THS3491: data on the GD2/HD3 validation

Hello Nishie-san,

  1)  We do not physically test for lot variation for distortion performance (HD2 and HD3). The variation we do test come from our DC specifications via ATE prior to unit distribution. AC specifications are difficult to physically test lot wise unless specified in the datasheet usually with open-loop gain or bandwidth. 

  2) I can look into if we have data on a sample size of THS3491 for distortion. However, this is unlikely, and in the original e2e thread the distribution profile (Gaussian) is based on design simulation and statistics rather than physical data. Therefore, you can expect around 99% of the device to be within +/-5 standard deviation from the listed typical in the datasheet. 

This e2e thread has really good information on gaussian distribution of our specifications. The same explanation would follow our HD2/HD3 specification. 

Thank you,

Sima 

Sima, I do not believe a large sample size was used to get a range on HD for the THS3491. I have done that work in the past (Including screening drivers for distortion for HP AWGs)  and the HD is more like a Poisson distribution being a cancellation effect. Using Guassian on a HD spec predicts a much too conservative min level. It is also requisite that you do your statistics using uV/V kind of data instead of dB data. We did put some min AC specs into some of the Burr Brown high speed parts, see for instance the OPA2677 spec tables (This device was for xDSL line drivers where HD is also critical)  - this approach was abandoned with the TI buyout. 

https://www.ti.com/lit/ds/sbos126i/sbos126i.pdf

Hello Michael,

   It looks like that yep, we do not have a large sample size on HD.

   Thanks for the information! That is really good to know that there was data backing a specific distribution. For Burr Brown, how did they determine the min AC spec? Was this via generalized Poisson distribution based on your findings or was this via design simulation? 

Thank you,
Sima 

Well Simi, in development the PDK's (Process Design Kits) always have process and temp variation in the models. The OPA2677 (and others at the time) were simulated by design over process and temp in CBC10 to build that spec table. The same could be done for more recent products but it is very time consuming (expensive). 

Working from uV/V type HD data, the HD2 shows a Poisson distribution - it is a cancellation effect up and down through the balanced op amp structure - sometimes you get perfect cancellation showing a very long tail going low in uV/V. The HD3 is really more Gaussian being curvature through zero crossing. 

I have seen really tragic cases of test groups doing this in dB terms are grossly underselling their performance. 

Hi Sima-san, Michael-san,

Thank you for your support.

Q3. 　I understand that the HD2 distribution follows Poisson distribution  and HD3 profile is Gaussian.

          Can you advise me how much performance degradation do you count on when less than 100 ppm failure rate is expected?

Best Regards,

Nishie

Hi Sima-san,

I am sorry that my explanation is not sufficient.

My intended question is how the HD2 and HD3 characteristics vary from device to device.
I thought that if we knew the distribution of the variation, we would be able to know the probability of including samples that did not meet the minimum HD2 and HD3 characteristics required by the customer.

Best Regards,

Nishie

Hi Team,

Please allow me to supplement Nishie-san’s question.

Each THS3491 shows different HD2/HD3 performance.

Some THS3491 shows better than –75 dBc of HD2 performance and some shows poor performance than it.

In the meantime, my system requires better than, says -70 dBc  of HD2 characteristics to achieve expected system goal.

The performance validation may follow statistical theory.

When you purchase million pieces of THS3491s, some of them do not meet this criterion.

From costumer’s stand point of view, they are “failure”. (The poor performance devices may be through away)

I want to know how much the HD2 and HD3 performance degradation you need to count on, to maintain less than

100 ppm failure rate.

I also want to know how much the HD2 and HD3 performance degradation depending on the supply voltage change (+/- 5%)

and temperature change (-40 to 85 degrees).

Mita

It's soooooooo good to have you here, dear Michael

I enjoy every post you write and learn, learn and learn...

Kai

Thanks Kai, 

And, I did find that presentation from a 2004 dynamic range - 

here you go, the inner workings of distortion, 

2004HS_2DynamicRange.ppt

Kai-san, Michael-san,

Thank you for your answer.

I understand that it is not feasible to estimate the relationship between HD2/3 minimum performance and failure rate.

I will try to find high speed AMP which should have adequate HD2/3 performance to meet system requests.

Mita

Hi Mita,

an OPAmp with a much higher supply current is a good candidate for lower distortion. A higher supply current allows the class AB output stage (and other internal stages) to be furtherly driven into the class A region.

But keep in mind that an OPAmp with a much higher supply current will hardly allow supply voltages up to 32V, because of the enormous heat dissipation. +/-15V times the 17mA of THS3491 is 510mW !

To be honest, the THS3491 appears to be a very good compromise between maximum supply voltage, load current, bandwidth, distortion and supply current. It will be hard to find a better OPAmp.

Kai

And, it will be even harder to find someone to provide this kind of requested data - easy to ask for, quite difficult to extract.