Module with temperature-stabilised LT1236-5LS8

[DeltaSigmaD]

Voltage Reference with a temperature-stabilised LT1236-5LS8

References with LTZ1000 or LM399 were discussed in many details in this metrology board. Both references have one weak point: the hermetic TO-package cannot be produced without Kovar or similar alloys for the package feed-throughs, which generate high thermal voltages against the Cu of the PCB. Therefore, the temperature gradients at the LTZ1000 pins must be stable to <0.1 K, while the chip is heated by typically more than 25 K. This is a demanding task, and the size of supporting electronics is large.

For my application, a smaller and less expensive reference circuit is required. But of course, the stability should be as good as possible. I was curious to know what is possible with a LT1236 5V reference IC. The drift of this reference is typically 2 or 5 ppm/K, and a temperature stabilisation is required for better stability. A temperature-stabilised reference module on an aluminium board (PCB) was developped and tested. The temperature dependency is reduced to less than 0.2 ppm/K by heating. Please read the attached blog if you are interested on details.

[ramon]

Hi, did you ordered the alumina pcb to some PCB manufacturer? or did you bought the substrate and later made the traces by yourself?

I have aluminum nitride bare ceramic substrates (82% thermal conductivity of pure Al) and would be interesting if I find the proper way to put copper traces over it.

[iMo]

Any schematics?

[razvan784]

I like your discussion of the various coefficients of thermal expansion.
I'd like to point out this appnote http://www.vishay.com/docs/28872/mechanicalstress.pdf where the authors also discuss the influence of the CTE of the solder joints.
They point out that even when mounting alumina components on an alumina substrate, the solder with its different CTE will still cause significant stress.

[branadic]

Reminds me of an idea I had to fit on a QH40A crystal heater, but was thinking of polyimide directly glued to the ceramic of the heater and without the aluminium in between.

-branadic-

[DeltaSigmaD]

The circuit must be as simple as possible in order to reduce its thermal mass. R1 and R2 are in series to simplify adjustment. But in most cases no adjustment will be necessary, since the semiconductor process tolerances are very tight.

[DeltaSigmaD]

Thank you for the link, I didn't know that paper.

The important role of isolation layers is no surprise. There seems to be no ideal solution yet. The leaded solder alloy in combination with temperature stabilisation is the best solution until the perfect one is found ...

[EmmanuelFaure]

An idea : You could control the temperature of the reference by heating directly the chip, i.e. not through a PCB. You can use a thin film PT100 sensor in a smd package, glued to the reference package, and use it both as a temperature sensor and a heating element.
With this technique a dead bug mounting technique can be used, and you're free of any mechanical induced drift.

[DeltaSigmaD]

To EmmanuelFaure:

Interesting idea to mount the reference IC top down (dead bug) and to use Cu wires to connect it. If only the ref IC is heated, there is a varying temperature gradient along the wires and an according thermal flow over the reference pins. The question is whether the thermal voltages at the transitions Cu - pins(W,Mo) - Au - Si are small enough. It is worth a try.

With the HLT1236M module, the temperature rise of the reference chip due to self-heating should be very constant, providing there is no significant load at the reference output. With dead bug mounting, the mechanical stress could be minimised, but the temperature sensing on the chip should be used anyway. But, the time constant is increased with the thermal resistance board to chip, which point is critical.

If you don't use on-chip temperature sensing, you could use the LT1027, which has lower noise.

[iMo]

That temperature stabilization would work best if the temperature was set to the point where dTC/dT of the 1236 is equal 0. Did you measure TC_1236 vs. temperature (most probably 2 points with dTC/dT=0)?

EDIT: fixed - TC has to be "dTC/dT"

[jbb]

Off topic, but does anyone know if Analog will release ~7V or 10V references in LS8?

[magic]

That temperature stabilization would work best if the temperature was set to the point where TC of the 1236 is equal 0.

Or tempco could be adjusted to almost zero using the DNC pins

I actually bought some LT1021 to experiment with but still have no time for it

[EmmanuelFaure]

To DeltaSigmaD :
* Using thin copper wires, the heat flux is negligible.
* The time constant in a 3D thermal system varies with 1/L², with L = a characteristic dimension. With a suitable control system you should easily achieve a time constant of some seconds at worst. With my idea of an external pt100 sensor/heater, you coud use a cascaded system to increase the performance even more. The PT100 with a fast response with a strong P and weak I, and the temperature measured through the Trim pin with a slow but stable response with a weak P and strong I.

