IC datasheet not matching the product performance

[EEEnthusiast]

Folks,
How many of you had experience with semiconductor products, where the product performance significantly deviated from the datasheet. Many a times, the manufacturer cuts corners in the product test cycle and arrive at specs which are half baked. Other times they quote the simulation data for parameters which cannot be measured easily but they never say that on the datasheet.
I had worked for one of the Top-5 semiconductor company in the world and I had seen this trend degrading over time, in order to save the overall cost.
Any experiences???

[donotdespisethesnake]

How many of you had experience with semiconductor products, where the product performance significantly deviated from the datasheet.

Sure, MCUs with half the specified RAM. Optos that go way off spec after a week in use. IGBTs that randomly blow up. Unfortunately, none of our production tests found the problems, but our customers did Warranty claims went through the roof, only then did management get seriously interested in quality. At least with the first problem I found in pre-production it was one bad batch of chips, but the production test would still miss it and deliver faulty products to customers if it happened again.

All components sourced from top distributors, no grey market stuff. It took a lot of investigation and evidence before the manufacturers finally admitted fault. Even then they claimed it was a few bad batches, which was obviously not true.

[EEEnthusiast]

That thing called corporate ethics is only to be found on the policy papers. When they have to save costs, they do that by any means which may include lying to the customers, manipulating the data sheets and shipping multiple production tests. The top executives show this money saved to the shareholders and increase their bonus payouts. Then they leave the company before the bugs are discovered....
Now that I design products using the same ICs that I helped produce once, I feel the fear to choose parts based on the datasheet without actually testing them...

[exe]

There are lies, big lies, and datasheets .

I haven't seen datasheet violation myself, but I've seen performance much worse than expected. I was always able to find a fine-print, or another parameter that could explain the behavior, but still . One common gray area is voltage references and voltage regulators. Esp. when it comes to hysteresis, drift and humidity. Basically, if these parameters are not specified, then the whole datasheet doesn't really make sense (for precision applications).

I'm also concerned about "typical values", esp. if min/max is not given.

[EEEnthusiast]

Yes, the typical value specification is where most of the misleading happens. There are cases where the worst case numbers can be significantly off from a normal distribution of data. Hence they just skip the min-max values and leave it to customer's discretion (bad luck i should say).

[exe]

So far this is my favorite case of not meeting specs: https://www.analog.com/media/en/reference-design-documentation/design-notes/dn229f.pdf . It talks about XFET references who didn't meet their long-term drift specs by two orders of magnitude:

Measured data shows this reference to have drift between 60ppm/kHr and 150ppm/kHr or 300 to 750 times worse than claimed.

[jmelson]

Folks,
How many of you had experience with semiconductor products, where the product performance significantly deviated from the datasheet. Many a times, the manufacturer cuts corners in the product test cycle and arrive at specs which are half baked.

Relatively few, and some were caused by poor implementation on our part.  But, I can remember a few.
We bought a bunch of digitally-controlled silicon delay lines.  They had a bunch of distributed silicon delay elements in a 1-2-4-8-etc pattern, and you could cut in the various length delays
with an 8-bit TTL input.  They were spec'ed to have just a couple percent tolerance.  Well, some of the chips didn't work at all, some of them were amazingly far out of tolerance, like +/- 25%
or worse, some drifted badly after a day, etc.  I think the whole project got trashed because these things didn't work.

We built a 32-channel constant fraction discriminator, that consists of 2 comparators and a little logic to trigger when both comparators are tripped, for each channel.  Well, I picked some jellybean single-element CMOS gate and FF chips to do that logic.  There was a huge problem with switching noise contaminating the whole board.  After much tribulation, I set about to measure the switching noise.
With a minimal series resistor on the isolated power islands for each channel, I measured the shoot-through pulse of the logic at something like 3 A for 3 ns.  This was a very approximate measurement due to parasitic inductance, but gave an order of magnitude indication of where the problem was.  I ended up completely redesigning the board with all different components, but I found some NXP gate chips that were touted as having a VERY small "equivalent gate charge", and with the same setup, I was not able to measure ANY transient pulse on the Vcc supply!

