I am not arguing about the existence of Hardvard arch. on MCU's but rather on the fact that most of the time, RAM and ROM memory accesses happen in the same add

I am not arguing about the existence of Hardvard arch. on MCU's but rather on the fact that most of the time, RAM and ROM memory accesses happen in the same address space. For sure, the underlying HW might connect you to different buses, but on a lot of MCU's, from the programmer's perspective, the address space is a big flat area containing everything. For sure, my experience has probably tainted my outlook. I have the most experience with Renesas MCU's and for example on them, I can surely run programs straight from the RAM, and depending what you're doing, you might sometimes be forced to do so. I worked with Cortex M also, and they're flat address space too. I did check out the specs for Atmega MCU's and it seems those, as you described have different address spaces for data and code areas. What I want to say, I guess is that you can have a huge variety of options and architectural approaches for MCU's.