Hardware first. Enough firmware to prove the hardware.

A list of skills can be written by anyone, so each group below points at the product that proves it. The work itself is in Projects, and eight of the boards can be opened in 3D.

01

Board design

Schematic capture and PCB layout, taken from a blank sheet to fabrication files. Power and protection circuits, switching supplies, galvanic isolation and creepage where a person can touch one side and the mains sits on the other, EMC and grounding, RF antenna layout and matching, low-power design, and layout that a factory can actually build.

EasyEDA ProMulti-sheet schematicsSMPS Reinforced isolationCreepage and clearance High-current polygon poursEMC and grounding RF matching networksLow-power design Design for manufacturing
PROVED BY

The three-phase energy meter holds a metering front end in full galvanic isolation from the host, with 8 mm creepage for reinforced class at 400 V line to line.

See the board
02

Embedded

Firmware written to exercise the hardware rather than to ship a product: bring-up, functional test, and the measurements that decide whether a design claim holds. Enough to find the fault, and enough to know when the fault is mine.

STM32 F4, G4, H5ESP32-C3 and S3nRF52832 CC++Python STM32CubeMXESP-IDFPlatformIO CMakeRTOS I²CSPIUARTBLEMQTT

Enabling the GPIO wake feature alongside an edge interrupt left the patient button drawing 18 mA instead of 0.4 mA. Removing it brought average current from 0.90 to 0.60 mA and battery life from 46 to 69 days.

See the board
03

Measurement and verification

The part I care about most. Checking how a number was produced before accepting what it says: whether the method suited the question, whether the reading was taken correctly, and whether the processing was right. That applies to my own results before anyone else's, and to data other people hand me.

OscilloscopeLogic analyserSpectrometer MultimeterCurrent measurement Sensor calibration against reference instruments Test plans traced to requirements Incoming acceptance criteria Reviewing methods and data produced by others Root-cause analysis

A proposed EMC fix measured 23.4 dB better, but at rest, in the condition that had already passed. In the one that had actually failed it changed nothing, so it was rejected. The product passed retest with 9.35 dB of margin.

See the board
04

Release and documentation

A design only counts when somebody else can build it. Fabrication files, the documents that justify them, and the coordination with the supplier and the fabrication house that turns a folder into a crate of boards.

GerberPick-and-placeBill of materials Design rule checkingWiring diagrams Verification documentsClause-by-clause gap analysis Supplier and fabrication house coordination

The incubator mainboard went out as Gerber data, pick-and-place coordinates, bill of materials, wiring diagrams and the verification documents behind them, design rule check clean.

See the board

Standards I design against

Not a reading list. Each of these has changed something in a board I drew, or decided whether a product could be released.

IEC 60601-1
Basic safety for medical equipment. Isolation, creepage, touch current.
IEC 60601-1-2
Electromagnetic compatibility. The clause behind the conducted emission retest.
IEC 60601-2-19
Baby incubators. Independent thermal cut-out, alarm thresholds, reconnection.
CISPR 11
Emission limits, Class A Group 1 for the operating table.
ISO 14971
Risk management. Hazard, cause, protection, verification, gap.
SNI equivalents
The Indonesian adoptions that national regulatory testing is run against.
The boards behind all of this

Five products, fourteen boards, and eight of them open in a 3D viewer where you can rotate them and look at the real component placement.

See the projects