Voyager: Thank you. That stuff is precisely what I was asking about.
So, it’s that automated these days? I suppose I shouldn’t be surprised; these are, after all, the most complex things humans have ever mass-produced (am I wrong in this statement?).
(And, if tape is no longer used, how do we currently apply the design to the die at the foundry? And here I thought I understood the sand-bending part.)
The fdiv bug wasn’t that subtle - someone forgot a few line of a PLA, and the verification processes didn’t catch it.
And it is that automated. I was shocked a few years ago when a good designer asked me to show him how to navigate net lists. Design for test people do this all the time, since test stuff is mostly inserted at the gate level, but most designers work on timing and optimizing paths, but not hacking gates.
Yes it is. But if you want to UNDERSTAND the process, you’re better off imagining yourself circa 1978, when so much of this was done by hand. That, you can comprehend. The automatation obscures the process.
Agreed. Automation is a great tool for doing something you understand fully more efficiently, not for learning how it’s actually done.
Plus, in 1978 chips were a whole lot simpler. I think I could just about comprehend the Intel 4004 or MOSTech 6502, but I doubt I could grok the Pentium M inside my laptop right now.
I don’t have much of a compsci background, but I agree that the best way to start understanding how microprocessors work is to look at how they used to work before they became so complicated. This book – Atari Roots, published in 1984 – was a good start for me. (There are some other books about assembly language on this site.) It explains how an old microprocessor (the 6502) works, in the context of programming in assembler. But not in enough detail to understand how someone would go about designing a microprocessor out of silicon, or even out of logic gates.
There are various open source microprocessor designs floating around. Go to opencores.org and look around. Sun has a Community Source License version of two old Sparc processors. You have to pay if you use them commercially, but they are free to download, after signing a license. But these tell you about the architecture of processors, and there is a lot more happening now.
It looks like we hit the wall on making processors faster architecturally. The new way is putting lots of processors on one chip.
Hennessy and Patterson is still the standard text, I believe.
