Author Topic: Voltage reference noise processes  (Read 3235 times)

0 Members and 1 Guest are viewing this topic.

Offline rhbTopic starter

  • Super Contributor
  • ***
  • Posts: 3487
  • Country: us
Voltage reference noise processes
« on: August 03, 2019, 10:00:29 pm »
Before delving into the "supercalifragilsticexpialidocious"  world of continuum mechanics one needs a good understanding of the statistical properties of the dominant noise processes as well as other error sources such as thermal EMF.

Attached is a scan from "Physics of Semiconductor Devices"  by S. M. Sze, 2nd ed which discusses, thermal, shot and flicker noise.

The important points here are that thermal and shot noise have been experimentally and theoretically quantified.  That leaves only flicker (1/f) noise to be determined empirically.  And,  that shot noise is a signed quantity which depends upon whether the PN junction is forward or reverse biased (eq 91 & 92 and intervening text).  As a consequence, an accurate empirical estimate of flicker noise requires that this be properly accounted for.

* sze-noise.pdf (53.69 kB - downloaded 159 times.)

So the journey begins. "Have data.  Will travel." to paraphrase Palladin's business card.

Have Fun!
Reg
 

Offline RandallMcRee

  • Frequent Contributor
  • **
  • Posts: 542
  • Country: us
Re: Voltage reference noise processes
« Reply #1 on: August 03, 2019, 10:59:15 pm »
 
The following users thanked this post: iMo, WillTurner

Offline rhbTopic starter

  • Super Contributor
  • ***
  • Posts: 3487
  • Country: us
Re: Voltage reference noise processes
« Reply #2 on: August 03, 2019, 11:17:53 pm »
Thank you.  I have a volume of the same title with a different 2nd author copyright 1973.   I shall order a copy of the new edition. 
 

Offline rhbTopic starter

  • Super Contributor
  • ***
  • Posts: 3487
  • Country: us
Re: Voltage reference noise processes
« Reply #3 on: August 04, 2019, 12:17:52 am »
I just ordered the 3rd ed of "Physics of Semiconductor Devices".  The 2nd came out in 1981 so I thought it well worth while getting an updated edition of the canonical reference.

While the 2nd ed doesn't discuss popcorn noise, that's sufficiently distinct as to be easy to filter out.

The signed nature of the shot noise makes for an interesting question.  Does one remove it or leave it in?  Both are simple to do, so it's really more  a matter of convention in how one characterizes a voltage reference.

Does anyone know what the national labs do?  They obviously have chosen a convention.

Edit:  Upon further consideration, I realized that the expression for the RMS value of shot noise (eq. 91) is governed by the measurement process.  The bandwidth, B, and the total charge, q, depend respectively on the  sample rate and length of the data series.  As a consequence, any meaningful comparison of voltage references *must* account for the shot noise.  It *cannot* be left to convention.
« Last Edit: August 04, 2019, 01:46:06 am by rhb »
 

Offline tomato

  • Regular Contributor
  • *
  • Posts: 206
  • Country: us
Re: Voltage reference noise processes
« Reply #4 on: August 04, 2019, 02:12:52 pm »
Upon further consideration, I realized that the expression for the RMS value of shot noise (eq. 91) is governed by the measurement process.  The bandwidth, B, and the total charge, q, depend respectively on the  sample rate and length of the data series. 

q does not depend on the measurement process; it is the charge of an electron. 
 

Online Kleinstein

  • Super Contributor
  • ***
  • Posts: 14564
  • Country: de
Re: Voltage reference noise processes
« Reply #5 on: August 04, 2019, 02:58:57 pm »
...
While the 2nd ed doesn't discuss popcorn noise, that's sufficiently distinct as to be easy to filter out.
...

Popcorn noise is only distinct if it is the dominant noise with a single or a few steps. If more pop-corn type noise sources are superimposed, one gets just some 1/f noise with no way to tell the actual process behind it.
I faintly remember an explanation of 1/f flicker noise as a superposition of pop-corn like noise sources.

 

Offline rhbTopic starter

  • Super Contributor
  • ***
  • Posts: 3487
  • Country: us
Re: Voltage reference noise processes
« Reply #6 on: August 04, 2019, 05:40:35 pm »
1)  q in Sze's equations is the total charge transferred, not the charge on an electron.  My apologies for not making that clear.  I should assume that he chose to do that to avoid cluttering simple equations with integrals.  If the current is constant, then q = I*t.  It is generally unwise to make assumptions about the notation in books like Sze's.   I have spent far more time trying to sort out other people's notation than I have the subject.  And in cases, been forced to contact the author for an explanation.

