The LTFLU (aka SZA263) reference zener diode circuit

[BU508A]

Hello noopy,

thanks a lot for these very nice pictures and explanations on your website. 

Thanks to branadic  I was able to take some more pictures of his LTFLU:


https://www.richis-lab.de/REF04.htm

[...]

One remark about this sentence on your website: "In der rechten oberen Ecke befinden sich die Zeichen FLU1."

The two signs below the "FLU1" are coming from the logo from Linear Technology.

[Dr. Frank]

These RefAmps all are relatively tightly specified concerning their initial reference voltage span, but especially for the span of the transistors collector current (20..200µA) where zero T.C. can be achieved, over defined operating temperature span, with zener current being 3mA.

Therefore it's clear, that they have to heat the device in situ, in turn to trim the transistor and zener to the correct above mentioned parameters.

Frank

[Kleinstein]

The adjustment range for the transistor size is surprisingly large: 8 fuse to for a kind of 8 bit weighted sum for the transistor area.
The trim may need some heater to check the TC. However they could as well use the zener current

4 separate zeners with extra resistors for current sharing could be better than one large one where current sharing may not always work.

[Noopy]

thanks a lot for these very nice pictures and explanations on your website. 

Thanks for the positive feedback! 

One remark about this sentence on your website: "In der rechten oberen Ecke befinden sich die Zeichen FLU1."

The two signs below the "FLU1" are coming from the logo from Linear Technology.

You are right. 
I thought most people would recognize the LT-Logo... 

...

...

An interesting explanation. Heating for adjusting the number of transistors and finding the right tc.
But wouldn´t that cost a lot of time? And a good accuracy of the temperature would be difficult to get? Hm... 

[BU508A]

One remark about this sentence on your website: "In der rechten oberen Ecke befinden sich die Zeichen FLU1."

The two signs below the "FLU1" are coming from the logo from Linear Technology.

You are right. 
I thought most people would recognize the LT-Logo... 

Yes, sure.
But sometimes, you know, you'll never know who will read it.   

Edit:
And, to be honest, when I saw this logo the first time, the first thing which was coming to my mind was "Klingons? What the heck .. ?"   

Anyway, thanks again, I've learnt a little bit on how to look at all these integrated circuits and identify some parts. Sort of.

[try]

Hi Frank,

These RefAmps all are relatively tightly specified concerning their initial reference voltage span, but especially for the span of the transistors collector current (20..200µA) where zero T.C. can be achieved, over defined operating temperature span, with zener current being 3mA.

Therefore it's clear, that they have to heat the device in situ, in turn to trim the transistor and zener to the correct above mentioned parameters.

Frank

what exactly is done during the trimming process that you describe?
I have an idea how metal foil transistors are trimmed to achieve a certain resistance value but how do you trim zeners and transistors
to achieve a combined temperature coefficient near zero?

[Noopy]

Hi Frank,

These RefAmps all are relatively tightly specified concerning their initial reference voltage span, but especially for the span of the transistors collector current (20..200µA) where zero T.C. can be achieved, over defined operating temperature span, with zener current being 3mA.

Therefore it's clear, that they have to heat the device in situ, in turn to trim the transistor and zener to the correct above mentioned parameters.

Frank

what exactly is done during the trimming process that you describe?
I have an idea how metal foil transistors are trimmed to achieve a certain resistance value but how do you trim zeners and transistors
to achieve a combined temperature coefficient near zero?

They burn fusable links like these:



By burning the fuses they can adjust how much transistors are in series with the zener. With less transistors you have more current per transistor. The Tc varies with current, so when you adjust the current per transistor you adjust the Tc. => Bob´s you uncle! 

[Kleinstein]

The reference is not an exactly cheap part - though one a price is hard to tell for a part that is fluke exclusive.
So a little more time for testing would no hurt that much, as it would with a cheap OP.

Chances are they would measure the ref. voltage during warm up, possibly with additional power to the heater or a pulse of higher zener current. Form that one can estimate the TC and get the right number of transistors. There seem to be just the 1:2:4:...  ratio seta and no extra spare. So they likely do the adjustment in one step and not as coarse and fine tune, as in some cases there is no way back.  For some cases one could still try it in steps and burn the corse fuses first only.  The really fine tuning of the TC is done with the external resistor(s).

