While waiting for new shift registers to arrive I resumed working on configuration tool. It is being written Haxe which is in my opinion an somewhat underrated programming language.
Reason for making a new keyboard configuration tool is that both electrical matrix and just amount of keys is more complicated than typical DIY keyboards which often have simple rectangular matrix with rows and columns going straight. It would be much nicer to define everything in a tool and have export scripts generate the required configurations and mappings than doing it manually. it specially when one of the middle layers change requiring to redo everything. The representations involved are:
In the previous log entry I described that hardware was more or less working and I could have continued with writing firmware. That was 2 years ago and there has been little progress. So I decided to change approach for something that will require less effort and produce usable result sooner.
One possibility could be replacing atmega8 with something that is pin compatible with more memory and use QMK. It seems that QMK support isn't great, at least it's not their primary platform, so I am not keen on this idea. If am moving away from minimalist approach and trying to squeeze too much in functionality in mcu that is too small for that I might as well go with overkill and use something more powerful.
First idea was using teensy 3.5 or 3.6. Turns out ChibiOS support for them is somewhat incomplete. I am also not a fan of proprietary bootloader chip which prevents debugging. Printf debugging isn't an option if you can't initialize the hardware properly. I might return to this once I have a working keyboard.
Together with teensy I had ordered some Blue Pill boards. So this seemed like next best thing. It doesn't have enough pins for controlling the keyboard matrix directly but I can use the shift registers that I planned to use in Initial iteration. Next problem is that TPIC6C595 I had only operates at 5v. After a few hours of searching catalogs I found NPIC6C596ADJ. I could probably make a level shifter from transistors I have laying around but I will need to solder a new board anyway and there are other parts of software that can be worked that I can work in while the new parts arrive.
First version of PCB was a failure. I didn't properly consider use of regular think the use of regular shift registers with push-pull outputs for driving rows. Only after the PCB was soldered I understood that pressing two buttons on the keyboard will cause rows to short-circuit. Not wanting to redo everything I tried to come up with a workaround, that resulted in more mistakes.
I tried to reduce the short-circuit current by covering contacts with slightly conductive ink made of glue and graphite. That bumped the resistance to 0.5K-2K . In combination with input scanning being done only a fraction of time that should have solved the excessive current problem. After observing that pressing two buttons in the same column still doesn't work I understood that I have made a voltage divider. There was no way of avoiding to redo the PCB.
This time I decided to order PCB from China instead of etching at home. I was getting mixed results with printing the transparent masks. My laser printer really didn't like the material and printing at copy shops were either too transparent or had lines more than half the time.
After soldering everything I was ready to start working on firmware. After loading the last test program it was recognized in the USB device list. I was starting to think that I have solved most of the hardware bugs. After starting to experiment with reading keyboard input results didn't look correct. Printing the number of pressed buttons on LEDs showed completely wrong result so I decided to print the whole matrix.
Interesting, pressing a single button resulted in streak. I guess the pull-up resistors built in atmega are too small for capacity of keyboard. That could be solved by adding some delay. It required 10-20uS delay for each row. Having 24 rows that seemed too much. I could switch the inputs to outputs for a cycle charge up the lines faster. Resetting shift register outputs would also take a while. Unfortunately due to lack of outputs for microcontroller and to simplify the layout I had chosen to connect the shift register disable line to ground. It would have been useful to have them now so that I can temporary disable their outputs without flashing the register state. One of the traces to disable pin was going under the chip so I couldn't cut and add a bodge wire. Not wanting to redo everything again I decided to take the risk and leave it as is.