What's in a datasheet?
Posted 7th October 2011 at 05:38 AM by abraxalito
Updated 14th November 2012 at 02:47 AM by abraxalito (Updated with link to new article 14th Nov 2012)
Updated 14th November 2012 at 02:47 AM by abraxalito (Updated with link to new article 14th Nov 2012)
Yesterday I had this very interesting exchange with RocketScientist about his open source design for a headphone amp, the O2.
https://www.diyaudio.com/forums/headp...ml#post2736266
The nub of the issue raised here is - should designers stick only to what datasheets tell them about parts or to what extent use what's 'common knowledge' about parts to eek out better performance?
I was surprised to learn from RS that offsets within dual opamps are so closely matched in practice - its a really new discovery for me. So why don't semiconductor manufacturers tout this feature? Or perhaps RS just 'got lucky' with the relatively few samples he tested?
My experience of reading opamp datasheets is that the specs for offsets (both the typicals and the max) degrade in going from single to dual devices, where the devices are all on one die. Let's have a look at a relevant opamp from the audio domain, the OPA211/2211 :https://www.ti.com/lit/gpn/opa211
Here the offset and bias current specs are worse for the dual than for the single - not by very much - but the trend is clear. They don't make any mention of the offsets in duals being closely aligned. If they did, some more interesting applications similar to those of RS's paralleling might be possible. I've already thought of one myself where the offset of one opamp is used to compensate for that in a second, on the same die. But I have no experience yet in measuring relative offsets in dual opamps - I plan to get some though
So what do you guys on diyaudio think - do you want designs where the designer is relying on some non-guaranteed aspect of a device's performance based on a few sample measurements? Or would you prefer that a designer sticks strictly to what a manufacturer is prepared to guarantee for their product in the official datasheet?
This is relevant to one or two of my designs that I've been working on with ARM microcontrollers. As many are aware, there's a phenomenon in PCs called 'overclocking' - running a CPU beyond its advertised performance specs. I have so far found that ARM CPUs will overclock quite well in practice, but how far to push that in published designs?
Update: Over one year later, and I happened upon this interesting article about the matching of the individual opamps within a dual, from TI -
https://e2e.ti.com/blogs_/b/thesignal...l-op-amps.aspx
There it is from the horse's mouth - dual opamp offsets aren't matched more than between individual opamps in separate packages. Looks like RS was just blowing smoke
https://www.diyaudio.com/forums/headp...ml#post2736266
The nub of the issue raised here is - should designers stick only to what datasheets tell them about parts or to what extent use what's 'common knowledge' about parts to eek out better performance?
I was surprised to learn from RS that offsets within dual opamps are so closely matched in practice - its a really new discovery for me. So why don't semiconductor manufacturers tout this feature? Or perhaps RS just 'got lucky' with the relatively few samples he tested?
My experience of reading opamp datasheets is that the specs for offsets (both the typicals and the max) degrade in going from single to dual devices, where the devices are all on one die. Let's have a look at a relevant opamp from the audio domain, the OPA211/2211 :https://www.ti.com/lit/gpn/opa211
Here the offset and bias current specs are worse for the dual than for the single - not by very much - but the trend is clear. They don't make any mention of the offsets in duals being closely aligned. If they did, some more interesting applications similar to those of RS's paralleling might be possible. I've already thought of one myself where the offset of one opamp is used to compensate for that in a second, on the same die. But I have no experience yet in measuring relative offsets in dual opamps - I plan to get some though
So what do you guys on diyaudio think - do you want designs where the designer is relying on some non-guaranteed aspect of a device's performance based on a few sample measurements? Or would you prefer that a designer sticks strictly to what a manufacturer is prepared to guarantee for their product in the official datasheet?
This is relevant to one or two of my designs that I've been working on with ARM microcontrollers. As many are aware, there's a phenomenon in PCs called 'overclocking' - running a CPU beyond its advertised performance specs. I have so far found that ARM CPUs will overclock quite well in practice, but how far to push that in published designs?
Update: Over one year later, and I happened upon this interesting article about the matching of the individual opamps within a dual, from TI -
https://e2e.ti.com/blogs_/b/thesignal...l-op-amps.aspx
There it is from the horse's mouth - dual opamp offsets aren't matched more than between individual opamps in separate packages. Looks like RS was just blowing smoke
Total Comments 7
Comments
-
Somebody once said "it's what they leave out of datasheets that is the most telling"Posted 7th October 2011 at 10:52 AM by jkeny
-
as a designer for production not DIY ie one offs, I can only rely on the data sheets ( esp. a concern with multiple vendor sources) using typical numbers and worst case limits and guessing on a sigma spread, so it's best IMO to design out that particular parameter with a different circuit if it is indeed a problem. Using your example Re DC offsets use 1) redesign AC coupling stage, 2) trim or adjust on the test line, 3) added vendor specs, or 4) cherry picking parts. If an important parameter is not specified an additional procedure 'select in test' SIT maybe required on the line (not desirable).
luckily DIY doesn't have deal with production yields per se. so typical #'s can be used more or less.Posted 8th October 2011 at 12:09 PM by infinia
-
With regards to selecting parts ("cherry picking"), you'd presumably have to test over the full temperature range.Posted 8th October 2011 at 05:38 PM by CopperTop
-
A dual (or quad) op-amp has all devices on one die. This increases the odds of having the devices match, which, in turn lowers offset. So far is common knowledge.
However, many parameters affect device matching. Dopant implant profiles, stresses, operating points, just to name a few.
The reason the one op-amp you found exhibit worse offset for a dual than for a single is likely that the die stresses are worse for the dual. The dual pin-out is also significantly different from a single pin-out, so the routing may be worse for the dual.
The reason you don't find matching specs for the two channels of a dual op-amp is likely that the manufacturer doesn't test it, hence, can't make any guarantees about it. If the parameter isn't important to the end customer it more often than not isn't tested.
~TomPosted 10th October 2011 at 06:41 PM by tomchr
-
Thanks Tom, sounds like you have useful inside info there. As an engineer it most certainly would be interesting to me to have a relative offset specification for dual opamps. Armed with that spec I would be able to design a circuit which allows me to use the offset of one of the pair to compensate for the other - in situations where it was important of course. This might allow (for example) a lower grade dual to work beyond the performance of the higher grade (and more expensive) part.Posted 11th October 2011 at 01:12 AM by abraxalito
-
"This might allow (for example) a lower grade dual to work beyond the performance of the higher grade (and more expensive) part"
Which may be why they don't tell you!
Tony.Posted 11th October 2011 at 01:51 AM by wintermute
-
Ha yes Tony but the lower grade part takes two opamps to achieve what the higher grade part realized with just one. This is relevant to the O2 design because that's thermally limited in its current guise. Going to two separate packages releases that constraint. Most opamp designs I would suggest aren't limited by their thermals.Posted 11th October 2011 at 09:10 AM by abraxalito
