Harvard architecture
Instructions in one memory, data in another — a split Aiken did not choose on principle but because punched tape was all he had.
The Harvard Mark I read its instructions from punched paper tape and held its numbers in counter wheels. The two lived in separate organs and were reached over separate paths, not as a stance but because 1944 offered no alternative. A year later the EDVAC report put program and data in the same store, and that became the norm. The phrase "Harvard architecture" was only coined in the 1970s, by people who needed a name for the thing that wasn't von Neumann.
The split buys two things: an instruction and a data word can be fetched in the same cycle, and program memory can be read-only and thus harder to shoot through. The cost is that a program cannot modify itself and memory cannot be reallocated — if the chip has 32 KB of flash and 2 KB of RAM, that is the ratio it lives with, whatever the program would have preferred.
Which is why the microcontroller is a Harvard machine: AVR, PIC and ARM Cortex-M all take separate paths to flash and RAM, as do signal processors. And nearly every modern CPU is "modified Harvard" — split instruction and data caches nearest the core, a unified main memory behind them. The division Aiken could not avoid turned out to be right exactly where speed is decided.