Most complicated error-free chip

As I tear my hair out dealing with errata on the chips I'm using, I find myself wondering what is the most complicated chip that's been designed that was error free from the time it was first release to the market?

Sylvia.

Reply to
Sylvia Else
Loading thread data ...

555 timer? I know many chips have part numbers of xxxA and xxxB which makes you think there must have been at least one before, no?
--

Rick C
Reply to
rickman

You are apparently judging chips only from a digital chip POV.

Over the past 55+ years I've churned out several hundred Analog chips... only 2 out of that amount did I @#$% up. ...Jim Thompson

--
| James E.Thompson                                 |    mens     | 
| Analog Innovations                               |     et      | 
 Click to see the full signature
Reply to
Jim Thompson

And more than half of that count were done BC (before CAD and simulators). ...Jim Thompson

--
| James E.Thompson                                 |    mens     | 
| Analog Innovations                               |     et      | 
 Click to see the full signature
Reply to
Jim Thompson

Probably the VIPER processor.

formatting link

The UK Royal Radar Establishment claimed that the chip had been designed us ing a formalism that allowed each stage of the development to be mathematic ally tested for errors and mathematically proven to be error free. I don't know what approach they used - the one I've encountered (but never came clo se to using) was

formatting link

No software guy I've run into take it seriously, but most of them were tran sparently cowboys, and the rest put their faith in simpler rules.

--
Bill Sloman, Sydney
Reply to
bill.sloman

How can any system be used to design error free chips when humans are involved. The North Anna nuclear power plant was supposed to have no single point of failure. When the earthquake shook the plant into shutdown one of the diesel generators failed to be replaced by a backup.

When they looked into the cause of the failure they found a head gasket had been installed incorrectly because the procedure for installation was faulty. There's your single point of failure. If the procedure is faulty, it can cause *every* generator to fail.

Humans were involved, so mistakes were made.

--

Rick C
Reply to
rickman

I believe their reset input is historically broken, tho... Does it still count as error-free?

Reply to
Johann Klammer

Yep, the 555 timer's reset line is bugged by the standards of some; it won't interface "correctly" like a logic device:

formatting link

Reply to
bitrex

When the Inmos "transputer" chip was designed (I don't remember which particular chip), the design was independently checked by a formal mathematical proof done by a university team, and by exhaustive checking of every combination for all instructions (using large numbers of chips running in parallel for half a year or so).

The two checks found exactly the same set of errors, which were then corrected.

Of course, there might have been problems or errata on issues other than the logical operation of the cpu.

Reply to
David Brown

LCF-LSM and ELLA. Messrs Cullyer and Pyggot.

formatting link

Reply to
Tom Gardner

David Brown :

Mathematical proofs are much like program code. In fact, it resembles assembly language code that is run in the human brain. Mistakes are easy to make.

Of course, automated proof checking alleviates this risk quite a bit, but you can still make mistakes in your modeling.

Now, automated proof checking *is* awe-inspiring: . No steps are skipped. For example, proving "2 + 2 = 4" involves 2,452 subtheorems: . And look at this gem: !

Marko

Reply to
Marko Rauhamaa

No, they are not like program code or assembly, and they do not "run in the human brain" like on a computer.

Mistakes can certainly be made. It is also possible to write proofs without making mistakes. In this particular case, the same design issues were spotted by both those doing the mathematical verification, and those doing the exhaustive testing of the hardware.

True.

That all depends on the kind of proofs you are looking for, what definitions you are using, what axioms you are using, and so on. In particular, this example is using complex numbers - the proof of 2 + 2 =

4 in natural numbers is far simpler. The example is also used to write things in the way that generates the most impressive numbers, with every sub-theorem expanded fully at every step, rather than the way proofs are /really/ done by proving each sub-theorem only once.
Reply to
David Brown

Even with proofs, you need to be sure that all the requirements are expressly stated so that they can be proved.

One annoying bug in the PIC32MX processors is that instructions that perform writes will sometimes do the write twice when interrupted - once before the interrupt occurs, and once after it returns. Microchip suggest that this is only a problem with certain peripherals, but it seems to me that program code could be sensitive to this behaviour, and consequently that it might cause a program failure.

The requirement is that each write operation is performed exactly once, but how sure are we that that anyone would think to include it, absent this bug mentioned in the errata.

Sylvia.

Reply to
Sylvia Else

David Brown :

No, Metamath doesn't expand subtheorems fully. It refers to them by name.

The way proofs are ordinarily done, though, at least in all student textbooks, is by doing a lot of hand-waving and expecting the reader to connect the dots. Metamath connects all the dots for you.

Marko

Reply to
Marko Rauhamaa

I'm not sure how entirely error-free it was, but the first ARM processor worked first time:

formatting link
formatting link

The thing that bothers me about Microchip's errata is that some bugs stay unchanged in the errata even when they have seemingly done at least one all-layer tape-out after they knew about the bug. In my opinion, any fault that they are not going to even *try* to fix in an all-layer tape-out should be described fully and honestly in the datasheet rather than being hidden in the errata, otherwise it borders on false advertising. Write endurance of the EEPROM being much lower than the datasheet says is one of the "bugs" that I am thinking about.

Reply to
Chris Jones

I once repaired a stereo amplifier for a friend. It was a transistorized unit from the 70's. The schematic showed an op amp. It also showed an insert of the op-amp itself which consisted or two transistors and a few resistors etc>

It was a potted device a roughly the size of a postage stamp and maybe

1/4" think>

It had failed and I could have just replaced it with two transistors, but instead used a 741. Supposedly a 741 was is not a good choice for audio, but it seemed to work fine.

Kind of shocked that I found it on my first Google hit

formatting link

Reply to
philo

Image here:

formatting link

Reply to
philo

It depends on how much gain you expect. The 741 was a big step forward from previous op-amps, no need for an external frequency compensation capacitor and other specs better. But if you expected much gain, the response tapered off fairly fast, I forget exact details. But everyone used them, though even in the seventies some were building their own op-amps out of discrete transistors. It didn't take that long for better op-amps to arrive, around 1977 or 78 there were cheap and better op-amps, but I think people got set in their ways, so the 741 is still available.

Michael

Reply to
Michael Black

A significant number of (many Sydney-based) "software guy"s take formal proof *very* seriously; they have produced the first ever machine-checked proof of an entire microkernel operating system; including verification that the generated ARM machine code matches the C source. But apparently that can be ignored, because it doesn't fit your narrative.

Clifford Heath.

Reply to
Clifford Heath

Nothing sane fits Slowman's narrative. ...Jim Thompson

--
| James E.Thompson                                 |    mens     | 
| Analog Innovations                               |     et      | 
 Click to see the full signature
Reply to
Jim Thompson

ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.