75% ISO Hall-Effect Keyboard PCB
The board design for a 75% ISO Hall-effect keyboard built around an STM32 microcontroller. Every key is read as an analog value rather than a switch closure, which allows adjustable actuation depth per key. Designed in KiCad with an emphasis on power integrity, signal integrity, and manufacturability.
- Stack
- STM32KiCadPCB DesignADCUSBSWDI2CSolidWorks
- Highlights
- Per-key analog sensing through analog multiplexers into the STM32's ADC
- USB, SWD debug/programming header, and an I2C OLED header
- Separate analog and digital power distribution with ferrite isolation
- Decoupling and ESD protection on every external-facing line
- Addressable RGB, rotary encoder, and a case designed in SolidWorks
The problem
A mechanical switch is a binary sensor: pressed or not. A Hall-effect switch reports a magnetic field strength that varies continuously with travel — which means the keyboard knows how far down each key is, and actuation depth becomes a per-key setting rather than a property of the switch you soldered in.
Getting that requires a board that can read an analog value from every key, cleanly, fast enough that no one would perceive latency. This project is that board: the schematic, layout, and case, designed to be fabricated.
Approach
Getting 80+ analog signals into one microcontroller. An STM32 doesn't have an ADC channel per key, so the Hall sensors feed banks of analog multiplexers. The STM32 drives the mux select lines and samples each mux output on an ADC input, so the whole board can be scanned key by key. The ADC routing from each mux to the microcontroller was laid out as its own concern, short and away from anything switching.
Analog integrity first. The signal being measured is a small voltage varying with magnet distance, on a board that also carries addressable RGB switching current at every key. Getting a clean reading is a layout problem before it is ever a firmware problem — ground pours, return paths, and keeping the analog sense lines away from everything noisy.
Separate analog and digital power. Per-key RGB plus an OLED plus an STM32 doing continuous ADC sampling adds up, and LED switching noise on the sensor supply shows up directly as jitter in the readings. The analog and digital rails are distributed separately, with ferrite beads isolating them and decoupling at every IC. Designing the power tree properly at the schematic stage is much cheaper than discovering brownouts on assembled hardware.
Designed to be debugged and plugged in. USB for the host connection, an SWD header for programming and on-chip debugging, an I2C header for the OLED, and ESD protection on the lines a user's hands and cables actually touch.
Manufacturability as a constraint, not an afterthought. Component choices, footprint decisions, and assembly tolerances were made against what can actually be fabricated and hand-assembled — a board that only works in CAD is not a board.
Case designed alongside the PCB. Mounting points, port cutouts, and the plate were modelled in SolidWorks against the same board outline, so board revisions and case revisions stayed in step.
What I got out of it
A complete board design, and a much better feel for the parts of embedded work that have nothing to do with code. Firmware bugs you can iterate on in seconds; a board respin costs weeks. That asymmetry changes how carefully you read a datasheet.