OpampÆs Aberrant Input Offset

[logbook]

I designed in a TLC082AID dual opamp into my brand new circuit. First batch worked fine. Second batch were a pile of pooh. Checked the opamps themselves and the input offsets were over a mV in some cases. The data sheet claimed 0.75mV worst case at room temperature.

So I have had to make a test jig to select-out the out-of-spec opamps and keep production going in the short term. I documented the test jig and the results and got in touch with the applications department of the offending company. They asked a few dull questions and asked for some parts to be returned. So I selected out the worst ones and asked where they would like them returned to. The answer: send them back via the distributor we bought them from. So I had to send them back to the sub-contract manufacturer for them to return.

Now it turns out that the TLC082AID manufacturer has changed the input offset specification by a factor of two. But this fact didn’t come from the apps department, but from a colleague who downloaded the data sheet directly from the manufacturer rather than from the distributor’s website like I did. Quite why the applications department didn’t bother reading the results and seeing that the parts met their "new revised specification" is another of life’s great mysteries.

This input offset voltage change was for a mature part four years after its initial release! I have never heard of a manufacturer doing such a thing and frankly it is appalling. Although all manufacturers put the standard legal disclaimer in that they can change the spec at will at any point in time, changing the input offset by this huge amount is scandalous. How they think this is acceptable is a complete mystery to me. Probably some overpaid marketing twat who wanted to "keep the yield up". Needless to say I will view future offerings from this offending company with deep suspicion.

Anyone else have a similar experience?

 

[zeitghost]

I worked for a bit player in the semiconductor business.

On introduction, the "new" cmos analogue switches were speced as having an analogue range of +/-15V for rails of +/-15V.

Sadly, the yields were dreadful on leakage.

So the analogue range was re-specced as +/-14V instead.

Sorted.

Apart from all the military stuff that still had to be tested at +/-15V analogue & specially selected...

And then there was the analogue switch that used to turn off in the dark & cold... like you get inside a satellite in orbit... oh dear... it was much more pronounced with a dark ceramic package than it was with the white ones, which must be slightly transparent.

 

[itsmoked]

Sad state of affairs... I would love to see the before and after data sheets you mentioned.. Got links?

Luckily you probably have a 100 other opamps you can drop right in.

I build a product that uses 8 op amps for 4 channels of measurement. I had the same "kind" of problem in that I would assemble a unit and have zeroing type problems. If I mixed the opamps around they might straighten things out. So needless to say the first production run was hideous. I had piles of parts some worked some wouldn't.

Next pass I put in offset adjustment pots.. Problem gone. But I still see huge variable differences from lot to lot. In some production runs no adjustment is needed... next run they all may need adjustment.

It's a National cheapo opamp so I can't complain at all since they are all within spec. If I really cared I could bump up to lower tolerance parts.


I also note in my product a large temperature dependence so be careful you are comparing results of products that are at the same running temps.

 

[felixc]

I remember an LF353 having this kind of problem, long ago. But National corrected the problem later. Meanwhile we used a compatible part from TI.

Where am I wrong? On the data sheet, the first page says offset voltage 60uV, and inside it says typical 390uV and max 2000uV. Where's the 60uV of the first page?

Can you find a compatible part from National or Linear?

 

[logbook]

itsmoked, if you go to the RS website, rs

http://www.com,

and type in their part number 462-1229 they may still have the old data sheet on display. I sent them a snotogram today mentioning that the data was out of date.

felixc, I don't ever read the crap they put on the front page of data sheets. I'm just old and grumpy I suppose, but the front page has little relationship to anything useful. It's like these slection guides you get on some semiconductor websites, you search on offset voltage and they give you typical values. Well on average the input offset is zero; what use is that!

 

[itsmoked]

I read it... I use it too. But I am also aware that the front sheet could well have been written before the part existed by some eager marketing hack and hence may not match the specs inside that were written by some design EE.


Thanks for the link.. Curiosity sated.

snotogram... nice.

 

[sreid]

My real education started when we discovered a problem with an early Intel 8085. Being a large company, a large account salesman not only told us we were correct but gave us the secret, internal eratta sheet deliniating all known anomolies that were to be fixed in later mask revisions (about 20). We could then code around those problems but I felt for the small developers who would not have access to that information.

 

[itsmoked]

That's pretty typical of the bigger companies esp. The big "I". Master's of over hype.

Remember how they would say, "But Our 8051 runs at 12MHz! And "their" CPU only runs at 4 MHz." Never mind "their" processor had only 4 T-States and I's needed 12 T-States.

We were a tiny company and had a problem with an Intersil 7660. You know, the old flying cap voltage doubler when it first came out. They had a souped up version that had what I needed so I used it. They started having problems in the field because they wouldn't start up occasionally. To Intersil's credit they sent someone over immediately who listened intently. Then left and came back in a week with new ones that had been corrected and disposed of all our stock. They then gave us enough of them that we never needed to buy anymore for the rest of that product's production.

 

[felixc]

Some are honest enough to make the erratas available to everyone, while some others...
Ever heard of a BIOS update to correct a CPU bug? Intel provides microcode to BIOS developers, to work around bugs down the road. Unless you're yellow or red book, don't try to find out about the solved bugs. Pretty frustrating when you are a developer.
I also remember Cirrus Logic, issuing errata sheets as we were finding bugs in their chips during our development.

The art of turning a bug into a feature.

 

[fsmyth]

I remember getting errata sheets from Intel ages ago (mostly
804x/805x and the 8500's) but didn't follow them after the
'186). I guess they turned into the gorilla, but it is still
a shame. Have never had the problem with TI, but you had to
prod them a bit sometimes. National stuff, you tested the
first 2 or 3, and then used the same batch.

Seems to me that before everbody started buying everyone
else, the stuff just worked as advertised. If it was
preliminary, it was clearly marked as such. But then, we
didn't have to lock the doors where I lived, and drugs were
still at the drug store.
<als>

 

[Skogsgurra]

One of my more frustrating experiences was an NS 10 bit A/D - I think that the number was something like 3010(?). It worked OK sometimes and had a terrible and varying offset other times. Turned out to be the ceramic housing. It was fabricated out of white ceramic and it let a lot of light through - which obviously influenced the comparator or some other part inside. Put black tape on it and then it worked. Never saw any warning about that in the data sheet. BTW: remember NS having problems with direction of time? I still have those old data books where time runs backwards.

Gunnar Englund

http://www.gke.org

 

[itsmoked]

No, I never saw that time thing and I probably still have a couple of their data books. I used to keep up with everyone's Every data book. What a hassle. What a load! Must have been about 1500lbs of books covering 60 linear feet of shelves. But finally thanks to the WEB I was able to heave them all. But I kept a few of the NS app books with the 300 opamp circuits etc.

And the HP opto books that were literally text books on the subject of LEDs.

For example: Did you know that if you blink LEDs at a much higher current but a lower average current Humans will perceive them as brighter?

I also had a light problem but it was with a light sensor. A TEOS system; semiconductor machinery cleaning process where a gas (TEOS) is stuffed into a chamber and a plasma reaction commences. Once the color of the plasma reaches a certain wavelength termination occurs. The sensor would flake out and do early terminations. Turned out if the control unit was mounted where ambient light could fall on the board the light would transit thru the actual standard FRP material between layers and get to the sensor.