Post the worst datasheet lies you've caught

[berke]

I propose 3 categories:

Category A - Flat lie - Device simply doesn't match the datasheet (example: goes poof at 5V even though supply range is said to be 3.3 to 6V)
Category B - Not a straight story - Datasheet is inconsistent, example it says 3.3V to 6V supply in the front page, but in footnote 3 of table 6. page 8 Vcc max is actually 5.3V
Category C - Lawyer talk - Datasheet makes you think something, but only upon very close examination it appears that that's not the case, or that some other parameter degrades so much that the device becomes useless.  Example: Sure our SPI port can run at 50 MHz!  (Yes well that's for output only, if you want to read MISO keep f < 27 MHz.)  Or sure the op-amp is rail-to-rail!  (But PSRR goes to trash.)

[berke]

For this first example, I think it's category C.

The iMX8M reference manual says this:

Upon which I was expecting SPI operation at 52 MHz.  But yes you can transmit SPI data at 52 MHz, so it's technically correct.

A separate document, the datasheet says this:

Looks like if you want to read MISO you better stay below 23 MHz.

[Psi]

Category B
I forget the datasheet, but it had text saying something like
"Pin X must be set low for correct operation"
and then it had a connection diagram that had pin X connected high. 

Also Category B i guess.
Another bad one is some of the ST LISxxxx accelerometers.
So much missing info or info from other accelerometers chips that doesn't apply to the one your reading.
Lots of head scratching involved and reading datasheets from other variants just so you can understand missing info from the chip you are using.

[schmitt trigger]

There is also a category D: one in which important operating characteristics are left out. Chinese data sheets are masters in this respect.

[tom66]

Category A, I guess.

I forget the exact device but I once designed an MCU into a product which had three SPI controllers advertised in the datasheet.

In the errata, there was this gem:
"Revision A1, A2.  Due to a silicon design error, SPI3 does not function.  Workaround: none."

So we ended up having to respin the board to connect to SPI2 instead.  Lesson learned...

[berke]

So we ended up having to respin the board to connect to SPI2 instead.  Lesson learned...

And that, kidsmanagers, is why we make prototypes and don't rush into production.

[joeqsmith]

Time stamped to NI datasheet for one of their DAS products:
https://youtu.be/TAsdr4ypBko?t=721

Over the years, I've encountered several problems with brand new ICs where the companies have altered their datasheets as a result.  In one case, company removed the IC from the market.   One IC I used,  I think the company revised the datasheet three times before I was done.  If someone unfamiliar with the part used it today, they would have no idea that several modes that the hardware supports because they are no longer documented.   

I've lost track of how many companies were bought out, and the company who takes them over strips details from the original datasheets, I assume because their marketing group doesn't understand engineering and the need for such details.   

Another category I have ran into many times is firmware group reporting problems that were a result of their lack of understanding.   A good example, we designed in a new micro which supported JTAG.  My plan was to use JTAG to test the hardware and program the peripherals during production.  Our firmware group determined JTAG did not work.  They raised the flag, generating knee jerk responses from mid management.   Me, not trusting anyone I work with, write my own code and make some custom hardware  to utilize the JTAG.  Worked fine.

[berke]

Over the years, I've encountered several problems with brand new ICs where the companies have altered their datasheets as a result.  In one case, company removed the IC from the market.   One IC I used,  I think the company revised the datasheet three times before I was done.  If someone unfamiliar with the part used it today, they would have no idea that several modes that the hardware supports because they are no longer documented.   

Were those modes terminally useless?  Maybe someone could have used them for something?

[joeqsmith]

Over the years, I've encountered several problems with brand new ICs where the companies have altered their datasheets as a result.  In one case, company removed the IC from the market.   One IC I used,  I think the company revised the datasheet three times before I was done.  If someone unfamiliar with the part used it today, they would have no idea that several modes that the hardware supports because they are no longer documented.   

Were those modes terminally useless?  Maybe someone could have used them for something?

In that particular case, I am using these undocumented modes for products we have in current production.  The problem from the IC manufacture's perspective is they would have had to provide details about the cases I uncovered in their datasheet.  While I would have thought errata sheets, I am guessing they just took the easy route and stripped the features all together.  Maybe I shook their confidence.   

From my experience, the story normally goes something like, I find a problem.  Contact company.  Company states I am an idiot and they can't replicate.   I provide bad parts.  Company eventually puts me in touch with higher level support who states they can't replicate.   In the case of Analog Devices, I get a call one day from the designer of the part.  Also get a call from their factory who provides me with details on how to screen for the problem.   They correct the IC in the next product line and they also add changes to their factory testing of the ICs.    It's always the same, time wasted dealing with the apps engineers, sales..... and finding the people who actually know what is going on.   

