Ultra Precision Reference LTZ1000

[Kleinstein]

Is there an explanation for the change in voltage due to the shorter leads  (130 µV noted down below) ? 

I am surprised to see so much change, as normally the slight higher thermal current should not change much. There is thermal EMF at the leads, but this should compensate to a very good approximation. The design must be terribly off if differences as large as 3 K exist.

It might be interesting to compare to the change when changing the thermal shield - this can also have an influence on the heater power.

[Andreas]

Is there an explanation for the change in voltage due to the shorter leads  (130 µV noted down below) ? 

The design must be terribly off if differences as large as 3 K exist.

130uV are around 18 ppm so less than 0.4K for the zener  (50-55 ppm/K without thermal regulation).
So how do you calculate 3K difference?

With best regards

Andreas

[Kleinstein]

The 3 K change would be from the about 40 µV/K thermal EMF. So this estimate only shows that this is not the source.

If you calculate back to the zener temperature, this would be 0.4 K - but this would be a huge change against a temperature regulator.  As the feedback loop itself has rather high gain and ideally (e.g perfect position of sensor/heater) should get the temperature stable to mK or better. So I really doubt the hole chip changes by 0.4 K - this would suggest a surprisingly poor thermal design. The more likely part would be having a change in the difference for the zener and compensating transistor. So more like a change in temperature difference by 0.06 K. Still this does not sound like a good thermal design for the non A LTZ1000.  In this case a different thermal isolation cap could also have quite some effect.

With so much sensitivity to the thermal design / heat flow one might really consider using a second layer of thermal stabilization.

[Galaxyrise]

The 3 K change would be from the about 40 µV/K thermal EMF. So this estimate only shows that this is not the source.

If you calculate back to the zener temperature, this would be 0.4 K - but this would be a huge change against a temperature regulator.  As the feedback loop itself has rather high gain and ideally (e.g perfect position of sensor/heater) should get the temperature stable to mK or better. So I really doubt the hole chip changes by 0.4 K - this would suggest a surprisingly poor thermal design. The more likely part would be having a change in the difference for the zener and compensating transistor. So more like a change in temperature difference by 0.06 K. Still this does not sound like a good thermal design for the non A LTZ1000.  In this case a different thermal isolation cap could also have quite some effect.

With so much sensitivity to the thermal design / heat flow one might really consider using a second layer of thermal stabilization.

Or perhaps that the old (long-leg) design was a poor choice for the non-A LTZ1000.  The datasheet does explicitly call for short legs, after all.

[Kleinstein]

I see that the data for the short legs look better (lower TC) - but this still means the point with a higher heater power and thus more internal temperature gradients. 

This makes me think, if if might be even better not to use the internal heater, but only the internal sensor and an external heater instead. As the heat flow would be only due to the rather constant internal loss, there should be much better temperature stabilization. I know an external heater would need more power and is slower, but who cares in a 24/7 line powered application.

[Andreas]

Or perhaps that the old (long-leg) design was a poor choice for the non-A LTZ1000.  The datasheet does explicitly call for short legs, after all.

Hello,

where have you found this?
Up to now I have only found that "thermal layout"  is essential to get full performance.

And in my opinion you get lesser thermal EMFs (more equal temperatures on solder joints) when leaving the leads long.

With best regards

Andreas

[Andreas]

I see that the data for the short legs look better (lower TC)

This is only true for my sample LTZ#4.
On LTZ#5  I have "zero T.C." with long legs.
So I guess I would get a +0.15 ppm/K T.C. if I shorten the legs on LTZ#5.

with best regards

Andreas

[Galaxyrise]

Or perhaps that the old (long-leg) design was a poor choice for the non-A LTZ1000.  The datasheet does explicitly call for short legs, after all.

where have you found this?

Whoops, I was looking at the wrong datasheet!  I didn't have enough of the filename showing, sorry. 

[Andreas]

Mhm,

would be anyway interesting which device calls for short legs.

By the way: I did some error in comparison of the power consumption.
The 25.2 mA are for LTC2057 instead of LT1013A.
After changeing back to LT1013A I have only 24.0 mA.
So only a increasement of 6%.

with best regards

Andreas

[branadic]

would be anyway interesting which device calls for short legs.

