I've been hacking on this cute little 8031 board that I picked up for $1 ... In spite of it being a Harvard CPU architecture (separate data and code spaces-- you can't download code into RAM and run it..) there is nonetheless a common hack of merging the EPROM and RAM together so that you can get around the limitation.
"With not use NVRAM?"
Googles... finds /no/ mentions of people doing that. Is there a reason why?