2) I was puzzled by Sze's omission of popcorn noise, but in reading Motchenbacher & Fitchen, the 1/f property of popcorn noise is mentioned.  Sze clearly considered the matter sufficiently well described as 1/f noise.  I shall be very interested to see if the 3rd ed discusses popcorn noise separately.  It is surface defect related, so I should not be surprised if he does not.

3) I do not  comment on such graphs without an explanation of how the data were collected and processed for display.  A very large proportion of spectra obtained by FFT are wrong because the person did not understand how to prepare the data before passing it to the transform routine.   If you have a legitimate question I shall attempt a proper answer.  If you want to play games I'm sorry, but I don't care to play.

This is not a contest.  I am not competing with anyone.  I never have and never will.  I only compete with what I was able to do yesterday.  I'd like to get to my real interest which is the aging and hysteresis effects observed in voltage references as quickly as possible.  But to do that I have to clear away a bunch of underbrush such as measurement noise. I don't recall the title of the thread, but I addressed the issue of the aliasing of white noise by the ADC design used in the 34401A.  That results in the estimated spectrum of white noise having an increasing amplitude with frequency.  Any estimate of the exponent of 1/f noise must take that into account.

I have attempted to initiate a discussion such as I might raise with someone over food and drinks at a week long professional society meeting.  My standard reference is "Random Data" by Bendat and Piersol.  If you are not conversant with that, you are not likely to be able to follow the discussion.  If sensible responses are unobtainable, I shall drop the subject and work on my own.

Have Fun!
Reg
 

Offline tomato

  • Regular Contributor
  • *
  • Posts: 206
  • Country: us
Re: Voltage reference noise processes
« Reply #7 on: August 04, 2019, 08:39:57 pm »
1)  q in Sze's equations is the total charge transferred, not the charge on an electron.  My apologies for not making that clear.  I should assume that he chose to do that to avoid cluttering simple equations with integrals.  If the current is constant, then q = I*t.

Equation 91 in the book you linked is the standard expression for electronic shot noise, and in that expression q is the electron charge (1.6E-19 Coulomb).  You can check a dozen other references if you like -- they will all confirm it.
 

Online David Hess

  • Super Contributor
  • ***
  • Posts: 16972
  • Country: us
  • DavidH
Re: Voltage reference noise processes
« Reply #8 on: August 05, 2019, 01:15:38 am »
Popcorn noise is only distinct if it is the dominant noise with a single or a few steps. If more pop-corn type noise sources are superimposed, one gets just some 1/f noise with no way to tell the actual process behind it.
I faintly remember an explanation of 1/f flicker noise as a superposition of pop-corn like noise sources.

That is what I remember also.

It would not surprise me if popcorn noise is only discussed for historical reasons if at all because it should never be an issue with modern parts.  Below are photographs of popcorn noise coming from inside the vertical channel switch of a late 1980s Tektronix oscilloscope.  It took weeks to track down the exact part because the noise was intermittent over a period of hours.
 

Online Kleinstein

  • Super Contributor
  • ***
  • Posts: 14564
  • Country: de
Re: Voltage reference noise processes
« Reply #9 on: August 05, 2019, 06:09:41 am »
Not really modern, but still used and produced, the LM399 references show quite some popcorn noise. I would even say that white noise plus popcorn (maybe 2 processes) could be a good approximation.

Popcorn noise is not limited to semiconductors even some resistors show popcorn noise, when there is a current flow.
 

Offline rhbTopic starter

  • Super Contributor
  • ***
  • Posts: 3487
  • Country: us
Re: Voltage reference noise processes
« Reply #10 on: August 05, 2019, 02:07:29 pm »
tomato:  Yes,  q in eq. 91 is the standard usage.  For some reason the phrasing chosen made me think otherwise. In any case it does not change the conclusion that one must account for the variation in shot noise between references.

imo: Before you can even begin to quantify 1/f noise you have to do a lot of data prep.  I'm still trying to sort out the extent of that.  The obvious first order problems are temperature and environmental noise.  And there is far more to doing a spectral estimate than just computing an FFT.  I have a 125 page book of theorems related to the Fourier transform for when I *really* need to be sure of what it says in the fine print.  In practice, once one has worked out using it for a task you never have to check again.  But in a case like this where I am doing something I've never even given serious thought to before I shall be reading it very carefully.

The 34401A uses LM399s, so LM399 noise is embedded in any data collected with them.  Along with a lot of other error sources.  So peeling the onion to get to the continuum mechanics level will take considerable work.

