LoadShed

LoadShed hardware

KiCad 9 project for the LoadShed PCB. Licensed under CERN-OHL-P v2.

Files

loadshed.kicad_pro / .kicad_sch / .kicad_pcb   KiCad 9 project
loadshed-schematic.pdf                         Schematic export (browse without KiCad)
sym-lib-table / fp-lib-table                   Project library tables (see JLCImport note)
fabrication-toolkit-options.json               Fabrication Toolkit settings (JLCPCB)
production/
  gerbers_v1.2.zip                             Gerbers as fabricated (v1.2)
  bom.csv                                      BOM with LCSC part numbers
  positions.csv / designators.csv              Pick-and-place files
  netlist.ipc                                  IPC netlist
bom/ibom.html                                  Interactive HTML BOM

Component libraries (JLCImport)

Several parts (regulators, inductors, polyfuses, battery holder, buttons…) were pulled from JLCPCB/LCSC using the kicad-jlcimport plugin, which converts EasyEDA CAD data into a project-local JLCImport library. Because that converted data comes from EasyEDA/LCSC, its redistribution licence is unclear, so the JLCImport symbol/footprint/3D-model libraries are not included in this repo.

This matters less than it sounds:

• KiCad embeds copies of every used symbol in the .kicad_sch and every footprint in the .kicad_pcb, so the project opens, edits, and fabricates fine without the libraries. Expect harmless "library not found" / "footprint not found in library" warnings in ERC/DRC, and missing 3D models for those parts in the 3D viewer.
• To restore the libraries (for placing new instances or 3D rendering): install JLC Import from KiCad's Plugin and Content Manager, open this project, and import each LCSC part number from production/bom.csv. The plugin recreates JLCImport.kicad_sym, JLCImport.pretty/ and JLCImport.3dshapes/ in the project directory, where the committed sym-lib-table / fp-lib-table already point. (easyeda2kicad is a CLI alternative that does the same conversion.)

Design overview

• Input: 12–32 V DC screw terminal, master fuse (F1), power switch. Sized for 12 V or 24 V battery banks; silkscreen says LiFePO₄ because that's what it was built for, but chemistry doesn't matter; set the voltage thresholds in the config tool to suit.
• 12 V load rail: AP64502 synchronous buck (rated 5 A; run here at 3.5 A). All four output channels switch loads powered from this rail, so loads see a constant 12 V regardless of battery voltage. This regulator is the hottest part on the board (the silkscreen says so).
• 3.3 V logic rail: AP63203 buck for the MCU, RTC, sensor and OLED.
• Outputs: 4 × AO3400A N-channel MOSFETs, low-side switching, gate driven directly from the MCU (3.3 V logic-level), 100 kΩ pull-downs, 100 Ω gate resistors, 3 A polyfuse and indicator LED per channel. Screw terminals carry +12 V and the switched return.
• MCU: ATtiny3216 (SOIC-20), UPDI on a 3-pin header, 20 MHz internal, no crystal needed.
• RTC: DS3231MZ (SOIC-8, integrated MEMS resonator, ±5 ppm) with CR2032 backup holder.
• Sensor: SHT4x on the I2C bus.
• Front panel: 4 push buttons, 4 pots (3.3 V–GND, wiper to ADC), header for the 128×32 SSD1306 OLED.
• Sensing: battery and 12 V rail via 100k/27k dividers into the ADC.

Input voltage

The silkscreen says LiFePO₄ IN (12–32 V). The technical ceiling is set by the smaller regulator: the AP63203 3.3 V buck accepts 3.8–32 V (the AP64502 12 V buck is good to 40 V). Two caveats above ~15 V:

• The battery-sense divider (100k/27k) full-scales at ≈ 15.5 V, so voltage readout and battery-protection thresholds only work meaningfully on a 12 V bank as built. For 24 V, change the divider ratio and update VBAT_DIVIDER in the firmware's config.h.
• The indicator LEDs use simple series resistors sized for 12 V-ish rails. Check their dissipation before running high input voltages continuously.

Known shortcomings (v1.2)

Candid notes for anyone considering a respin:

• No relay channels. All four outputs are MOSFET/PWM. One or two dry-contact relays would have covered loads that don't share ground or need galvanic isolation.
• No output clamping. There are no on-board flyback diodes or TVS on the MOSFET drains, so external flyback diodes are mandatory for inductive loads, including external relay coils (fit them at the load). A respin should add clamping per channel.
• SOT-23 MOSFETs. The AO3400A is fine at a couple of amps but has no thermal headroom for sustained high-current PWM into long cable runs.
• No reverse-polarity protection. The input fuse covers overcurrent only. A reversed battery would push a large fault current through the regulators' internal body diodes; the fuse should open quickly, but the regulators surviving the event is not guaranteed (and hasn't been tested). Connect it correctly.

For the record, v1.2 fits all of its designer's needs and there was never a reason for a second spin. A v2 wishlist, should anyone take one on: one or two dry-contact relay channels, per-channel clamping, input reverse-polarity protection, and selectable output sourcing (unregulated battery through the MOSFETs, or additional regulated rails such as 5 V routable to a channel).

Fabrication

Costs, assembly details and the offer of at-cost boards live in the main README. The mechanics:

• Production files in production/ were generated with the Fabrication Toolkit KiCad plugin (settings committed in fabrication-toolkit-options.json).
• bom.csv carries LCSC part numbers for every placed part; JLCPCB economic assembly places all SMD parts from it.
• Hand-solder after assembly: screw terminals, buttons, pots, headers, power switch, OLED. Nothing fine-pitch.
• Open bom/ibom.html in a browser for an interactive placement view.