Well, LM199 calls for short legs, refering to AN-161 and similar thermal mass at each pin:

"...Thermocouple effects can also use errors. The kovar leads from the LM199 package from a thermocouple
with copper printed circuit board traces. Since the package of the 199 is heated, there is a heat flow along
the leads of the LM199 package. If the leads terminate into unequal sizes of copper on the p.c. board
greater heat will be absorbed by the larger copper trace and a temperature difference will develop. A
temperature difference of 1°C between the two leads of the reference will generate about 30 ?A.
Therefore, the copper traces to the zener should be equal in size. This will generally keep the errors due
to thermocouple effects under about 15 ?V.
The LM199 should be mounted flush on the p.c. board with a minimum of space between the thermal
shield and the boards. This minimizes air flow across the kovar leads on the board surface, which also
can cause thermocouple voltages. Air currents across the leads usually appear as ultra-low frequency
noise of about 10 ?V to 20 ?V amplitude..."

[Kleinstein]

There are several differences between long and short leads:

As advantage of short leads and thus not so good thermal insulation there is less temperature rise due to the internal power - so the set temperature might be chosen slightly lower (e.g. 2-5 K). Also thermal regulation might be slightly faster and thus could allow a faster regulation loop - but that is likely not an issue anyway. At least usual the thermal regulation is not build for ultimate speed.

Thermal fluctuations under the chip can also be avoided with a spacer or padding - so this is a rather weak argument for short leads.

With short leads one has a higher heat-flow through the leads and thus more sensitivity to unbalanced thermal layout. Also more power is needed to stabilize the temperature so chip internal temperature differences are also expected to be larger. This last point might be rather critical - at least for me it's the best explanation for the change in voltage just from changing the thermal setup. Often extra insulation is used to keep the thermal resistance high - so I see no good reason not to do that with leads too.

It might be interesting to check how much influence the thermal insulation has - it is well possible to have the TC separated into an effect of heat flow through the leads and heat flow through the cap part. I would also expect the TC is more an effect of the heat flows and less of the direct TC, as at least the sensor part of the chip should have a very stable temperature. With an effect due to heat flow it somewhat makes sense to have the 400 K compensation resistor from the heater voltage, though the heater power and thus the heater voltage squared should be more suitable.

[TiN]

I'm under impression that long leads cause large thermal gradient from zener to copper tracks on PCB, causing uneven EMF. No matter how balanced is layout, it would not be exactly perfect anyway. By having chip close to PCB (mostly it's bottom sitting right on PCB surface!) PCB get warmer and there is less temperature gradient.

[IanJ]

Hi all,

Just a wee note on the LTC2057..............I tried it on my precision digital voltage source on the final output after the DAC a few months ago but found it had a rather annoying glitch that I couldn't get rid of. Don't remember the exact nature of the glitch apart from it was just a few tens of uV.
I did try a few (same batch) to no avail, but changing to a different device the problem went away immediately.

Ian.

[Galaxyrise]

would be anyway interesting which device calls for short legs.

I remembered the one from the LM399, but looked at the LT1007 datasheet when I checked if the LTZ1000 had a similar entry.  But the LT1007 doesn't have much self-heating to worry about, and the LM399 doesn't usually have its leads under a wind screen.

[Andreas]

Hello Ian,

sorry but the info from your side is a bit vague for me.

How did you measure the tens of uV change? (Oscilloscope or multimeter)
Which bandwith/NPLC settings did you use?
Wiring capacitance of your setup? (long coax-line?)
What was the other part number with no issues? (bipolar or CMOS?)
Is your output stage a voltage follower or a inverting cirquit?

I also had some problems with some uV drift measured on DMM for the LTC2057 in voltage follower cirquit.
I blame it on the CMOS protection diodes at the negative input.
But as Ken already described this can be solved with a output filter capacitor and a capacitive isolation cirquit.
It is generally a good idea to have some kind of isolation cirquit on precision outputs.

With best regards

Andreas

[IanJ]

Hello Ian,

sorry but the info from your side is a bit vague for me.

How did you measure the tens of uV change? (Oscilloscope or multimeter)
Which bandwith/NPLC settings did you use?
Wiring capacitance of your setup? (long coax-line?)
What was the other part number with no issues? (bipolar or CMOS?)
Is your output stage a voltage follower or a inverting cirquit?

