And for those who just love seeing silicon dies up and close, you might want to check out these amazing pictures to really put things into perspective:
Really cool to see the different designs used in various components, or even compared to other CPUs/MCUs. The pictures above were taken by dissolving off the inline package casing with sulfuric acid. No 8085, but there's a picture of the Z80A available, and you can make out the similar areas to the 8085 (register file, control logic, ALU, etc... the quality is insane, the Z80A picture is 6200x6000 px). There's also some earlier photos at http://zeptobars.ru/en/read/how-to-open-microchip-asic-what-... if you want some more high-res silicon imagery.
Nice post, Ken. I've been appreciating your work since xfractint in 1994 or so.
Note that in a wired-AND-capable current-mode logic, like open-collector TTL, you can get a similar RS latch out of a couple of inverters wired into a Henkin-sentence circuit like this, with the latch inputs being the latch outputs; but the behavior is slightly different than what Ken documents here. Here he says that, essentially, the 8085 tolerates bus-fight, but the bus drivers in the ALU are stronger than the bus drivers in the registers, and so after a brief bus fight, the register bit flips.
But with the wired-AND approach, bus fight doesn't exist; you just don't hook up any external zeroes to the Q and /Q lines if you want to read the latch, or to write it, you assert an external zero on one of them.
The trouble with open-collector logic, as I understand it (I've never built or debugged an open-collector circuit in my life) is that it's kind of slow. If I understand Ken's post, the circuit described here is a kind of optimization of the wired-AND latch I described above: there's a weak positive driver stage to speed up positive-going edges on the output. Is that right?
It's always interesting to see how real-world circuits depart from the purely digital, RTL ideal in order to save some resource. Jeri Ellsworth's work reverse-engineering old chips (which I hope to God is not secret!) found another example: dynamic stages in pipelines, where rather than using a flip-flop or even an entire latch, you just rely on the capacitance of an input to hold the previous value for long enough that the next stage in the pipeline can do its job. Works as long as you clock it fast enough, fails when you try to single-step the chip.
(Also, thank God, an actual hacking-related story on Hacker News, as opposed to "I lost 3lbs and a percent of body fat in a week" or "Every night when my dad would come home from work he would be filthy from working at the junkyard all day.")
Thanks for the detailed comment, Kragen. I didn't think anyone remembered xfractint :-) I think your analysis is correct, although I'll need to investigate Henkin-sentence circuits some more
You mention dynamic stages that rely on capacitance - these pass-transistor circuits are a standard technique, written up in books. The key is the nearly-infinite input impedance of MOS transistors, which lets the charge hang around for a while. Interestingly, the Z-80 doesn't use this, so its clock speed can be reduced down to DC, unlike the 6502, 8085, etc that have a minimum clock speed as you say.
Oh, a Henkin sentence is a sentence that asserts its own truth: "This sentence is true" — the opposite of the Liar's Paradox. You can self-consistently treat a Henkin sentence as either true or false: it's true if it's true, and false if it's false. A circuit analogue might be a buffer connected to its own input, or two inverters wired in the fashion you describe: "This sentence is not false."
This last week I've been using the nearly-infinite input impedance of MOS transistors to use LEDs as light sensors, as described in the Yerazunis's MERL paper. I'm not getting speeds anywhere close to the speeds they were reporting, though: closer to Hz than to MHz measurements.
If Xfractint is the "stone soup" fractal program then I remember. :)
Probably the first program I actually remember was a ratty ol' Mandlebrot fractal generator in Turbo Pascal 7.0 from probably around 1994-95 or so. Then I saw programs that were coded by people who actually knew what they were doing and my head asploded. :)
The "stone soup" fractal program is Fractint. Xfractint was my port of it to Unix/X, which was a bit tricky since parts of Fractint were in extremely tight 386 assembler that did crazy stuff like simulating floats with integers for performance.
It appears to be common nomenclature dating back to the 1960s-70s [1] to refer to the registers in a CPU as a single "file" in RAM, although the term "file" denotes the fact that you can access multiple parts of it at once.
From [2]: "A register file is a critical component of a modern superscalar processor. It has a large number of entries and read/write ports in order to enable high levels of instruction parallelism."
You might want to think of a register file as "a group of registers with several (parallel) input/output ports", as opposed to a component which can only perform a single read/write at a time. Basically, a really fast multi-channel RAM which can do simultaneous reads/writes. The number of read/write ports is usually dictated by your hardware, instruction set, parallelism, and scheduling [3] [4].
----
[1] Bollesen, V.; , "Core memories," Magnetics, IEEE Transactions on , vol.6, no.3, pp. 534- 535, Sep 1970
[2] Gonzalez, R.; Cristal, A.; Ortega, D.; Veidenbaum, A.; Valero, M.; , "A content aware integer register file organization," Computer Architecture, 2004. Proceedings. 31st Annual International Symposium on , vol., no., pp. 314- 324, 19-23 June 2004.
[3] Wallace, S.; Bagherzadeh, N.; , "A scalable register file architecture for dynamically scheduled processors," Parallel Architectures and Compilation Techniques, 1996., Proceedings of the 1996 Conference on , vol., no., pp.179-184, Oct 1996
[4] Balasubramonian, R.; Dwarkadas, S.; Albonesi, D.H.; , "Reducing the complexity of the register file in dynamic superscalar processors," Microarchitecture, 2001. MICRO-34. Proceedings. 34th ACM/IEEE International Symposium on , vol., no., pp. 237- 248, 1-5 Dec. 2001
That's an interesting question. Wikipedia doesn't say. [1] I did a brief Google Books search and found use of "register file" dating back to 1966. I assume it's related to filing-cabinet file, not file system file.
Etymology of the word "file" is interesting [2]. It comes from Middle French filer meaning 'string documents on a wire for preservation or reference'. That's surprisingly close to what a computer's registers do, right down to using wires :-)
http://zeptobars.ru/en/read/open-microchip-asic-what-inside-...
Really cool to see the different designs used in various components, or even compared to other CPUs/MCUs. The pictures above were taken by dissolving off the inline package casing with sulfuric acid. No 8085, but there's a picture of the Z80A available, and you can make out the similar areas to the 8085 (register file, control logic, ALU, etc... the quality is insane, the Z80A picture is 6200x6000 px). There's also some earlier photos at http://zeptobars.ru/en/read/how-to-open-microchip-asic-what-... if you want some more high-res silicon imagery.