What the designer remembers
One of the people who drew this chip, checked against it.
Every other page here reads the die. This one reads a person. Bill Mensch co-designed the 6502 and described, four decades later, what went into it and why. Some of what he says can be asked of the silicon directly, so it is, one claim at a time. The most interesting answer is not agreement.
Loading the chip…
Written, not measured
The interview
続きを読む
In August 2015, Kevin Savetz recorded an interview with Bill Mensch for ANTIC, a podcast about the Atari 8-bit computers. Mensch is one of the four people named as inventors of the 6502, alongside Chuck Peddle, Rod Orgill and Wil Mathys. He had come from Motorola, where he had worked on the 6800 and its peripheral chips, as part of the group that left to build a cheaper processor at MOS Technology. He has spent the years since running the company that still sells the design.
This is a different kind of source from the talk on the other page, and the difference matters for how it is read. A reverse engineer describes an artefact, and if the description is wrong the artefact will say so. A designer describes intent, from memory, about decisions taken forty years earlier. Intent cannot be checked at all. What can be checked is the handful of places where he attaches a number or a structure to it.
So this page does not attempt to grade him. It asks the small number of his statements that are answerable, and the answers turn out to divide cleanly: the structural claims are right, and the one quantity is not.
Written, not measured
The clock that made it a 6502
続きを読む
The story he tells is about the difference between two part numbers. The 6501 was built to drop into a 6800 socket. The 6800 took its two clock phases from a separate generator chip, which Motorola sold alongside it, and Mensch's argument was that this was a liability rather than a feature: the generator was manufactured on a different process from the processor it drove, so a slow generator could be paired with a fast part and the two phases could overlap. Overlapping phases in a chip that stores its state on the wires between them is not a performance problem, it is a wrong answer.
His proposal was to build the generator into the processor. He recalls arguing that it would cost nothing worth counting: it would not make the die bigger, because the transistors could be hidden under the metal that was already routed over that area, and it would not change the price. The part with the generator on it was the 6502. Motorola then sued over the socket-compatible 6501 and it stopped being sold, which settled a bet he had made with Rod Orgill about which of the two would sell more.
Two things in that are answerable. He says the change was worth about a dozen transistors, and he says the point of it was to guarantee the phases never overlap. The second is a claim about the shape of a circuit, and this die has that circuit on it.
Measured, not written
The claims, re-asked of the chip
Each row pairs what he said with what this simulation answers when the same question is put to it. Nothing below is written down as a verdict: every row carries a function that re-derives the answer from the published measurements, and the verdict is the comparison. If the chip stops agreeing, the page changes on its own.
Derived from the switch network
What the clock generator actually is
Nothing on this site had drawn this circuit before, so it is worth setting out what the walk finds. Starting at the clk0 pad and following the wiring forward, the generator is … nodes carrying … transistors, which is …% of the chip.
It is two symmetric halves. The input pad drives a pair of inverters, and each of those feeds a chain that ends in one of the four clocks: two used inside the chip, and two driven back out of the package on pins so the rest of the system can be built around them. Most of the transistor count is not logic at all. Only … of them decide anything; the remaining … are the four output stages, wide banks of transistors in parallel that exist to shift the load. Driving a clock is mostly a question of strength: the internal one alone opens … switches across the chip.
The interlock is the part worth looking at. There are exactly … transistors that carry a finished clock back into the generator, and each lands in one of the two halves. That feedback is the non-overlap: each phase is held off until the other has gone. The guarantee Mensch describes as the reason for the entire change is those two devices, and it is a guarantee about the chip rather than about the board it is soldered to, which is the whole point of moving it inside.
The count is derived by a rule rather than a list, which is what makes it a measurement: walk forward from the pad, and include the four clocks without expanding them. Expanding them is not an option, incidentally, which is why the rule has to say so. The chip's own boundary is the answer to where the generator stops.
What the difference is worth
Being wrong about the number and right about the argument
The one row that differs is worth more than the four that agree, and it is not a gotcha. A designer recalling a change from four decades earlier is off by roughly a factor of two on the transistor count, and is exactly right about everything the count was being used to support.
His argument was never really that the number was twelve. It was that the cost was negligible, that it would not grow the die, and that it bought a guarantee no external part could offer. All three survive the measurement: the circuit is …% of the chip, and the guarantee is there in the silicon as the feedback that makes the phases exclusive. What is off is the figure attached to the argument, which is the part a person is least likely to carry accurately and the only part anybody could check.
There is also a limit on this page that no amount of care removes. There is no 6501 on this die. Nothing here can measure a difference between two chips, only the size of the circuit on the one that survives, so the row is worded as that and its verdict means only that the circuit is larger than the recollection. If the two parts differed in some other way as well, this page would have no way of knowing.
Credit
Sources
続きを読む
The interview is Kevin Savetz's, recorded for ANTIC, and the recollections are Bill Mensch's. This page is our own reading of it and our own measurements against it. Where the two disagree the disagreement is on the page rather than smoothed away, and where his account cannot be checked at all we have said so rather than implying otherwise.
The die trace is by Greg James, Brian Silverman, Barry Silverman and Ed Spittles, and is licensed CC BY-NC-SA 3.0. Those terms carry to everything on this site.