LM741: Ibias specification

I noticed an error, please consider this as the second question:

2)Can I consider datasheet maximum value as the 6-sigma?

The distribution should look like half Gaussian (negative value are not allowed), so I suppose 6-sigma is measured from zero...

Well you have possibly picked the oldest still in production op amp to peruse. 

Hard to tell on this data sheet, but for most op amps that are not CMOS or JFET input, the input bias and offset current are tested at final ATE.  So the march to product release involves truncating the actual distribution at an acceptable yield (25C only) limit. Similar for offset voltage. 

Hello Massimo

1) Yes mean value 

2) Probably true.

3) IIB is based on transistor gain and bias current in input stage. input stage bias varies with ICC. So bias current is complicated.

4) Maximum is high enough that it should (almost) never happen.

IIB does flow in one direction as it is dominated by bipolar base current.

We have many other devices with lower or nearly no IIB.

Hello,

thank you for your replies. My last statement derived from what I read in SLOA011:

"Most parameters have a statistically normal distribution. The
typical value published in the data sheet is the mean or average
value of the distribution, with one exception, offset voltage. The
average offset voltage is normally zero (or very close to zero).
Therefore, the typical value listed for offset voltage is the 1s value.
This means that in 68% of the devices tested the parameter was
found to be ± the typical value or better. The definition of minimum
and maximum values has changed over the years. Texas
Instruments currently publishes a conservative 6s value."

I was wondering if this statistical explanation regarding Vos is also applicable to other parameters "centered around zero" (like Ibias of OPA188), that would have a mean equal to zero (I think).

Thank you and regards,

Massimo

Massimo,

This chart comes from OPA188 data sheet that states a 6uV typical (mean) while disregarding polarity otherwise mean would be 0uV

Hello Massimo, 

Well that SLOA011 comment about using 6sigma is not universally correct. This was written by Jim Karki in Dallas 1998 before the BurrBrown acquisition when they were just kind of learning op amps again out there. Many of the TI - Tucson(BurrBrown) have a trimmed and tested offset spec (for instance) where the tails of the distribution are cut off in final test at minimal yield loss but much tighter limits. 

Here, for instance, are ATE offset histograms for the OPA837 3V supply test. The limits here are set to +/-120uV where that bar of 5 units at the right are exceeding that and failing - discarded. It is very rare to set 6sigma limits on precision specs if you are doing ATE anyway, 4 to 5sig maybe, but I have never heard of 6 sigma. 

Hello Michael,

thank you for your observations about the 6-sigma limit, reported in SLOA011.

The 4sigma limit seems consistent also with OPA837 datasheet: the typical value (1sigma) is +-30uV and you have reported an ATE test at +-120uV, which is really 4sigma. The larger range +-165uV reported in datasheet, could lead to an assumption of a final test set to 5.5sigma (165/30) , but probably it is written to have some margin, since the tighter limits +-120uV are used by ATE.

Coming back to SLOA011:

"The typical value published in the data sheet mean or average value of the distribution, with one exception, offset voltage...Therefore, the typical value listed for offset voltage is the 1sigma value."

I am wondering if Vos is really the only exception or if I can also extend this assumption to other parameters with zero mean.

In other words, the key point is to understand if typical value is interpreted as mean or standard deviation, since it changes the circuit analysis: if the datasheet parameter is a mean value (I made the example to Ibias of LM741 specification) it can be used "as is" to estimate the effects on Vout, whereas if it is intended to be 1sigma, I have inevitably to take +- signs into account despite datasheet could only show a positive value  (as for Vos of some OpAmps, for example).  So I think that the correct interpretation of the "typical" datasheet value (mean vs standard deviation) must be clear, otherwise it can lead to errors in circuit analysis.

If you can suggest updated documents regarding the proper interpretation of specs, I would really appreciate them.

Regards,

Massimo

I found myself ruminating about these issues as well yesterday - there are in fact many layers to each DC spec line that are not exposed to the design community. And, over time and across different groups, the details are often different - and rarely documented (publically)

1. The spec expansion you note on the OPA837 is also partially due to what we call test repeatability guardband - that is accounting for measurement spread (for a single device) across ATE platforms and boards - this is sort of a CYA across test hardware. 

2. If a DC spec has a 0 mean (like offset voltage and sometimes offset current), the practice is to use a 1sigma number as zero does not give you any information. 

3. Many specs are unipolar (like pnp input bias current like the OPA837) where the mean "means" something. Others, like the OPA690 or bias current cancelled inputs, are bipolar input bias currents where you might not have a zero mean, but often those are mean +/-1sig perhaps. 

4. The most recent conundrum I ran into was whether CMRR is unipolar? I think it is in a lot (but maybe not all) cases. The sim models certainly produce a single polarity, and sometimes that reverses polarity as models are updated - who knew?? There is some talk of CMRR being a centered Guassian uV/V distribution - I find that amusing as having never seen anything but a one sided distribution. Another CYA?

I did kind of drill down on this last issue a bit in this article, not resolved yet just an observation. Where this started for me is the CMRR spec on a current feedback is attributed to the buffer gain being <1.0000 and is always a "gain" compression factor for a non-inverting design. not sure yet about the vast range of other devices out there, 

This CMRR discussion in here should have been in #3, but I did not run into at that point in time, 

https://www.planetanalog.com/dc-precision-considerations-for-high-speed-current-feedback-and-fully-differential-amplifiers-insight-4/

You bet Massimo,

Not a simple question,nor a simple answer.