I also had some problems with some uV drift measured on DMM for the LTC2057 in voltage follower cirquit.
I blame it on the CMOS protection diodes at the negative input.
But as Ken already described this can be solved with a output filter capacitor and a capacitive isolation cirquit.
It is generally a good idea to have some kind of isolation cirquit on precision outputs.

With best regards

Andreas

Hi,

Yes a bit vague because it was a few months ago and I don't remember much about the exact nature of the noise. I do remember though that it wasn't at the chopper switching freq, it was a lot less.....I think at the time I thought it was picking up noise from something else on the pcb (lcd, Atmel uC etc)......and that I was a bit surprised to see it given I had tested quite a few other chopper op-amps and none of the rest exhibited the same issue......at all!

To answer your questions:-
- Not drift, but glitching. I have never had a problem with drift using any of the choppers I tested.
- Oscilloscope (Rigol 2202) and using a ground clip on the standard Rigol scope lead.
- The 2202 is a 200MHz scope, I tried with the 20MHz limit in & out (as I usually do).
- Output not connected to any DMM or any other leads used when testing noise with the scope.
- Voltage follower x2 gain (5ppm/degC resistors)
- My circuit has a similar setup by way of series resistor to isolate, and a snubber etc.
- The feedback cap is not in my circuit as standard, but only because it made no difference. My circuit is quite noise free (well, as much as the noise floor of the Rigol will allow). I remember trying the feedback cap on the LTC2057 but it made no difference.

I have tried a couple of other precision choppers.

LTC2050 - Limited to 11.5vdc supply max, nice op-amp, but output relatively easily damaged. No current limit or short circuit protected.

LTC1250 - Advertised as a bridge amplifier, but works great in all respects. The best all round solution for me, albeit the most expensive op-amp. Robust output.

I tried a bunch of others.......don't have the part numbers to hand.

Hope that helps.

Ian.

[Kleinstein]

Chopper OPs can be a little more sensitive to EMI problems and poor decoupling. Also high impedance sources can be a problem and may require an extra capacitance to ground. The input bias current is different from other FET OPs: as charge injection current the Bias is about opposite sign for both inputs - so no compensation  trough equal source impedance. Also the temperature dependence is not as strong as with other FET OPs. So the 30 pA for the LTC2057 are also valid at 50°C not only at 20 C.

For the LTZ1000 Reference I see not real need for a chopper OP, as reference chip has gain inside. The buffer Chip right after the reference is way more critical.

For the thermal design, the thermal resistance of a board without strong thermal cutouts will be somewhat comparable to the thermal resistance of the chip and pins - so the temperature of the solder joints will not follow the chip temperature, but only about half the way - maybe a little more with cutouts and less with a solid board with much copper. What really helps is that only the difference between the solder joints is what counts, and the coupling between the joints is usually way better than to the outside. Making the board thermal resistance really large compared to the pins is not that easy - one would need rather thin copper lines (e.g. 0.1-0.2 mm over 50 mm length). So I would still see an slight advantage for the longer pins - though I don't think the thermal EMF is really a problem in both cases.

[zlymex]

Another consideration of the opamp is the tempco of the bias current, better to be small, to allow the divider pair to have larger values other than 1k:13k, for instance 2k:26k, 3k:39k or any value that not too large and having the same ratio. This is very helpful when buying off-shell parts where the value selection is limited.

Looking at the datasheet of LT1013, the temco of the bias current is about 30pA(per deg C), less than 0.1ppm of the current of the divider pair(around 500uA). We all know that the divider pair has about 1:100 ratio on final voltage, so this 30pA variation will affect the final voltage by 0.001ppm(=0.1ppm/100).
Therefore, if you don't mind "increasing" the tempco by 0.001ppm/K, you can use an 2k:26k pair.

While the voltage noise of a LT1013 looks a bit high(a little below half of the typical LTZ1k), the current noise of the bias is small enough to be ignore safely.
Apart from choppers, there are other options to replace LT1013, such as OPA2188 and OP727.

[Kleinstein]

Due to the internal amplification, the OP used directly at the LTZ1000 is not that critical. So the LT1013 is fine and
there is little reason to use an auto zero OP, as the low drift, low LF noise and high DC gain are not needed.
More important parameters are power consumption and stability in the loop with extra gain  - some AZ OPs could have a problem here and may need adaption of the circuit.