This thread will likely go dormant for a considerable period while I build a scanner and write software to collect data.  At the moment I'm just trying to figure out how much material I don't know that I need to master. That will let my subconscious work on some of it and allow time for the required reference material to accumulate as well as parts for side experiments.

Sourcing  PCB edge connectors for the 44421/2A relay boards is very difficult.  When 3497A systems got scrapped, the terminal blocks generally went in  the scrap wire bin.  PCB edge connectors are not generally used in new designs so finding them is a very tedious NOS hunt.  I did find a place that claimed to have some, but never got a response to my email.

Have Fun!
Reg

 

Offline MaxFrister

  • Regular Contributor
  • *
  • Posts: 140
  • Country: 00
Re: Voltage reference noise processes
« Reply #11 on: August 05, 2019, 02:47:30 pm »
At first glance, it looks like a generic edge connection.  Digikey shows 200k kinds in the world with 1.5k in stock.  It should be possible to find something that fits.  Remember that it can have more contacts than necessary.

The 3488A and 3499B mainframes are much more common and it is possible to find cards with the connectors still attached.  Have you checked to see if those cards will fit your mainframe?  The older ones have similar part numbers.

It is always possible to solder wires directly to the boards.

I recently designed a connection board for the N2262A.

« Last Edit: August 05, 2019, 02:59:50 pm by MaxFrister »
 

Offline rhbTopic starter

  • Super Contributor
  • ***
  • Posts: 3487
  • Country: us
Re: Voltage reference noise processes
« Reply #12 on: August 05, 2019, 07:43:46 pm »
I have a backplane card from a 3497A, so I'm set on that end.  It is the other end that I need.  Both are hard to find.  I've spent many hours looking.  Now if I wanted 10,000 pieces it would not be a problem.  But I just want a few.

Board edge:

Width: 4.605"
Depth: 0.305"
Thickness: 0.068"

2 x 36 pads at 0.1218" spacing center to center (measured across 36 pads)

I'd steal a connector from one of my 44421A terminal blocks before I'd solder to the pads.

I had a part number, but it is MIA at the moment.
 

Online David Hess

  • Super Contributor
  • ***
  • Posts: 16972
  • Country: us
  • DavidH
Re: Voltage reference noise processes
« Reply #13 on: August 06, 2019, 06:30:25 am »
Not really modern, but still used and produced, the LM399 references show quite some popcorn noise. I would even say that white noise plus popcorn (maybe 2 processes) could be a good approximation.

I have noticed that before.  Is there something inherent about the LM399 other than low noise which makes for greater popcorn noise?  Or is it just that there are a lot of old LM399s out there which were manufactured before the cause of popcorn noise was understood?  What about other buried zener references?
 

Online Kleinstein

  • Super Contributor
  • ***
  • Posts: 14564
  • Country: de
Re: Voltage reference noise processes
« Reply #14 on: August 06, 2019, 10:03:47 am »
An LM399 without the popcorn noise would be great. However I don't think even the selected ones for the better DMMs reach no popcorn noise, but only lower levels. I saw popcorn noise with LM399 with date codes from 2017 and 2018 (not sure if both, but visible in the ratio). So it is not a thing of the old ones only - newer ones may be better.

I an not so sure there is a single possible source of popcorn noise - there are different possible defects / impurities that can cause such noise, though with different amplitude and frequency range. Even if know which defect is the really bad one does not mean one is able to 100% eliminate it.
TiN and a few other had LTZ1000 that did show popcorn noise. I would be surprised if other reference can be made free of popcorn noise. Even of not obvious it may still be there. The popcorn noise in the LM399 is also often relatively small amplitude - so if one does not look very careful, one may not recognize it as such. The classical cases where more obvious.

AFAIK to get visible pop-corn noise it takes a defect at a position where it has a large effect so that such single electronic state has an visible effect. Chances are there are many other positions, where a single defect is not directly visible, but such smaller effects can still add up.
 

Offline iMo

  • Super Contributor
  • ***
  • Posts: 4987
  • Country: cv
Re: Voltage reference noise processes
« Reply #15 on: August 06, 2019, 01:04:46 pm »
It could be all references made with the ion implantation process (ie. burried zeners) may show popcorn noise. You shoot ions with pretty large energies into the lattice such the damages are rather high. They recover somehow in time (ie. because of a poor thermal annealing damage recovery of the particular sample) producing popcorn noise, imho.
I've been seeing 2-3uV jumps in an LT1021-5, pretty slow process (10PLC), however.

https://en.wikipedia.org/wiki/Ion_implantation
« Last Edit: August 06, 2019, 01:49:57 pm by imo »
 

Offline rhbTopic starter

  • Super Contributor
  • ***
  • Posts: 3487
  • Country: us
Re: Voltage reference noise processes
« Reply #16 on: August 06, 2019, 01:48:53 pm »
According to Sze, ion implantation damage to the lattice structure is removed by annealing at 700 C or less.  Laser beams are preferred as that allows controlling the depth and location.  However, that reference is almost 40 years old.

