I Applied to the DigiKey Make ONE Challenge 2026 and Didn't Make It — What Was in My Submitted Development Log

Overview

This is a maker contest hosted by electronics distributor DigiKey. Entries were submitted by registering a project on ProtoPedia and filling out a Google Form. Any award would have led to an exhibition at Maker Faire Tokyo 2026.

The result: eliminated in the first round.

What I Submitted

I compiled the entire development log for the ReLow60 L-HE into a ProtoPedia project page — covering the full process of designing and building everything solo: circuits, metal fabrication drawings, firmware, and configuration tool.

I understood this contest to be asking “what did you build and how,” so I kept philosophical discussion to a minimum and focused on the process of measuring, comparing, and deciding. Below are excerpts from the key sections of the submission.


Archetype09 — Starting with a 9-Key Proof-of-Concept

Jumping straight to a 60-key PCB carried too much risk. I first designed a 9-key tester board called Archetype09 and started with sensor and bypass capacitor selection.

Comparing Three Hall Sensors

I mounted three Hall sensors — DRV5055A3 (Texas Instruments), MT9102ET (MagnTek), and GH39FKSW (GH) — on the same board and ran 8 tests covering noise, sensitivity, linearity, thermal drift, crosstalk, dynamic response, settling time, and scan rate.

Distance–ADC response comparison of three Hall sensors
Distance–ADC response of the three Hall sensors. The GH39FKSW shows inverse polarity: ADC decreases on keypress.

The results showed that MT9102ET was overwhelmingly superior in thermal drift: ±9.2 LSB vs. ±33.8 LSB for DRV5055A3 and ±48.2 LSB for GH39FKSW.

Thermal drift comparison of three Hall sensors
MT9102ET thermal drift is 1/4 to 1/5 that of the others — the most critical metric for keyboard sensor selection.

The MT9102ET is inferior in some noise and sensitivity metrics, but I chose it because keyboards experience temperature changes during use, making a small baseline shift the most practically significant factor. I selected based on actual use conditions, not raw catalog specifications.

Comparing Bypass Capacitors

A bypass capacitor is connected to the Hall sensor output pin to suppress high-frequency noise. Together with the sensor’s output impedance, this capacitor forms a low-pass filter that attenuates MUX switching noise and MCU clock noise. However, too large a capacitance can degrade the sensor’s response speed.

I therefore compared noise and settling time across four conditions: 0 nF (none), 1 nF, 2.2 nF, and 4.7 nF.

Noise comparison by bypass capacitor value
At 2.2 nF, maximum σ decreased from 2.33 to 1.72 — equivalent to 4.7 nF but with lower cost and smaller footprint.

At 2.2 nF, noise was reduced by approximately 20%, achieving the same effect as 4.7 nF with no measurable impact on settling time (theoretical RC time constant τ = 10 Ω × 2.2 nF = 22 ns, well below the ADC conversion time of ~700 ns). I adopted 2.2 nF.

Difficulties Encountered with the Tester Board

The Archetype09 development itself was far from smooth. I documented everything transparently.

  • All 5 boards failed to boot — The MCU (AT32F405) could not exit reset after power-on; USB and SWD were both unresponsive. After days of investigation and suspecting chip defects, a senior colleague pointed out that the NRST pull-up resistor was connected to GND instead of 3.3V — a schematic error.
  • OLED not displaying — The AT32F405 has no Alternate Function mapped to the I2C SDA pin (PB5). Worked around temporarily with bit-bang I2C, later rerouted with a jumper wire to hardware I2C.
  • MUX channel mapping offset — Missed the fact that the 4051’s channel numbers and pin numbers are non-sequential.

Every one of these failures informed the design of the ReLow60 main board.


Main Board — Scaling Up to ~70 Sensors

Using the knowledge gained from Archetype09, I designed the main board. It aggregates approximately 70 Hall sensors into the MCU’s ADC via 8 MUXes (SN74LV4051A-Q1).

The PCB is a 2-layer design with a GND pour on the back. Noise mitigation measures include separation of analog and digital signal traces, impedance-controlled USB differential pairs, and copper-free zones directly above sensors (to prevent copper from interfering with the Hall effect).

Scaling up introduced new challenges that were invisible at 9 keys: MUX switching settling time, crosstalk between adjacent channels, and loop time when scanning all keys simultaneously. Settling was resolved by adjusting the ADC sampling delay from 5 μs to 8 μs; crosstalk remained within spec at a maximum of 4 LSB.

The submission also covers the sheet-metal case design, firmware (a fork of libhmk), ReConf, bugs fixed through real-world testing, and distribution at Gaming Bazaar.


Reflections After the Rejection

Since I was eliminated in the first round, I received no comments from the judges. I don’t know the reason.

That said, the ProtoPedia page itself is now the best-organized record I have of my own development work. Despite the rejection, I’ve submitted it as a reference document in subsequent applications and used it to explain “what I’ve built” in conversations related to an AI grant program.

What I wrote for the contest has ended up doing work outside of it. Nothing I wrote was wasted.