CMOS op amp datasheet: Lies! All lies!

They've been going to seminars titled "How to Write Data Sheets the Takata Way"?

Maybe they forgot to power it up when they did the measurement. Or they accidentally typed "f" when they meant "n".

That sure seems atrocious.

Hoping for a learning experience here I will ask some dumb questions...

-The 0.6fA/rtHz is a _typical_ spec; and there appears to be no input protection. Did you try more than one unit, hopefully with parts from another batch? Ideally babied through the build process to guarantee no zaps during construction?

-The spec is at 1kHz, and your measurement was at 1Hz. Do you know how much the noise current of the part varies with frequency?

Thanks!

There's always input protection--see the I_in vs V_in curve for negative voltages. That 0.6 fA is a very suspicious number--it's exactly the predicted shot noise of the 1 pA typical bias current. I expect that nobody at Diodes Inc. has ever done the measurement.

Did you try more than one unit, hopefully with parts from

It's possible it could be a bad or damaged unit, but it works fine otherwise.

If there were as much as a nanoamp of input current, it would be very obvious in the output voltage, but there isn't. The offset is less than

I say "1-Hz noise" so that I can quote noise density without having to write "per root hertz" all the time. I did measurements at 100 Hz, 1 kHz, and 10 kHz. The input capacitance is big enough that there's a lot of gain peaking in the kilohertzes when you're using a 200M feedback resistor. (I did the measurement with my HP 35665A DSA.)

Suggestions for better high-Z jellybean duals, anybody?

Cheers

Phil Hobbs

I checked the second unit, and it's just like the first one--the 1-Hz output noise is 20 uV at 8 Hz and goes up like sqrt(f) until it rolls over at about 1 kHz. (It's 168 uV at 8 kHz.) That's almost certainly some 1/sqrt(f) contribution getting differentiated by some internal capacitance that isn't isolated.

The TLC2262 version is flat at 10 uV near DC, rising to 23 uV at 1 kHz due to the input capacitance.

I've ordered some of the following jellybeans to try:

TS272IDT TS27L2CDT MCPL02T MCP6002T MCP6242

We'll see!

Cheers

Phil Hobbs

Hi, Phil:-

A pet peeve of mine that they don't tell you what the noise current

Used quite a few TLC27L2 chips in the past- they're solid and second sourced. The front end is something like 100 interdigitated transistors in parallel for each side, so the DC characteristics are excellent (well, for MOSFETs anyway).

--sp

Have you tried the original TI version?

No, I haven't--probably should add that to the list.

Cheers

Phil Hobbs

For a SWAG I normally assume that it's at the quoted frequency, so that it's never less than the quoted value in the flatband and rises like 1/f below there.

Terrific, thanks. I should have some on Thursday.

Cheers

Phil Hobbs

Unless it's like with bicycle frames where in the end they all come from the same factory, just the labels are different.

As someone else mentioned, I noticed but never tried the second sources (I think ST was making them)- it's probably worth comparing the original TI parts.

--sp

I assume they set up ST with the LinCMOS process so that picky buyers would have a bona-fide second source, but that may not have been the case.

Maybe someone here knows.

--sp

Spehro, (As I've only recently learned... AoE3) Isn't the corner in the current noise where it crosses the voltage noise times the input capacitance. (Well assuming all noise is "ideal"..which is not always the case :^) I've found voltage noise peaks in opamps that weren't in the data sheet. (nothing like Phil's 600x.)

lotsa opamps though... George H.

I don't trust "typical" specs. I figure that no matter how bad the part may be, if I could even get the attention of the manufacturer they'd just look at me blankly and say "but that's the _typical_ number -- you must just have had bad luck".

Tim, that was certainly true when I entered the industry over 30 years ago, but is much less so now, at least where I work.

There are typical specs and there are typical specs. Sometimes the spec is something that would require a lot of test time and hence would be extremely expensive to do (really small leakage currents can fall into this category, as can things like sub-100ps rise/fall times). Sometimes it's a parameter with a demonstrated (in characterization) reasonable correlation to some other parameter that actually is measured. And, true, sometimes the manufacturer is too lazy or just doesn't care about their customers. Personally I've always fought to make typical specs on my chips representative of real performance and not simply blow them off. I say all this as someone who's designed quite a few chips and written even more datasheets.

For the last 20+ years I've worked on really high-speed high-performance custom and semi-custom ICs used in the big-iron automatic test equipment industry (Teradyne, Advantest, guys like that). Most of these are custom jobs, but a recent datasheet you can see is ADATE320. These are not inexpensive chips and can have quite substantial test times, but there are still a lot of "typical" specs. We still take those typical specs very seriously - if you look at the datasheet you'll see a column headed "Test Level" that explains it in great detail - and pretty much every "typical" spec reflects the mean of characterization lab measurements of samples from multiple lots. If the standard deviation is large we'll often bulk up the spec to reduce the likelihood of a customer getting a part that is vastly different from the datasheet.

Really cheap CMOS op-amps are a different arena. Customers only want to pay pennies for these things, and the manufacturers need to make a profit to stay in business, so pA and lower leakage currents simply can't be measured in production as the test time would cost more than the sell price. This is where you have to build and measure prototypes so you can develop trust your supplier. And as others have pointed out, second sources chosen purely on a cost basis can be very risky!

Understood. But in the present case, I find it hard to believe that that fa ctor of 400 in the current noise is a lot-to-lot variation, so they only ha d to measure it once and 'fess up.

Cheers

Phil Hobbs

Like that nasty bump at ~400kHz in the LT1028?

About 7:1 - it's in the datasheet (page 10), just not on the front page curve where they tout the low noise.

Second sources have a vested interest in just copying the data sheet specs from the original manufacturer (unless they can do significantly better, and that matters to customers).

If they meet all the guaranteed specs, but their 'typical' numbers are very different from the original..

--sp

Well at least it's hidden in the data sheet somewhere. This was years ago and I'd have to go digging in my note books, or redo it. But the opa124 has a broad peak in voltage noise... part of me remembers ~20 kHz, but it could have been 200 kHz. (Or some different frequency.) No hint in the data sheet....

Except the noise at 10 kHz is "guaranteed by design"

George H.

It's in the _revised_ datasheet. They had to be called out on it first.

customers might know what they were buying? Whoda thunk?

Cheers

Phil Hobbs