Ultra Precision Reference LTZ1000

[Andreas]

Hmm,

did anyone already recognize that the Package of the LTZ1000 (non A) has connection to the chip?

Today I measured 0.477V between housing and cirquit ground which is nearly the same as R1 = 0.465V = Pin 4 of the LTZ1000.
Resistance between Pin 4 and housing is around 65 Ohms (both directions).
So it seems that the housing is high ohmic connected to substrate.
I could also draw ~5mA between housing and ground in mA-Mode of the DMM.
(of course the output voltage changed immediately).

I had expected that the housing is connected to pin 7 (most negative point) but not at a voltage somewhere in between.
So you have to take care not to make a short cirquit via the housing.

with best regards

Andreas

[Edwin G. Pettis]

@Kleinstein.....

Why do you think multiple resistors are needed in the temperature setting divider?  ±0.1% with low TCR is all that is needed, anything more is not only overkill but quite unnecessary.  The temperature setting does not require that precise of setting for the chip temperature, ±0.1% is already a fraction of a degree, and the on chip temperature isn't that accurate according to spec, what more do you need?  Stability is not dependent on tolerance, only resistor stability and even there it isn't all that tight.  Most multi-resistor chips that offer very low TCR tend to be more expensive than a couple of ±0.1% resistors,  PWW or film.

[Kleinstein]

Temperature setting does only need a divider.  Using 8 equal resistors from an array may still be an attractive option and the price can be competitive: The ones I had in mind are NOMCA/NOMCT or 2xMORN with 8 x 5 K as a lower cost option (some $4) and 2xLT5400 (2x some $8), but 10 K. The absolute TCR may not be that good, but the matching usually is. The more problematic parameter with individual resistors is the long time drift anyway, not so much the TCR. Naturally data on long time drift are tricky so the TCR get more attention as the long time stable resistors usually are also low TCR.

No need for high accuracy, so no need for the higher grades with even better matching. The easy to reach combinations and thus temperature settings are limited. 1:13 looks like the most practical to reach and one can get it with combining always 2 resistors in pairs, so that one can also use 2 arrays of 4. The main downside is the higher impedance, especially with the LT5400.

It makes sense to have the case connected to the substrate. This is kind of the normal case if the chip is soldered to the case. Thus planar BJTs usually have the collector at the case / back tap. JFETs tend to have the gate (which is the substrate) at the case.

[Andreas]

Hello,

I had a strange "ageing effect" on my LTZ#2 which I cannot completely explain.
The 9 year old battery pack was now rather weak.
And last week I also forgot to recharge over night.
So it happened that during two of my daily automated measurements that LTZ#2 was switched off by the battery monitor.
That is no problem for a absolute measurement.
But I am also doing difference measurements between two references as sanity ckeck for my ADCs.
The ADCs have a unbuffered LTC1043 2:1 divider at the input.
To protect the references in case of failures I have two 4,7K resistors between the references and the LTC1043 2:1 divider.
The voltage is checked to be within 1 mV against the expected target value. If the voltage is ok two relais contact short the 4K7 resistors to reduce noise.
But if one of the references is switched off there is a path with about 10K between the active and the switched off voltage reference. So the switched off reference is back-supplied by about 0.5 mA from the resistors and the input protection diode of the LTC1043.

In my opinion this should be no problem but on the measurement after exchange of the battery pack with a new one I recognized that the output voltage had increased by about 3.1ppm. (see day 2200)
Interestingly that is the opposite direction which I usually get when shorting the unbuffered output of the LTZ. (see day 1250)

with best regards

Andreas

[Kleinstein]

I also would not expect back-feeding the LTZ1000 circuit with a limited current (less than the zener current)  to be a problem.
How does the change compare to a simple power down and thus thermal hysteresis ?

Shortening a LTZ1000 ref circuit is different: removing the output upsets the temperature regulation and would turn on the heater full power. The damage / change could than be from over-temperature. Discharging some caps could be another problem - though I would not expect this to cause damage.

[dietert1]

Andreas, your graph shows several years of data with two events, right? The more visible -5.5 ppm step happened in 2017, but you are now talking about the +3 ppm step at the end of the curve. This time the event seems to revert some of the previous aging. Could this be an effect of handling the devices? Are you regularly handling the devices or do you keep them in a "don't touch" corner?

Would be interesting to see the curve after some more years. This business really requires patience.

Regards, Dieter

[Dr. Frank]

Hello Andreas,

when you overheated the LTZ1000 circuit at day 1250, you probably introduced a permanent hysteresis, which is of the type that does not slowly drift back by itself. The decrease of the output after overheating is the correct direction, anyhow.
Now, as the oven failed after years, the LTZ maybe simply jumped back in direction to the initial state, i.e. before day 1250.
That's also the correct direction, because if you cool down the LTZ1000, like below room temperature, or even below 0°C, its value will always go upwards.

