diode B540C on the GND_MOT island at the bulk capacitors.

Open issues and what could not be verified

Tracked in docs/open-issues.md (Q49 layers, M25-M27 and M29-M30 cable and connector checks, P24-P28). In short:

• Mechanical: room for the mated PH plug in the gap between the motors; XT30 tails and pigtail bend at the left edge (M27); HCTL GH-6 mating with genuine GHR housings (M29).
• Footprints: DRV8243 mask polygons approximated; TPS62826, TPS2121, SK9822 pads from drawing labels; XT30 drill 1.8 mm vs AMASS 1.9 mm; no 3D models for the new footprints; rabbit-body-c.kicad_dru still names the old USB4105 for J50.
• Pinouts: TPD4E05U06 pins from memory; PESD2CANFD24V unchecked; pin 1 of the kit motor cable, the C1 lidar cable and the xArm cable to be checked with a meter; XT polarity on the moulded housings.
• Electrical: DRV8243 thermals assume 2 oz outer copper and via-stitched pads; ILM readback below 0.5 A unspecified; BSC010N04LS SOA margin under the LM5069 small; ITRIP live switching with a FET ladder untested; +5V worst case 5.30 V vs the C1's 5.2 V (P26); SERVO_RAW unclamped after a buck high-side short (P25); brake resistors 4 × their rating while on (P24); PCA9685 can phantom-power unpowered servos (P27, firmware).
• Stale text in the generator (not netlist): design.py docstring and comments (9 INA238, 8 ToF, ADS7128 ×3, "+5V17", "56k Rp", PCAL INT on GPIO22, VBUS sense on PC3), the bottom silk "4L 1.6 mm" and the routing.py docstring ("In5 signals") (P28).