TI acquisition of National

For those there is still SO8. If a chip is never ported to SMT (which we already used throughout in 1986 ...) then it is usually a sign for me to steer clear. I can't have stuff go unobtanium on my clients.

High frequency noise can be filtered out. The 1/f noise was the killer here, and the LM4140 has proved to be very good in this respect. We used the AD706 to follow it, and check the LF noise by reading a Keithley 6517 5-digit Voltmeter over several hours. We have to warm up the Keithley for a day before starting measurements.

I'll check out these others, thanks.

Jon

Ah, the LM329 is a 6.9 V ref, but we only have nominally 6 V unregulated input. Same problem with the LT1021. Anyway, both these regs only spec LF noise down to 10 Hz, and even the 10 Hz noise is no better than the LM4140. We have tested the noise of the LM4140 out to at least .0001 Hz (2 hour run) and STILL do not know where the knee is! It is down to the LEFT of .0001 Hz! One little trick we found, almost by pure accident, is that reducing supply voltage to the ref improves the 1/f noise considerably. We run them at about 4 V input.

Given the sample plots in the data sheets, I'll put the LM4140 up against the other two devices, at least, for the extreme LF noise range. The chips we are powering with this ref are REALLY sensitive to small changes in the supply rail, so if the ref misbehaves, we WILL know about it. We make histograms of the signal amplitude, and if the supply voltage jumps, the centroid of the histogram shifts appreciably.

Jon

My

er

up

ly

is

This is the kind of information that makes this user-group useful.

The flame wars are what we do when there's a shortage of useful information to share .

-- Bill Sloman, Nijmegen

A 1/f knee left of 0.0001Hz is so unusually low, it makes me wonder if it isn't really buried below a stronger flat thermal source....

You might try the LTC6655, btw. There is a J.Williams appnote, too, on how to measure the noise of the 6655 properly.

regards, Gerhard

Thermal regulation helps these things a lot, as it helps crystal oscillators. The ultimate voltage reference would be a *lot* of low-noise chips on a constant-temperature block, with the outputs averaged.

John

I'm doing that averaging currently with lots of ADA4898-2s in the baseband and Wilkinsons & SMA-Amps on RF :-)

Gerhard

Yeah, that's the noisy version of the LT1028.

John

No, it's the quiet version. It doesn't have the 300 kHz noise peak, and because it has the usual two gain stages instead of three, it isn't as squirrelly to compensate.

Cheers

Phil Hobbs

Actually those buffers were really good. But that is old National. The place has been a mess for decades.

I recall a story from some ex-National employees about having to walk through some puddle in the fab. Turned out to be something toxic.

Certainly for the last two decades, National deserved the reputation as the place to mentally retire. Even if you found some part you think can't be matched, do you honestly trust National to test to spec? These guys were caught for failing to test military grade chips properly.

Ship the parts and let God sort them out.

Who the f*ck said it was trivial? All I said it is done and done all the time. You can't even repair a lot of old fab equipment. Nothing is really static in the analog realm.

You do realize that even without changing fabs, the process can change. Changing wafer vendors for instance.

Even tooling gets changed. Shit happens in the fab all the time. Worse, your tooling might get sabotaged, especially at a foundry. Or you change steppers and that requires new alignment targets. Etc etc etc.

Some companies will rev the mask layer for new tooling. Some will add numbers elsewhere and leave the numbers alone. Once place I worked kept track on tooling changes with Chinese numbers.

There is no assurance that all the parts are even run on the same testers.

Don't know about that. However, they came out with some super fast ADC that caused "Oh s..t!" faces at the competition. Also, they tend to be able to deliver production quantities, in contrast to some other company that we all know ... and that usually matters more than anything else.

Thanks!

We had an application with a custom chip that has a very sensitive charge-integrating amplifier. The source of the input device has to be held to an insanely clean and stable supply rail. I knew this was going to be harder than anything I had done before, and was prepared for the worst. When looking for a good reference, I selected the LM4140 first, based on the spec sheet. I used an OP275, my usual choice for a general op-amp, and had big noise problems. After long perusal of the AD op-amp book, I found that the AD706 had a noise spec that went lower in frequency than just about any other part. I had already verified with the Keithley meter that the problem was clearly 1/f noise, as everything would be stable for several minutes, then we'd get a big jump in offset that was several times the distribution of our test signal. The AD706 nearly fixed the problem, then dropping the supply voltage to the LM4140 ref was the last tweak.

I was nearly stumbling around blind, but found a solution that worked. Part of the problem was that it was very hard to measure what I was looking for, and I had to use the system under test (our custom chip and support circuits) as part of the measurement apparatus!

Jon

The first cut at our ASIC had a massive thermal sensitivity in a source follower transistor that wasn't well-appreciated by the chip designer. This was not causing 1/f-type jumps, but a large thermal drift that just moved up and down constantly, as drafts in the room changed the cooling of the chip. We had to put cooling bars across the boards and have a water-bath circulator holding temperature to .1 C to stabilize that mess. We didn't really even SEE the 1/f jumps until we had the drift under control.

OK, well we are not going to change anything now that we have something that works, but I will check out that part.

Jon

Hmm, note 7 gives some REAL weasel-words about noise lower than .1Hz. Basically, they are not guaranteeing ANYTHING at all below .1 Hz. While the LM4140 spec doesn't really go a whole lot lower than that, we KNOW that the part actually does stay quiet to much lower frequency. The LTC6655 datasheet says that 50% of the 10 second sampling intervals will be quieter than the guaranteed spec! WOW, was that written by lawyers? What the HELL happens in the other 50% of the 10 sec samples? Anything at all, the way I read it!

Jon