The non-A version is more likely to show such hysteresis, whereas the A version often shows short dips, when not initially pre-treated correctly.

Maybe you remember, that John R. Pickering exactly does this in his 7000 reference, cycling the oven between room temperature and maybe 20..30°C above the set oven temperature of about 50°C, to get rid of any hysteresis.

I also don't think that it has anything to do with the back fed current.

Frank

[Andreas]

Could this be an effect of handling the devices? Are you regularly handling the devices or do you keep them in a "don't touch" corner?

Ok, I did not tell the complete story:
Usually my LTZ#1 + LTZ#2 references are running 24/7 on my measurement table and are only measured once per day with a wired relay multiplexer.
Once a week I also do a manual comparison against LTZ#3-LTZ#8.
1-2 times a year I try to get a calibration either from a friendly volt-nut or when the calibration guy at work calibrates stuff at our location with his Fluke 5520A.

The last successful automated measurement was on tuesday.
On wednesday + thursday the battery of LTZ#2 was down during measurement in the morning.
On friday I transported some of my references to calibration with the Fluke (unfortunately no measurement in the morning).
At work after 2 hours temperature settling I recognized that the battery voltage was unusual low on LTZ#2 (14.8V) although I had charged in the morning before transport.
At home doing the daily measurement after recharging I finally found out that against tuesday the voltage had increased by +22uV.
With the reading of the Fluke 5520A (with a 34401A in 100mV range as zero-voltmeter) I also get 22uV increased voltage difference between LTZ#1 and LTZ#2. So the +3.1 ppm shift was before calibration. (either the battery flat or the transport to calibration).

But on previous transports I never had any voltage shift on my powered LTZ references.

How does the change compare to a simple power down and thus thermal hysteresis ?

Shortening a LTZ1000 ref circuit is different:

With a simple power down I never have a voltage shift that I can measure (noise limit is around 2uV with 2:1 divider).
The setpoint temperature of around 52 deg C is rather low so that there is also low hysteresis between room temperature and powered LTZ.
After the shorting of the unbuffered output (see day 1250) it is of course different. In this case I could reduce the initial hysteresis by several power downs from a initial much higher value to around -5 ppm.

Hello Andreas,

when you overheated the LTZ1000 circuit at day 1250, you probably introduced a permanent hysteresis, which is of the type that does not slowly drift back by itself. The decrease of the output after overheating is the correct direction, anyhow.
Now, as the oven failed after years, the LTZ maybe simply jumped back in direction to the initial state, i.e. before day 1250.
That's also the correct direction, because if you cool down the LTZ1000, like below room temperature, or even below 0°C, its value will always go upwards.

I also don't think that it has anything to do with the back fed current.

Hello Frank,

I think that is the correct Idea. I usually had only short power interruptions of some minutes for tests.
Now with the battery flat the power outage might have been for 2 half days giving more time to relax into the previous direction.
Should switch off the device for a longer time to see what happens.

I also think that there is no correlation with any back fed current (since it is in same direction as normal current and much lower).

At the moment I make a longer measurement (~18 hours) to check if the Allan deviation behaves unusual.

with best regards

Andreas

[floobydust]

I wonder if something happened as the battery voltage fell. It didn't just "go flat" ?
Have to try run a reference undervoltaged, where the zener is starved and off and see if the op-amp oscillates or misbehaves or latches up. From LT1013 datasheet:
"...There is one circumstance, however, under which the phase reversal protection circuitry does not function: when the other op amp on the LT1013, or one specific amplifier of the other three on the LT1014, is driven hard into negative saturation at the output."

So it shares an internal common bias, they are not totally independent op-amps.

[Andreas]

I wonder if something happened as the battery voltage fell. It didn't just "go flat" ?

In my case the LT1763 voltage regulator (set to 14V) is switched off at 13.8V by the battery monitor.
Switching on happens at 15V (hysteresis).

So it shares an internal common bias, they are not totally independent op-amps.

Didn´t know that. Interesting.
But so far I have tested switch on / off behavior by replacing the battery monitor with a frequency generator.
(the scope pictures are somewhere in this thread: no oscillations).

Edit: the scope pictures are beginning from here:
https://www.eevblog.com/forum/metrology/ultra-precision-reference-ltz1000/msg832030/#msg832030

With best regards

Andreas

[Kleinstein]

I wonder if something happened as the battery voltage fell. It didn't just "go flat" ?
Have to try run a reference undervoltaged, where the zener is starved and off and see if the op-amp oscillates or misbehaves or latches up. From LT1013 datasheet:
"...There is one circumstance, however, under which the phase reversal protection circuitry does not function: when the other op amp on the LT1013, or one specific amplifier of the other three on the LT1014, is driven hard into negative saturation at the output."

So it shares an internal common bias, they are not totally independent op-amps.