I am not sure if the is includes in normal production test, but the by far slowest test would be checking for excessive popcorn noise. This test could be separate already with the case.

[try]

Hello dietert1, hello Noopy, hello Frank!

Anyone with the capability to sample the zener voltages of an open LTFLU chip? I am wondering whether it makes sense to implement four seperate zeners on one die. This may be more or less useless because the zeners will be extremely similar.

Regards, Dieter

Similar in relation to what property or properties exactly?

@Noopy:

I assumed Frank was talking about another treatment than just triggering fuses.
I thought he was refering to manipulating zeners and transistors directly.

Why bother with all these tedious micro adjusting stuff when I can set the collector current externally?

From my layman's view a reference amplifier has three connectors. What is "e" used for?

Best regards
try

[Noopy]

I assumed Frank was talking about another treatment than just triggering fuses.
I thought he was refering to manipulating zeners and transistors directly.

Hm, I don´t know what that should be. But let´s wait for an answer from Frank.

Why bother with all these tedious micro adjusting stuff when I can set the collector current externally?

In the circuits I know you adjust the current through the zener and the transistor but you can´t or don´t want to adjust the two currents on their own. *
With the adjusting of the current per transistor on the die you can be sure that the current through the active transistors is ideal to compensate the Tc of the zener.

* Not absolutely correct. You do adjust the current but only in a very small window.

[iMo]

@Noopy: what could be the structure between the zeners?

[Noopy]

@Noopy: what could be the structure between the zeners?

1. In my view the two structures are connected only to the metal layer above them (z/GND).

2. It looks like a transistor or something similar.

3. I didn´t find an obvious contact to the substrat on the die.

=> I asssume it´s something connecting ground to the substrate. Probably a pn-structure (shorted transistor) giving the substrat a little lower potential than ground to get a better isolation.

[dietert1]

I know a little bit about large mosfets made by paralleling lots of smaller mosfets on a single chip and about the negative TC that makes this possible (no unstable hotspots). I have also seen that two 7.5 V zeners in series get you lower differential resistance than one 15V zener.
But i have no idea whether and why it makes sense to divide a larger zener into several smaller ones. I know very little about the current sharing resistors or to what extent they are doing something meaningful. I think you want the current sharing resistors as small as possible for obvious reasons. If they are "small" and the zeners very similar, is there still some reason to divide the zener into four regions? Very similar means equal structure resulting in equal electrical and physical parameters (knee shape, TC, differential resistance etc.)

For me these are the open questions with respect to LTFLU. If possible someone should measure the current sharing resistors and the zener voltage differences of a LTFLU.

Regards, Dieter

[iMo]

The 4 zeners are not the same. They are always different even on the same die. They do it to lower the noise (2x in this case) and statistically TC.

[Dr. Frank]

Well,
The external fine trimming of the overall T.C. by changing the collector current from 20 to 200µA will change the -2mV/°C very lightly , I would estimate maybe from -2.0mV/°C to -2.1mV/°C.
This should mostly account for the variations of T.C. at different operating temperatures, either 23°C nominal for non stabilized use case, and about 45°C for an ovenized use case.

For this RefAmp, the T.C. of the zener and the transistors BE diode then have to match exactly in terms of these -2.0 ..-2.1mV/°C for the transistor and +2.0..+2.1mV/°C for the zener, which has about 6.4V.
When you convert these latter values for the zener to a relative T.C. = dU_ref/U_ref, that gives a quite narrow window of +312 .. +328 ppm/°C, which the zener structure has to be trimmed for, otherwise you will not get an overall zero T.C.

Even if the collector current variation changes the T.C. of the transistors to a much higher degree, you will anyhow get a quite narrow span for the zener, due to its 13 times bigger voltage.

I don't know, how you can change a zeners T.C., but one parameter is its absolute voltage.. the 1N821..829 have nearly zero T.C @ 6.2V, and higer zener voltages will have more positive T.C.s.
The T.C. of these type of zeners vary greatly from 5ppm/°C (1N829) to 100ppm/°C (1N821), which is accomplished by selection only. Here you see, that precision zener diodes usually have a big manufacturing variation.
So LT had to invent something to precisely trim this T.C. of the buried zener to the desired value.