[iMo]

That temperature stabilization would work best if the temperature was set to the point where TC of the 1236 is equal 0.

Or tempco could be adjusted to almost zero using the DNC pins

I actually bought some LT1021 to experiment with but still have no time for it

Let me precise what I wrote above:
The 1236 would be placed into an "optimal" temperature stabilization range if the oven temperature was set to the point [T1 or T2] where dTC/dT of the 1236 is equal 0 (as, afaik, all LT102x and LT12xx show "S" shaped TC on T dependency).. 

[Conrad Hoffman]

A useful trick, though not sure it's useful here, is to connect using fine gage manganin wire. It has near zero TC with copper and you can solder it. It also has lousy thermal conductivity, which is great when connecting to heated or cooled circuits if you want to avoid outside influence. The only downside is resistance, but that's usually not an issue as it will stay fairly constant.

[Andreas]

That temperature stabilization would work best if the temperature was set to the point where TC of the 1236 is equal 0.

Or tempco could be adjusted to almost zero using the DNC pins

I actually bought some LT1021 to experiment with but still have no time for it

Let me precise what I wrote above:
The 1236 would be placed into an "optimal" temperature stabilization range if the oven temperature was set to the point [T1 or T2] where dTC/dT of the 1236 is equal 0 (as, afaik, all LT102x and LT12xx show "S" shaped TC on T dependency).. 

No chance with the LT1236:
Nearly all LT1236AILS8-5 that I have measured have a nearly linear tempco between +2-4 ppm/K (typical +3ppm/K)

For a S-shaped curve near room temperature you will need either a LT1027 or a AD586

with best regards

Andreas

[iMo]

..Nearly all LT1236AILS8-5 that I have measured have a nearly linear tempco between +2-4 ppm/K (typical +3ppm/K)
For a S-shaped curve near room temperature you will need either a LT1027 or a AD586

In what temperature range did you measure? Could it be the T1 and T2 were outside your range?
PS: ideally the T2 would be at 40-50C..

[Andreas]

Hello,

One example measurement: Usually I try to measure from 10-40 deg C.
Depending on environment temperature I reach also higher values.

The bad thing that I observed with all LS8-packages is that it has a relative large hysteresis. (>> 1ppm)

Note that I measure indirectly with the LT1236 on the reference input of a LTC2400. The A/D-input is a stable 7V reference with a precision 2:1 divider (LTC1043). So if you see a negative T.C. in the diagram the reference has a positive T.C.

with best regards

Andreas

[DeltaSigmaD]

To EmmanuelFaure:
Of course, the time constant could be drastcally reduced, and suitable adaption of the controller parameters will lead to good settling. This could be made even with the existing circuit. But a short time constant is combined with a higher heat flow with accordingly higher gradients. A top-down mounting will be necessary in this case. A problem will be the dead time between heater action and chip response as with many temperature controllers.

[DeltaSigmaD]

To Andreas:
All LT1236 references I measured had a roughly linear temperature coefficient. It has obviously no curvature correction such as LT027.

To the upper ambient temperature limit of the reference: this limit is determined by the setpoint temp which could be set also to much higher temperatures. Already tested setpoints where 31°C, 38°C, and 43°C, and about 6 K are required for regulation (temperature rise by quiescent current). The on-off-on hysteresis test for 43°C has still to be made.

And I agree that the LS8 package shows a relatively high hysteresis, therefore temperature stabilisation. My idea was to have a temp setpoint low enough to obtain a small hysteresis with power failure. Do you know anything about the functional dependency of the hysteresis on the temperature range of change?

[Andreas]

Hello,

from theory the hysteresis goes with square from temperature change.

But I did not measure that with references. From precision resistors I have the impression that reducing the temperature swing below +/-7 deg C gives nearly no hysteresis (at least I cannot measure it because it is within my measurement noise).

So the stragegy of a heated reference short above the maximum operating temperature is the way to go.

With best regards

Andreas