A good find, but I doubt this would be a problem here. The OPs work with one side at some 0.5 V, never higher than one diode drop. So there should be no need to even rely on the phase reversal protection.
Oscillation when the supply to the OP is too low may become an issue, but this would drive the OPs output close to the positive output. The FB loop is a little tricky, and could oscillate when the OP gets to slow or with extra delays from saturation recovery. However in this case there is not much power to drive too much zener current, so even if it starts oscillating it should not damage the reference.

[Cerebus]

From LT1013 datasheet:
"...There is one circumstance, however, under which the phase reversal protection circuitry does not function: when the other op amp on the LT1013, or one specific amplifier of the other three on the LT1014, is driven hard into negative saturation at the output."

So it shares an internal common bias, they are not totally independent op-amps.

Interesting, and it underscores something that I have learned over the years:
Read the data sheet from beginning to end, do not merely skim it for the characteristics that you are looking for. Things to  particularly watch out for are caveats that undermine the headline features (e.g. current noise on op amps with impressive 'headline' noise figures) and things related to substrate connections (e.g. a small footnote saying "never let pin x become reverse biased relative to pin y"). I've seen people come a cropper more than once from missing some vital bit of information that was under their noses if they had only bothered to look.

[Crossphased]

Hi guys,
I have two questions regarding the circuit for an LTZ1000A. I just finished putting together my first board using the LTZ. I've read lots of forum posts, and pored through the datasheet to understand it better. One thing I noticed is this voltage reference can be damaged if incorrect voltages are applied to its pins. Before I insert the LTZ into the board, are there some known voltages I can check each pin for, to make sure the surrounding circuit is operating properly?

*EDIT removed a question I was able to figure out

[dietert1]

A complete answer to your question is difficult. Maybe you can first make a LTZ1000 substitute from discrete parts plus some wires? It's a minor effort but may save your real LTZ1000s life.

Regards, Dieter

[Kleinstein]

A substitute for the voltage loop part is relatively simple (transistor + Zener)
For the heater loop, one could limit the heater voltage (e.g. at the base of the heater driver transistor), so that no excessive power is applied. This could be part of the normal circuit, not just for the initial test.

Another point is checking the circuit once more.  I known trap for the young players is the unusual pin-out of the LT1013 in SO8 case.   

[Crossphased]

Great thanks very much guys, will make a substitute LTZ from discrete components- good idea!

[alex-sh]

Quick temperature oven question.
I have the following temperature graph meaning there is about 0.6C fluctuation. I wonder if this is due to the oven not being properly insulated?

EDIT: The amplitude is 0.3C (max 35.1, 34.75 min). I have TE Connectivity GA10K3A1A 3976K NTC +/- 0.1C. The oven should stabilise to 0.1C

[Vgkid]

If that scale is in minutes, and from initial turn on. Give it time to stabilize.

[alex-sh]

Thank you. The problem is that it’s not stabilising.

[Kleinstein]

It looks like a regulator that is rather close to oscillation. Much added insulation will change the system and thus have different parameters for the controller.  It may still stabilize over a long time, but it would be better to optimize the regulator tuning.

[Andreas]

I wonder if this is due to the oven not being properly insulated?

I fear that the NTC for the temperature controller is too far from the heat source.

Thats why I typically use 2 temperature sensors.
One for the heater and one for the point of interest.

With best regards

Andreas

[alex-sh]

I wonder if this is due to the oven not being properly insulated?

I fear that the NTC for the temperature controller is too far from the heat source.

Thats why I typically use 2 temperature sensors.
One for the heater and one for the point of interest.

With best regards

Andreas

Hi Andreas,

Interesting point. I have a metal enclosure and two NTC sensors (I specified the type two posts above) - One sensor is attached to the centre of the enclosure (metal part) and another sensor is also in the middle measuring the air. Is this a wrong approach?

Thanks
Alex

[alex-sh]

Much added insulation will change the system and thus have different parameters for the controller. 

Yeah. I have lost  insulation material and now have to order a new one. Any recommendation on insulation please?
I think insulation is a must. Otherwise  the metal enclosure is losing heat very quickly and it takes time to heat up the oven (four 10W resistors). Hence 0.3C oscillation

[floobydust]

The period looks like 160 seconds or 2.7 minutes?
I've seen temperature graphs with that long of a time-constant due to the lab HVAC cycling. Make sure there is no draft or airflow from a vent influencing your test. You can put a piece of cardboard over it to see if that makes a difference.

[alex-sh]

The period looks like 160 seconds or 2.7 minutes?
I've seen temperature graphs with that long of a time-constant due to the lab HVAC cycling. Make sure there is no draft or airflow from a vent influencing your test. You can put a piece of cardboard over it to see if that makes a difference.

There is no aircon. However, you are correct there is a temp change / draft. It comes back to a proper insulation