Hello Guillaume,
NPR is currently not well designed for "backbone-point-to-point" topologies. I know it. OK, I could maybe add features later, but I have lots of work to do before. One question for you : on which condition do you think we could "wake-up" part of an IPv4 or Ethernet network? Protocols above the L2 or L3 (VoIP for example) are usually very chatty, and keep sending packets, even when no real user activity.
The question of how to wake up an L2 connection is an interesting one. My first thought would be to possibly send Ethernet frames with different flags, etc. I don't have one of your kits (yet) so I'm speculating a bit here but if the master has a list of all 7 clients and their MACs, you can begin to make it begin to act like an Ethernet switch. If a remote station is "registered" with the master (aka.. hard coded in), the master could track the last time it was heard from. If you have non-broadcast traffic for that remote station and it hasn't been recently heard, you could send these alternative "wake up" Ethernet frames. The remote client would have a setting that would allow for remote wake-up (default on) and if it heard these frames, it would wake up and talk to the master using regular IP Ethernet frames. The question here is if the STM32 boards expose some of these alternative Ethernet frames for you to make decisions on.
You might be mistaken. The L432KC is 80MHz RAM_64kB Flash_256kB, whereas the F303K8 is 72MHz RAM_16kB Flash_64kB which is way too low for the NPR firmware. The L432KC is really the most powerfull for this form-factor, inside the STM32 offer.
Hmmm.. ok. I'm not that familar with the STM32 line so I apologize. I had been looking at some data sheets that showed some versions with more RAM, etc.
--David KI6ZHD