The higher frequency noise is usually set by the OP, due to the filter function of the circuit, so here the noise of the OP might be important.

There is one more effect of longer leads - the long leads could have a resistance up to the 0.1 Ohms range and this way have an significant influence at pin4 with the typical 120 Ohms resistor. As the resistance of kovar is not that stable with temperature it can also contribute to the TC.  However resistance of the wires also sets a limit on how good thermal isolation on the board can be made.

[xyrtek]

Anyone got working LTZ1000 board PCBs to sell?

For anyone interested in the pcb design please check this.
https://xdevs.com/article/kx-ref/

My apologies for cluttering the thread.

[Andreas]

There is one more effect of longer leads - the long leads could have a resistance up to the 0.1 Ohms range and this way have an significant influence at pin4 with the typical 120 Ohms resistor. As the resistance of kovar is not that stable with temperature it can also contribute to the TC.  However resistance of the wires also sets a limit on how good thermal isolation on the board can be made.

When calculating with the data:
http://www.calfinewire.com/datasheets/100105-kovar.html

I get about 3.1 Ohms/m for the 0.45 mm diameter of the wire.
with 12.7mm (shortable) wire length this gives about 39mOhms for Pin 1 where I have influence.
Over 20 degs temperature difference the 0.35%/K give a factor 1.07 or 2.7 mOhm difference.
so against the 120 ohms around 23 ppm resistor change.
With the factor 0.14/100ppm this gives a error 0.032 ppm over a 20 deg C range.
Or 0.002ppm/K.

I guess I cannot measure that.
So the thermal isolation has much more effect.

With best regards

Andreas

[Kleinstein]

The wire resistance has an effect mainly with pins 3 and 4, as here the highest current flows.
With pin 3 and 4 mA zener current, a 2.7 mOhms change in resistance for a 20 K increase in temperature gives 10-12 µV of extra voltage. However the Kovar wire will also partially profit from temperature stabilization, so a smaller effect of maybe half the size is more realistic. So about an additional +0.04ppm/K for the TC with longer leads. That is quite a lot for chip that should reach 0.05 ppm/K.

The other effect would be from Pin 4 - from Andreas' calculation this effect is much smaller. So even with opposite direction it can not fully compensate.

So the length of the wires is not only important for the thermal part.

Besides the extra TC the lead resistance can also explain much of the change in absolute value.

[Andreas]

The wire resistance has an effect mainly with pins 3 and 4, as here the highest current flows.
With pin 3 and 4 mA zener current, a 2.7 mOhms change in resistance for a 20 K increase in temperature gives 10-12 µV of extra voltage. However the Kovar wire will

At least for pin 4 the behaviour is not linear.
A 100 ppm change on R1 gives 0.14 ppm change of the zener voltage.
So for Pin 4 its only 0.22 uV (and not 5-6) over 20 deg C.

With best regards

Andreas

[Andreas]

Hello,

Measurement on LTZ#4 with short leads. LT1013A again as OP-Amp.
And R9 = 402K to compensate the remaining -0.1ppm/K.

For R9 = 1 Meg I get a T.C. increasement of around +0.04ppm/K
100K give around +0.4ppm/K over 20 deg C.

While the K2000 had some warm-up delay for the first hour,
both the HP34401A in 100mV range against LTZ#5 and ADC16 (with 2:1 divider)
show a nearly zero T.C. (less than 2uV change) between 20 and 32 deg C environment temperature range.

With best regards

Andreas

[Kleinstein]

The extra resistance of the longer leads at pin 3 can explain rather well the drop in voltage from shortening the leads. However one would extect an more negative TC with the shorter leads, but the observed change is in the other direction  - so the change in TC must have another explanation. The wires at pin 4 should have even less influence, so they cant be the cause either.

The with the compensation with R9, the TC gets smaller, but the curve also get quite nonlinear. Here I don't think the effect is due to nonlinear influence at pin4, but due to the nonlinear relation ship with heater voltage. So the heater voltage might not be such a good choice for compensation. As the relationship heat voltage to power gets more linear at higher power / higher set temperature, the curvature could well be less critical at higher temperature setting.