In the case of previous part, I had found more than one problem. Once I worked the channels the first time,  finding other root problems went smoother.  They knew who I was and I had a bit more clout.   Life of a designer.   

Anymore, I don't do much with new parts so I dare say most of what I run into it the company take over, marketing/sales messing up the new datasheets.   

[HalFET]

Change of footprint in every minor revision of the datasheet, courtesy of ST. Even their own distributor couldn't get a straight answer out of them.

[JPortici]

From my experience, the story normally goes something like, I find a problem.  Contact company.  Company states I am an idiot and they can't replicate.   I provide bad parts.  Company eventually puts me in touch with higher level support who states they can't replicate.   In the case of Analog Devices, I get a call one day from the designer of the part.  Also get a call from their factory who provides me with details on how to screen for the problem.   They correct the IC in the next product line and they also add changes to their factory testing of the ICs.    It's always the same, time wasted dealing with the apps engineers, sales..... and finding the people who actually know what is going on.   

In the case of previous part, I had found more than one problem. Once I worked the channels the first time,  finding other root problems went smoother.  They knew who I was and I had a bit more clout.   Life of a designer.   

can't say i can blame them for this approach In the firmware world, do you know how many "compiler bugs" which are actually "lack of understanding in how to write even simple statements" support receives? I have reported actual compiler bugs a number of times and the process is always the same: get the useless wall of text from first level support, have them suggest i don't knwo what i'm talking about when i reply that they sent me is completely unrelated, please reread the request and let me try with simpler words if you still don't understand, then second level asks a couple of more info that were already in the first text, then i get messaged from a member of the compiler team that is able to reproduce the problem with all the data that was there in the first post

usually the middle step is skipped if i previously opened a thread in the company's forum

at least i don't get sales droids to insult me asking how many units of X i plan to buy anymore, it is I who should be given a reel of parts instead whenever i find problems (sometimes really nasty ones) that they didn't manage to catch 

[vk6zgo]

Over the years, I've encountered several problems with brand new ICs where the companies have altered their datasheets as a result.  In one case, company removed the IC from the market.   One IC I used,  I think the company revised the datasheet three times before I was done.  If someone unfamiliar with the part used it today, they would have no idea that several modes that the hardware supports because they are no longer documented.   

Were those modes terminally useless?  Maybe someone could have used them for something?

In that particular case, I am using these undocumented modes for products we have in current production.  The problem from the IC manufacture's perspective is they would have had to provide details about the cases I uncovered in their datasheet.  While I would have thought errata sheets, I am guessing they just took the easy route and stripped the features all together.  Maybe I shook their confidence.   

From my experience, the story normally goes something like, I find a problem.  Contact company.  Company states I am an idiot and they can't replicate.   I provide bad parts.  Company eventually puts me in touch with higher level support who states they can't replicate.   In the case of Analog Devices, I get a call one day from the designer of the part.  Also get a call from their factory who provides me with details on how to screen for the problem.   They correct the IC in the next product line and they also add changes to their factory testing of the ICs.    It's always the same, time wasted dealing with the apps engineers, sales..... and finding the people who actually know what is going on.   

In the case of previous part, I had found more than one problem. Once I worked the channels the first time,  finding other root problems went smoother.  They knew who I was and I had a bit more clout.   Life of a designer.   

Anymore, I don't do much with new parts so I dare say most of what I run into it the company take over, marketing/sales messing up the new datasheets.

I've moaned on this forum before about the 4000 series monostables we used at my old work in (luckily) just a couple of transmitter site remote controls, built "in house".
It's too long ago to remember the exact type number, but we used the devices to provide a quite long time delay, which was well within the range quoted in the data sheet.

We had to perform a modification which required messing with the wire wrap connections to the socket holding the device, so I carefully removed the IC & promptly lost it into the "black hole" which seems to surround workbenches.

No problem, plenty in the store!
Big problem!---- the time delay function didn't work properly with the new one!

Stealing a device from another part of the equipment, all was well, so the next trick was to check every device of that type number in the store, it was found that none of them worked in that circuit.

Digging through a ton of datasheets, we eventually found, on a loose, daggy looking errata sheet that some people had been having trouble achieving short delays with the devices, so the manufacturers modified the device to work well at such delays, buggering up its performance for long delays.

To "add insult to injury" they retained the original part number for the modified version & relabeled the original version with a new type number.
We just bought a bunch of the relabeled ones, changed our records/schematics & were "good to go" as we only had two of those remote control units in service, but imagine if someone had thousands of them in production or "out in the field!

