I finaly fixed my sodering station so I could finish the Ethernet on the board below as this was critical to test before I ordered my new design. And a good thing I did because Ethernet did not work straight up at first.
I ported Ethernet using a Nucleo-144 board with the same MCI in 144 pin configuration – this helps because you then work on verified hardware as you port the full stack. Once that was working I moved the code to my board and nothing worked – on the contrary my SWD connection went unstable, and Ethernet was dead – that was two days ago. My first aproach then stuck is to test openai for ideas or hints based on what I see – it pointed me to the 50Mhz clock that was not working, but gave me a lot of bullshit so I abandoned the AI aproach.
I have copied the reference schematics – which is a good start because it means lots of otheres have worked on this as well – and reading up I discovered that someone else had a similar problem because they had forgotten a pull-down on a LED pin. Looking at my own I recognized the same error, so with a bit of creative soldering art I added a 10K pull-down and the 50Mhz clock came up.
It turns out that this LED pin have a purpose at power up so the LAN8742 clock does not start unless the pin is down. I think the signal maybe should be square, but I will dig into that later. Seeing 50Mhz means that clock is working and 25Mhz crystal is working fine, but still no Ethernet.
Checking settings on my nRST signal I see that is low and never started – voila – once that is fixed Ethernet lights up and run fine. But, my SWD is still down and ustable – I have a hard time loading code or doing any work. And my laptop is whining ping-pong all the time.
Accidentally I start charging my iPhone and notice that gives a pingt-pong sound on/off as well. Needless to say SDW adapter is USB as well, so after a short reorganization of my USB network SWD also works fine. I admit I was so focused on Ethernet that I incorrectly assumed everything was related to me adding the Ethernet circuit on the board.
This is the cost of doing this for the very first time. I copied an Ethernet circuit I did not 100% understand before, but guess what – now it works – not bad for6-7 hours work. Soldering a QFN package was easier than expected – TBH this is my 3rd as I have started soldering more of these by hand. But, soldering up 34 components by hand is a lot of work.