Back in the mid-late '70's, the 8080/8085 and Z80 along with the MC6801/02, etc, were pretty popular, but along came the MOS-Technology guys with their 6501 and, later, 6502, and "bought" a huge portion of market share by virtue of their low price.

That was possible because the chip size, which, ultimately amounts to selling silicon by the pound, was said to be between 1/4 and 1/3 the size of their competitors. Maybe that architecture, for which there are still useable cross-development tools lying about, would be a good choice for soft-core development.

It seems to me that an ALU and a 16-byte RAM is sufficient to generate that entire core. Careful examination of the instruction set reveals that no instruction operates on more than one register during any given clock phase, of which there are two. An architecture implementation using the same ALU to perform register arithmetic and address arithmetic is easily implemented when arbitrated by its two clock phases.

It seems more practical to me than a 4004, though I guess a really useful application of a 4004 soft-core in today's FPGA technology would be a fun thing to do. The boss would fire you, though, if he knew that's what you'd done.

RE