[srb1954]

Category A
I was using a 30ns multiplier-accumulator chip in a system running at a 32ns cycle. However, this chip occasionally spat out incorrect calculations for some multiplicand values. After a lot of time, may be 40-50 hours, investigating and attempting to replicate the worst case condition I eventually determined that the chip was not meeting its specified speed rating and was only capable of returning correct calculations in every case when operated at a 38ns cycle time or slower.

Reported this to the chip distributor who admitted that their factory had known for some time that their test system set-up was not catching some marginal chips off the production line. They refunded our money for the bad chips but we were still out of pocket for the engineering time lost investigating their bad chips.

[David Hess]

This one falls under category A.

Long ago Texas Instruments released a new integrating analog-to-digital converter, and one of its advertised features, which was suppose to be a great improvement over earlier models, was a true single count zero instead of positive and negative zero, which would require external logic to fix.

So we designed and built new boards with this improved converter and ... it still reported positive and negative zero.  We contacted Texas Instruments and confirmed our results, but as far as I know, they never changed the datasheets and never advertised the problem.

[joeqsmith]

I've moaned on this forum before about the 4000 series monostables we used at my old work in ...

It's too long ago to remember the exact type number, but we used the devices to provide a quite long time delay,...

Funny, we also used some of these old 4000 series CMOS parts for timers as well in the minute range.   buffered, unbuffered, brand, all made a difference for leakage.  We were pretty limited what we could use.  Seems like we found Motorola UBCP worked well.   We didn't want to pay extra for the higher temp parts and qualified with the cheapest.   Most parameters we didn't care about.   The caps were also a problem.    Cheap, crap design but worked alright.     

[berke]

From my experience, the story normally goes something like, I find a problem.  Contact company.  Company states I am an idiot and they can't replicate....

They refunded our money for the bad chips but we were still out of pocket for the engineering time lost investigating their bad chips.

So we designed and built new boards with this improved converter and ... it still reported positive and negative zero.  We contacted Texas Instruments and confirmed our results, but as far as I know, they never changed the datasheets and never advertised the problem.

A pattern seems to be emerging.

1. Find issue
2. Triple-confirm, have someone else look over it and prepare nice write-up with data
3. Send mini-report to manufacturer.  To customers this kind of report would be sent with a five-digit invoice, but for the gods of big IC this is an offering.  Secretly expect a handwritten thanks note from the CTO and a box of Champagne within a week.
4. Instead, quickly get a very short response from illiterate tech support with the equivalent of "Have you tried turning it on and off?".
5. Manage to write a polite response, include supplementary material.
6. Level 1 support recognizes a difficult customer and transfers the request to a less incompetent team.
7. Get response from higher-level tech support team, asking for clarifications (which of course were already included in the first two exchanges)
8. Re-send data highlighting the relevant parts, include cartoonish drawing making situation clear to even 5 year olds ("Bob is a transistor.  Where does it hurt, Bob?  Bob: I'm too hot.  Daddy said my collector can take lots of electricity.")
9. Two months pass.
10. "After totally unprompted investigations which are a regular part of our ISO9001 quality assurance program, our engineering team has determined that some batches of product XYZ were found to exhibit slightly degraded performance characteristics under very particular circumstances where no customers are likely to try to use them, and if they do they're probably of the annoying kind and they won't be getting free samples anymore."
11. Datasheet revision B, footnote 7. "Performance not guaranteed if device operated beyond typical parameters."

[SeanB]

I've moaned on this forum before about the 4000 series monostables we used at my old work in ...

It's too long ago to remember the exact type number, but we used the devices to provide a quite long time delay,...

Funny, we also used some of these old 4000 series CMOS parts for timers as well in the minute range.   buffered, unbuffered, brand, all made a difference for leakage.  We were pretty limited what we could use.  Seems like we found Motorola UBCP worked well.   We didn't want to pay extra for the higher temp parts and qualified with the cheapest.   Most parameters we didn't care about.   The caps were also a problem.    Cheap, crap design but worked alright.     

And I would bet the manufacturer was unconcerned, as they all were within the maximum leakage range, just that none would ever be in the "typical" part of the range at all. Those all were pulled off for the industrial part line, not the commercial line.

[joeqsmith]

I've moaned on this forum before about the 4000 series monostables we used at my old work in ...

It's too long ago to remember the exact type number, but we used the devices to provide a quite long time delay,...

