Yeah, that's basically doing the whole thing using specialised "lisp circuitry". Approaching the task from the other extreme, by just implementing the interpret

Yeah, that's basically doing the whole thing using specialised "lisp circuitry". Approaching the task from the other extreme, by just implementing the interpreter in microcode, would be much simpler - and I doubt that the end result would differ wildly in the number of gates used. I think this is basically a kind of minimal description length problem. And yes, you can do a crap job with anything ;) Edit: Btw. LISP 1.5 for the IBM 704 seems to be something like 10k lines of assembly code.