Custom EtherCAT device

you sure you are not messing up ms und µs somewhere? 6MB/s is 4 full (1500bytes) ethernet frames per millisecond.

Yes I'm sure. I used a timer to measure in microseconds and I got about 1560. Like I said the library is very basic. And it does writes for every byte. Currently I'm using HAL in STM32Cube IDE and those SPI calls are a little bloated. For now I'm trying to interface with an Analog servo driver from SanyoDenki, so I'm not working on optimizations right now

I didn't get a chance to review all the specifications, but here is A launchxl board, this should chop the forrest a bit for you

www.digikey.com/en/products/detail/texas...CHXL-F28069M/5138373

Thanks, as far as I can see the TMS320F28069 does not have ethercat. It only has the standard stuff like uart, spi, can..
The ones I was able to dig seem to be out of stock.
Its always nice to know there are options.

Non of the MCU has EtherCAT support via hardware you always need to use an ASIC. Tipically it connects via SPI or QSPI. With an Application Specific Integrated Circuit you can use the MCU for exchanging the data with the master and soing other stuff whil minimizing latency and jitter on the bus itself. Because the ASIC is responsible to recieve the packets on one port and while extracting data it also forwards the packet to the output port for the next slave.
I succesfully made the slave and interfaced it with an analog servo.
I setup the machine so one turn of the motor represents 4mm of movement eith this I achived 14m/min speed vith 0.1mm error while moving. This was dons only with FF1 and P tunning. If I spet more time on this I could lower the error. The movement is smooth without any problems

would it be possible to use the EtherCAT LAN9252 without a MCU and have the I/O directly at the chip like in the Datasheet "Digital I/O Mode"?
This would provide 16 very fast I/O at minimal cost. (10$ for the chip, maybe 20$ for the PCB and other components)
There is also the very similar LAN9254 with 32 I/O pins, for a 12$ per chip.
Directly output the Step/Dir signal from the LAN chip into the stepper drivers (obviously some opto-couplers and such in between).
I think this would be a elegant solution, as all the software would be on the LCNC side and no additional latency added.