According to AoE 3rd, popcorn noise is very rare in modern fab processes even with old designs such as the 741 op amp.
 

Offline iMo

  • Super Contributor
  • ***
  • Posts: 4987
  • Country: cv
Re: Voltage reference noise processes
« Reply #17 on: August 06, 2019, 01:56:10 pm »
Quote
According to Sze, ion implantation damage to the lattice structure is removed by annealing at 700 C or less.  Laser beams are preferred as that allows controlling the depth and location.  However, that reference is almost 40 years old.
There is not such thing like "removed" in microelectronics. The entire semiconductor business is based on "probabilities" and large effects caused by small amounts of impurities. The annealing at 700degC may remove 99.98% of the damages per wafer, but the remaining few will cause the popcorn noise in the LM399..
 

Offline rhbTopic starter

  • Super Contributor
  • ***
  • Posts: 3487
  • Country: us
Re: Voltage reference noise processes
« Reply #18 on: August 06, 2019, 01:57:39 pm »
I'll let you argue the point with Dr. Sze. c.f. p 66 2nd  &   AoE 3rd p 477
« Last Edit: August 06, 2019, 02:38:04 pm by rhb »
 

Offline iMo

  • Super Contributor
  • ***
  • Posts: 4987
  • Country: cv
Re: Voltage reference noise processes
« Reply #19 on: August 06, 2019, 04:26:53 pm »
Dr.Sze understands what we talk about, sure.
The LM399 is a mass produced cheapest burried zener Vref today, afaik. ADI is not messing too much with it, imho, they anneal once, they test the heater and 6.9V output, when ok the chip pass. The production costs are in XXcents levels (ie. the yield 90%). The chip is intended for general use in the industry and with mainstream meters till 6.5digits where the few uV popcorn does not mean much (moreover it is a "high noise" device, perhaps because of wide spectra of popcorns as Kleinstein mentioned above).

The production process of the LTZ1000 and LTFLU is different, imo. They care much, may be they do several special additional processes after the ion implantation, and provide careful tests. The yield could be much lower (easily 5-10%), therefore the cost per chip could be is ZZx higher. It does not mean the LTZ1000 or LTFLU is popcorn free by design ,however.
« Last Edit: August 06, 2019, 04:50:49 pm by imo »
 


Offline rhbTopic starter

  • Super Contributor
  • ***
  • Posts: 3487
  • Country: us
Re: Voltage reference noise processes
« Reply #21 on: August 07, 2019, 12:24:46 am »
The best reference by far is:

Random Data
Bendat & Piersol
Wiley

I have relied on it for 30+ years starting in grad school with the 2nd ed. and continuing on  through the 4th ed. which will be the last.  Allan Piersol passed away while it was in preparation and Julius Bendat deserves to retire from the job.  No matter how exotic the problem, it is likely they have a brief treatment.  They even treat sign bit, truncation of an analog signal to 0 or 1, in the 4th ed.  Sam Allen would be pleased.

While there are many more important cases treated in the 4th ed., if you're tight on money, an older edition will serve you very well.  I have most of the major texts which consume about 20 ft of shelf space.  I almost never look at the others.  The only real reason to look at them is someone cites them and I want to check the citation.

Have Fun!
Reg
 

Offline graybeard

  • Frequent Contributor
  • **
  • Posts: 437
  • Country: us
  • Consulting III-V RF/mixed signal/device engineer
    • Chris Grossman
Re: Voltage reference noise processes
« Reply #22 on: August 09, 2019, 05:11:06 pm »
Over the years I have observed popcorn (or burst) noise in both Si and III-V bipolar and FET devices.  It is almost always wafer.dependent and rare in occurrence.

If you have a batch of devices that exhibit burst noise, dump them and get more from another wafer or production lot.

Offline rhbTopic starter

  • Super Contributor
  • ***
  • Posts: 3487
  • Country: us
Re: Voltage reference noise processes
« Reply #23 on: August 09, 2019, 06:24:11 pm »
That comports with the comments by H&H in AoE 3rd.
« Last Edit: August 17, 2019, 06:17:43 pm by rhb »
 


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf