[volt-nuts] Some questions to zeners (1N823-1N829)

[WarrenS]

Andreas

1st my two cents worth concerning comments of others that are being more 
practical than nutty.

"it is not easy to achieve these specs",
Too True,
If you do not already know this, then I'd suggest you back off and do some 
major readjustment of your requirements.
Assuming you already know it is not easy and why, then
IMHO, what you want to make is doable with the right compromises, some of 
which being
A little extra power.
a lot of extra time
a lot of test selecting
some extra circuits or S/W

"your expectations are not realistic "
True,  if you think it can be done from a data sheet and a few of off the 
shelf parts in a production device.
Not true, if you want to go totally nuts and have the time to do it.


"The thermoelectric effects causing error would swamp the performance"
Could be true for some. If you do not already know enough to keep these 
effects well under <<0.1PPM then should find another starter project.


"To keep everything below the 1 ppm/deg C range you would have to put the 
entire circuit in controlled temperature"
Too true, But for these requirements, fortunately you do not have to keep 
*Everything* under 1 ppm/deg C
ALL you have to do is to keep the sum total of Everything under 1ppm.  Big 
difference!
And that is very easy to do, by a fact of ten plus if desired.


"long term drift and noise will be intrinsic to the devices, and 
unpredictable except in a statistical sense."
Half true,
Using the *right device*, and a lot of time, noise can be low enough to be 
mostly insignificant
and long term drift will follow a predicable slope.
It ain't easy and it's going to take a lot of time. If you think otherwise, 
you need a new project..


"provide near zero tempco at one temperature only"
Answer too limited to be useful
zero TCing a circuit, can provide no voltage difference at any two 
temperatures, (to the limit of the repeatability)
Depending how flat you need it between these two temperatures, is the only 
issue.
Best to plan on having a second order  TC method as well.
A total max total deviation of 1ppm and a 0.1 PPM /deg is not too hard to 
get.


"One of the best voltage standard Datron 4910AV (4x LTZ1000) have only 1 
ppm/year drift"
OK, so that shows there is at least one way to do it, so it is not 
impossible.
Now just need to find the best way for you to do it.

"Don't bother with TC zeners"
Not a bad idea, but Unfortunately Not a lot of other choices considering 
your requirements.
The other choice you have (that you should consider if time is a big issue)
is to get some three terminal Fluke voltage references.
If necessary by removing them from old test equipment such as Fluke 731
Those have already been selected and aged, and you'd just then need to work 
on the long term drift compensation selection method.


"There are lots of nice IC references available"
True, and if you can live with data sheet specs, they are much better 
choices.
But hard if not impossible to find anything that will compare to 1/F pop 
corn noise, and long term stability


"I doubt that any TCZ will match an LM399"
True when considering a wider temperature range, and it sure makes things 
easier,
so a good suggestion if TC was your main problem.
But TC need not be a problem, and most any good zener will outperform most 
any selected LM399 in low freq noise and stability.


"so would have to be ovenized to get *best* performance".
True of Any circuit when "best" TC is concerned, so not very relevant,
The question is can it be made good enough without an oven?
And the answer is defiantly yes with the right tempco circuit.
And if you want it even better, can make a very low power mini-oven.

Back to your email ******************

> Do you have typical values over a 64-90 ?F range. Will it be above 1ppm/K 
> or  below?
I do not have any data over that Temp range in front of me.
Over My room changes (about 1/2 that temp range), can keep the Total voltage 
error down to around 1PPM, (not just /deg)
If it is an important consideration, I may try and see what happens over a 
wider range.
For a backup plan, considering having an addition second order TC 
compensating method such as S/W.
Overall compensating to 1ppm / C is no problem,
which is Total change +- 7PPM over a +- 7 deg C range
Ten times better is possible over narrow temp ranges like that.

It is the hysteresis and stability that is going to be a limiting factor, 
and last I looked,
The zeners I tested had no measurable Hysteresis over a much wider temp 
range than that.

>>From your plot it would be 0.33ppm per 3 degrees in narrow range i.e. 0.1 
>>ppm per degree.
Correct, but not a relevant measurement.
ANY straight line TC as seen there can be zeroed out.
That was just a pre-test plot to see if the part was low enough noise to do 
a more accurate Zero TC.

> By the way: is it degrees Fahrenheit or degrees Celsius (= 3 Kelvin)?
My world is mostly in F, and since I was not even plotting temp at the same 
time, pretty irrelevant for that data set.

> If I understand you right then you would not use this device because it 
> does
> behave other than the others?
Your understaning is not correct, not even close. My comment applies to a 
preselection process of 1N82x parts.
In the case you stated, your preselection process (assuming you have enough 
to pick from) would be to select all the parts that are less than say 1PPM, 
with maybe a 20% yield.

If the yield is too low then make a second pass for parts that are say 2 +-1 
PPM,
and in that case, then yes the "best" part would not go into that batch, 
because it will likely need a different compensation than the others.


> On the other side it seems to be the device with the largest ageing rate 
> of  the 5 pieces.
likely too little data to be important, Just as likely random luck as 
anything important.
Until after you do some pre-aging, (whatever that may mean for that part) I 
would not even bother looking at that data this early on in the selection 
process.


> So I still hope that anyone has experiences with hysteresis of the zeners.
Turn on after power down, and hysteresis, repeatable, etc altogether is 
under 1PPM.
As part of my pre-test TC procedure, I hit them with cold spray to 0 C and 
heat them with a heat gun to ~50C, a few times,
If they are not repeatable they are not used in further testing.

I have a question about two of your requirements
AJ> tempco below 1ppm/K
AJ> hysteresis in the 10-40 degree range well below 1ppm

This suggest to me that what you are really planning on making is something 
with an overall compensted TC that is well below a 1 ppm/K,
otherwise your hysteresis requirement does not make a lot of sense to me.
So if what you are really after is more like < 1 PPM total error over time 
and temp, all this extra trouble now makes a lot more sense.


Now the disclaimer, I have no idea if any of the now available 1N825 or 
better parts work like I've described.
It is VERY much a manufacture sensitive thing.
Short term noise being the biggest rejecting thing I've seen in the past.
0.5 PPM (3uv steps) are not untypical for some batches.

If someone that has enough of these parts from a single batch and 
manufacture to make it worth while testing them, and will make them 
available to others, then I'll test and report the results and compare them 
to what I have.

There may be another less direct way to get where you want to go (You did 
not say where that was)
Divide the task into two or more sections.
As an example, One being the low power short term stable DVM device that 
works over a limited temperature.
There would be no problem getting 0.1 PPM accuracy/repeatable for that part.
and a second higher power device that is mostly off, and powered up Just 
long enough to calibrate the low power xfer DVM above.

And if you can then include some sort of cal exchange program for the higher 
powered but less used device, this would make a very hard project into a 
relative easy project.


ws

> ----------------------------------------------------------------------
>
> Subject: Re: [volt-nuts] Some questions to zeners (1N823-1N829)
>
> Hello Warren,
>
> many thanks for your valuable response.
>
>> If you want to make something in volume
> The planned number of devices is more in the range of 2 up to four.
>
>> one of only a very few possible solutions.
> I guess that the others are a LM399 or a LTZ1000 based solution (both with
> heaters).
>
>> Don't just solder the parts down on a PCB without a stress relieve loop 
>> in
>> the leads.
> Thats a good hint otherwise I will get the humidity changes of the PCB as
> stress on the device.
>
>> The zero TC current can be set so that the voltage at most any two
>> temperatures will be the same. (<< 1PPM)
> Do you have typical values over a 64-90 ?F range. Will it be above 1ppm/K 
> or
> below?
>>From your plot it would be 0.33ppm per 3 degrees in narrow range i.e. 0.1
> ppm per degree.
> By the way: is it degrees Fahrenheit or degrees Celsius (= 3 Kelvin)?
>
>> From a given batch, any that are considerable different, I don't use
>> because they may have something else wrong going on.
> Thats interesting since I have 5 pieces of the brand new LT1236AILS-5
> devices.
> 4 of them have a tempco of  2-3 ppm/K around room temperature.
> 1 piece has a very flat tempco of around 0.2ppm/K (picture attached)
> If I understand you right then you would not use this device because it 
> does
> behave other than the others?
> On the other side it seems to be the device with the largest ageing rate 
> of
> the 5 pieces.
>
> One experience that I did is that devices out of one batch which have a 
> low
>  tempco around room temperature tend to have a larger hysteresis and vice 
> versa.
>
> So I still hope that anyone has experiences with hysteresis of the zeners.
>
> With best regards
>
> Andreas
>
> *****************************************

Desired requirements from Andreas:

AJ> I want to have high end specs with the power consumption
AJ> of a battery supplied DMM.
AJ> The planned number of devices is more in the range of 2 up to four.
AJ> that is:
AJ> tempco below 1ppm/K
AJ> hysteresis in the 10-40 degree range well below 1ppm
AJ> no sensitivity to humidity (so all plastic housings will fall out)
AJ> ageing in the range of 1 ppm/year after some pre-ageing

*************************
>
>> Andreas
>>
>> If you want to make something in volume which is just pretty good, I 
>> would
>> not recommend this method for a new design.
>> On the other hand, if this is a nuts thing to make the very best, this is
>> one of only a very few possible solutions.
>>
>> Here are some general answers, most of my experience with these parts is
>> pretty dated, (i.e long ago).
>> As when pushing the performance limit of any reference, there is a  lot 
>> of
>> variation between parts and even more so between manufactures.
>> Solution is, select, Age, select, test, grade, select.
>>
>> As far as my experience with 1N823, performance depends on the run and
>> what
>> is left after the manufacture has selected out the others.
>> With 1N823's, Yield can often go to zero. With 1N825's a typical yield
>> I've
>> seen is around 25-50% (from the right manufacture and batch)
>>
>> Yes the main difference is the zero TC current, with some parts there is
>> no
>> zero TC current.
>> So yes you are more likely to get a lower current TC such as 5 ma from a
>> 1N823 or 1N825 than a 1N829, but it could of course be > than 7.5 ma.
>> I don't use anything that does not have a zero TC between ~ 4 and 10 ma
>> I found TC to be very much a Batch thing, with up to 50% of the majority
>> of
>> a batch, tending to be similar.
>> From a given batch, any that are considerable different, I don't use
>> because
>> they may have something else wrong going on.
>>
>> Another thing that needs to be selected for in high end references, and
>> will
>> vary by manufacturer, is 1/F noise.  The random jumps in the voltage.
>>
>> For me, hysteresis has not been a issue over room temperature changes for
>> the most part, but something that has to be checked.
>> Some Manufactures are better than others, and hysteresis can and will be
>> effected by assembly, layout, or anything that puts any stress on the
>> part.
>> Don't just solder the parts down on a PCB without a stress relieve loop 
>> in
>> the leads.
>>
>> The zero TC current can be set so that the voltage at most any two
>> temperatures will be the same. (<< 1PPM)
>> If the voltage change in-between those two temperatures is too much, lots
>> of
>> ways to add an additional second order temperature compensation.
>>
>> For the best TC performance, consider the mini-oven idea with the zener,
>> heater resistor, and thermistor all  heat shrunk together.
>> With a lot of outer insulation, it could be done low power by adding an
>> addition 0 to 5 ma to the heater resistor.
>>
>> Everything has it's trade offs.
>> The trade off using these zeners is time and complexity.
>> For these parts, 5ma is about as low as you are going to get.
>> For low power, there are many things Much better.
>> The trade off you make to get low power is "Noise" & stability.
>> The trade off you make for the good TC of LM399 is long term stability,
>> PPM
>> noise, and the high cost of selection fall outs.
>>
>> The best solution will depend on many things including the desired
>> performance, how many you want to make, the cost you place on selection
>> time,
>> and if you can still find the 1N825's at a reasonable price like they
>> where
>> in the 70s & 80s. (&  $0.10 in 2000s)
>>
>> For a xfer standard, the Most important criteria is 1/f random noise. 
>> Most
>> everything else can be compensated out.
>> For that, it is hard to build anything better than using a 1N825 selected
>> device.  Plot attached.
>>
>>
>> ws
>>
>> ************************************
>>> From: "Andreas Jahn" <Andreas_-_Jahn at t-online.de>
>>> Subject: [volt-nuts] Some questions to zeners (1N823-1N829)
>>>
>>> Hello all,
>>> Hello Warren,
>>>
>>> after having experimented a lot with 5V monolithic zener references
>>> and still not found the ideal solution I want to try a 1N82x based
>>> solution.
>>>
>>> For me a extended room temperature range of
>>> 25 degrees centigrade +/- 7 degrees (64-90 ?F)  is of interest.
>>> Since I plan to have battery supplied instruments a lower supply current
>>> would be of interest.
>>>
>>> For the zeners a zero TC current is stated.
>>> Over which temperature range the TC is nearly zero.
>>> How large is the voltage deviation in the above mentioned range?
>>>
>>> Does it play a role for the absolute temperature deviation if a 1N823 or
>>> a
>>> 1N829 is used?
>>> Or is the behaviour equally when the individual zero TC current is used?
>>> Is the only difference between the selections that the zero TC current 
>>> is
>>> more near the 7.5mA value on the 1N829?
>>>
>>> So is it more likely to get a low zero TC current of 4 mA on a 1N823
>>> device than on the 1N829?
>>> Or should I go for the 1N829A for the lowest absolute TC?
>>>
>>> How large is the hysteresis on the zeners in a temperature range of 
>>> 10-40
>>> degrees celsius (50-104?F).
>>> On monolithic unheated reference voltages with hermetic case I have up 
>>> to
>>> around 2 ppm hysteresis difference
>>> on temperature cycling. (see attached picture with 10-45 degrees celsius
>>> on X-Axis for an ADC with a 5V reference
>>> measuring a LM399 heated reference over a 2:1 precision voltage divider.
>>> The ADC with the 5V reference is temperature cycled).
>>>
>>> I blame the temperature hysteresis on the die attach to the lead frame
>>> which seems to be usually a silver filled epoxy compound.
>>> I hope that the hysteresis on a discrete zener is much lower.
>>>
>>> With best regards
>>>
>>> Andreas