Those are two that come to mind immediately.

Jon

[KrudyZ]

So far this is my favorite case of not meeting specs: https://www.analog.com/media/en/reference-design-documentation/design-notes/dn229f.pdf . It talks about XFET references who didn't meet their long-term drift specs by two orders of magnitude:

Measured data shows this reference to have drift between 60ppm/kHr and 150ppm/kHr or 300 to 750 times worse than claimed.

What is funny about this one is that Linear Tech is dissing an Analog Devices part and this is now pulled from the ADI web site since they bought Linear...

[EEEnthusiast]

What is funny about this one is that Linear Tech is dissing an Analog Devices part and this is now pulled from the ADI web site since they bought Linear...

[magic]

That's not strictly a case of "not meeting the spec" because there is no guaranteed spec here, but I found this little gem in Texas Instruments OP-07 datasheet.

"Functional block diagram" they call it, although it pretends to be a full schematic, unlike other vendors OP-07 schematics which show the general 3-stage topology of the opamp with implementation details omitted for brevity.

This one of course is different, showing a 2-stage topology which looks like an illegitimate child of LM101 and RC4558
No trace of the annoying input protection, which is perhaps the first part of the schematic actual designers would want to see
Offset null pins are shown referenced to GND, but in reality they are referenced to VCC and the datasheet says so on the next page

They even include a "components count" table which shows the number of transistors etc on their fake schematic (I counted). What are the chances that this reflects the number of components in the actual chip?

And the best thing: somebody was paid to sit for a few days and invent this nonexistent circuit and draw the schematic, which is neither true nor helpful. Why?

[tszaboo]

It is usually easy to detect these kind of issues. They write "Microchip" on those chips.
Also, they place Maxim logo on chips, that are obsolete or "out of stock, we never even had this chip".

[VK3DRB]

Honeywell had these expensive airflow sensors worth around $250 each. I bought about 6 of them from Element 14. Problem is Honeywell had changed the pin-out of the sensors, but they never changed the part number . To make matters worse the datasheet enclosed in each airflow sensor box was for the sensors prior to the pin-out change. Bloody idiots. I wired them up as per the datasheet exactly and found I could not get them to work. Eventually after trying the fourth sensor,  I worked out they were actually destroyed due to incorrect wiring. I provided a fully detailed report with my findings to both Honeywell and Element 14.

Honeywell's response: Not our problem, return to the place of purchase.
Element 14's response: Not our problem, return to the manufacturer.
Rinse and repeat for a few months with phone calls and emails until I just gave up in frustration.

My message to Honeywell: You are inept. I have avoided using any of your products in my designs for the past 10 years.
My message to Element 14: You broke Australian consumer law and I should have reported you to the ACCC. You have been my last resort when buying parts over the past 10 years.

Thank goodness for reputable companies like Digikey and Mouser - never had a problem with either of them. Brilliant service, and great websites too.

[exe]

Ah, pinout, this reminded me one case from many years ago. It was early 2000-s, I was student, I used all my pocket money on an atmel mcu and a 16x2 dotmatrix display for it (I was trying build a PDA, a pocket computer that used to be popular in pre-smartphone era). For an obscure reason to me, even the cheapest LCD displays were very expensive (for a first-year student, it was ~$15, which was my month scholarship back in the day). Guess what, the display didn't want to work at all.

Turned out there were two similar displays with similar characteristics, with the only significant difference is Vcc and GND were swapped. And the datasheet on the distributor's website was from the wrong model. So I fried the display. I wasn't able to return it, or get a reimbursement, the distributor said that it's my problem to get correct datasheets. I Had to spend last money on another display, which, thankfully, worked. Eventually datasheet was quietly updated.

Nowadays I can buy much more advanced displays (read graphical TFT) from aliexpress for much cheaper, and even have some rudimentary buyers protection. And displays are still expensive when buying from big distributors.