Funny, we also used some of these old 4000 series CMOS parts for timers as well in the minute range.   buffered, unbuffered, brand, all made a difference for leakage.  We were pretty limited what we could use.  Seems like we found Motorola UBCP worked well.   We didn't want to pay extra for the higher temp parts and qualified with the cheapest.   Most parameters we didn't care about.   The caps were also a problem.    Cheap, crap design but worked alright.     

And I would bet the manufacturer was unconcerned, as they all were within the maximum leakage range, just that none would ever be in the "typical" part of the range at all. Those all were pulled off for the industrial part line, not the commercial line.

We bought enough parts we would typically work with the factory to screen certain parameters we were interest in.  It was more cost effective.

[Siwastaja]

In the errata, there was this gem:
"Revision A1, A2.  Due to a silicon design error, SPI3 does not function.  Workaround: none."

Sounds like ST . Their process in general is hilarious: documentation is written before the product exist, describing their internal hopes of how it would function. When the device is finally actually designed (at HDL level, or even manufactured), the documentation is not updated to reflect the actual behavior. Instead, they make a mental diff, list of differences during design, and then roll dice on each difference. If five or six, they add the diff on a separate "errata" document. The thing is never fixed for any silicon revision so they don't consider them errors, "errata" for them just means "the actual documentation". If they throw 1-4, they just leave it out because the errata would be too long.

I'm quite positive they only hire sadists, because any normal human being would just modify the original documentation - much faster to do anyway!

[PwrElectronics]

A  for me was those International Rectifier power Fet datasheets.

TO-220 or D2PAK device, maybe others.  The front page says it can do 300A (!!!!!).

Fine print and a few calculations later...  One learns that that figure assumes somehow you keep the case at 25C and the die junction is at 175C max spec.  Elsewhere one learns that the pins on those packages are only rated for 75A anyway (also at 25C).

[Psi]

A  for me was those International Rectifier power Fet datasheets.

TO-220 or D2PAK device, maybe others.  The front page says it can do 300A (!!!!!).

Fine print and a few calculations later...  One learns that that figure assumes somehow you keep the case at 25C and the die junction is at 175C max spec.  Elsewhere one learns that the pins on those packages are only rated for 75A anyway (also at 25C).

Did it say 300A continuous though? 

[tom66]

Did it say 300A continuous though?

Indeed, every time I've seen those very high power FETs it notes that the current is limited by the package first.  The high current rating of the actual silicon I guess is for assurance (and pulse applications where the leads survive high current momentarily.)  Anyone putting 300A continuously through a TO220 is in for a bad time.

[Psi]

I wonder if they are using the same silicon die in their TO220 as their PowerSOIC(8-DFN 5x6).
Might explain where the ratings come from.

[David Hess]

Anyone putting 300A continuously through a TO220 is in for a bad time.

What about 300A through a TO-92, and at 35 ohms?

[EPAIII]

I think I had one of those 300 A, TO-220 transistors once. I believe my fingerprint is still melted into it.

Not a joke. Real fingerprint in whatever the plating alloy was. OUCH!

And it did survive and continue to operate! So did my finger.

Did it say 300A continuous though?

Indeed, every time I've seen those very high power FETs it notes that the current is limited by the package first.  The high current rating of the actual silicon I guess is for assurance (and pulse applications where the leads survive high current momentarily.)  Anyone putting 300A continuously through a TO220 is in for a bad time.

[JPortici]

Anyone putting 300A continuously through a TO220 is in for a bad time.

What about 300A through a TO-92, and at 35 ohms?

Which of course was meant to be milliamps
https://docs.rs-online.com/4c4b/0900766b8137e252.pdf all pages mentions milliamps, besides the offending graph
and on the current version
https://ww1.microchip.com/downloads/aemDocuments/documents/APID/ProductDocuments/DataSheets/DN2535-Vertical-DMOS-FET-Data-Sheet-DS20005541.pdf
report milliamps

[aduinstat]

The i.MX6 datasheet claims it can gigabit. It cannot do gigabit. It does like half gigabit.

[berke]

The i.MX6 datasheet claims it can gigabit. It cannot do gigabit. It does like half gigabit.

RX, TX or both at the same time?  Can it jumbo?  IIRC the i.MX8 has trouble there as well.

[aduinstat]

The i.MX6 datasheet claims it can gigabit. It cannot do gigabit. It does like half gigabit.

RX, TX or both at the same time?  Can it jumbo?  IIRC the i.MX8 has trouble there as well.

It's been a while so I don't remember all the details, but I think we had jumbo packets and RX and TX tested individually.

[Smokey]

How about every RRIO opamp ever..
https://www.eevblog.com/forum/projects/how-rail-to-rail-are-we-really-talking-here/