Frank
 

[Noopy]

Hello Frank,

is it really necessary to trim the T.C. of the zener?
Trimming the T.C. of the transistor to match both should work too?

Greetings!

[Dr. Frank]

Hello Frank,

is it really necessary to trim the T.C. of the zener?
Trimming the T.C. of the transistor to match both should work too?

Greetings!

That will not work at all.
I think it's impossible to shift the T.C. of a Si pn junction by a factor of 13 by any means.
Check the appropriate diode equations.

Frank

[Noopy]

Hello Frank,

is it really necessary to trim the T.C. of the zener?
Trimming the T.C. of the transistor to match both should work too?

Greetings!

That will not work at all.
I think it's impossible to shift the T.C. of a Si pn junction by a factor of 13 by any means.
Check the appropriate diode equations.

Frank

Right, didn´t see that…

But you don´t see any trimming around the zener.
Perhaps they managed to establish a manufacturing process that is extremly good and then there is some sorting?

[iMo]

FYI - An info on 1N821-9 and other "zero TC" diodes. They combine a zener with one or two or more pn junctions to get near zero TC.
https://www.microsemi.com/document-portal/doc_download/14616-zero-tc-reference-diodes

[Dr. Frank]

@Noopy: by these fuses you can divide the external collector current further down binarily, by sending it through 1, 2...., 16 transistors. As the Shockley equation gives a logarithmic behaviour for the T.C. (I think?!!) this 16 fold current variation will still not give a 13 fold T.C. variation (I assume!??)

@imo: well, I know that already, but does that change anything about the conclusion, that zener diodes obviously have a great production variation of their T.C. ?

[iMo]

@imo: well, I know that already, but does that change anything about the conclusion, that zener diodes obviously have a great production variation of their T.C. ?

I think the LUTFU is not trimmed for Voltage, the voltage comes from the process variation as-is.
They trim for zero TC by selecting the 1..256 emitters at a specific temperature given by the customer.

[Kleinstein]

Changing the transistor area is vey similar to changing the transistor current. The voltage is set by the current density. The transistor area change may be good for a factor of 100 at most, probably better less, not to make the transistor too small and thus more 1/f noise. There are not just several large transistors but also a few smaller ones. A factor of 100 in the current shifts the voltage by a little over 100 mV or the TC by 100 mV/300 K = 0.33 mV/K.  Changing the external current by a factor of 100 would be possible too, but would change the circuit more than just a simple change in resistor value: the FB divider may have to get lower impedance with more base current.

100 ppm/K for the 6.5 V zener voltage is some 0,65 mV/K, still much smaller then the about 2 mV/K of zener or diode individually. Still this would be too much to adjust with the transistor current density. So it needs reasonable tight control over the process too. Here the buried zener may be more predictable than a classical zener in the 1N82x. At least the LM329 usually come on 50 ppm/ K for the lowest grade - so this seems possible. To get zero TC with a useful transistor size and current one may be able to adjust some 0.4 -0.5 mV/K or some 70 ppm. So the production TC should be within some +-30 ppm/K, preferably smaller.  This may mean that the yield may not be 100% but could still be good.

The "E" contact is there so one can set the transistor independent from the zener current.
For the zener diodes at higher current density the TC does not change very much anymore. So there is some room for changing the zener TC, but this would also effect the noise level and heat. So the practical range is smaller (e.g. 2-10 mA).

[Dr. Frank]

Yes, I have forgotten to take the transistor areas into account.
Therefore that structure is analogue to a collector current division from /1, /2, /3, ..., /256. 
I still wonder, if the T.C. can be changed that much, as required.

[iMo]

I've done a naive simulation of my model from past, removed the Q9 and the resistors to its base.
I've changed the number of emitters from 4 to 192 and run a Temp sweep 20-70C of the entire voltage reference, not touching anything else.
The TC uV/C is slope of the Vref from 20-70degC.
Below the TC (y axis) of the Vref vs. number of emitters (x axis).

[branadic]

Interestingly LTFLU is lower in voltage than any other zener reference and tighter in spec for output voltage. Attached a table I once found on the web.

-branadic-