A few years back some really smart guys figured out how to make optical diodes. That means optical computers, that means really really fast computers. They said the problem was the the size of the diode was on the order 400nm as opposed to todays transistors being around 40nm or so. What did I think? Screw that! I'll take a 1 THz 6502 over all these bloated multi-core pipeliney chips any day of the week!
I hadn't heard about super fast optical transistors, but a few months ago I was reading about ultra fast transistors that operate at 600GHz.
The idea came to me, to write a x86 emulator in 6502 assembly. Run it on a 6502 CPU built using ultra fast transistors, and boom, you've got a faster x86, just it's an emulated x86, MMU and all!
Typical 6502 instructions take 2-8 cycles if I remember rightly, so as long as it takes under 200 6502 cycles to emulate one x86 instruction, you're beating a 4GHz x86 processor.
I'm sure there are practical problems that make it difficult or impossible in reality. But still fun to think about. :)
- 3 8bit registers vs. ~8 32bit or ~20 64bit registers. (so you're essentially working on memory at all times, and have to emulate all "longer" commands, such as addition, to do 4-8 carries manually)
- 64k of memory (which includes the emulator), so you need to add some banking mechanism.
- no support for integer multiplication/division or even float/SIMD in your instruction set.
But you probably can provide a fast implementation of the decimal addition instruction :-)
