Ultra Low Noise Reference 2DW232, 2DW233, 2DW23x

[branadic]

I completed my design for a 10V reference with the first 2DW233 sample. As the crystal heater is mounted to an aluminium rod with a diameter of 16mm and the reference diode with a diameter of 8.2mm is stuck into it, there is still enough material around the diode to insert all resistors necessary into it as well. I will use PTF56 for that purpose. So I need a single sided board on each side to connect everything. A LT1001 in smd package will thus be temperature stabilized too. This is some sort of 3D mounting concept and thus a pretty compact design.

-branadic-

[GigaJoe]

some diodes drifting up to 20mV / day.   so I hook all diodes with 1K resistor  (approx 2ma per single internal diode) I put everything in a box tossed it under bed.  will wait a month till it settle ...  then measure\see\figureout ... guess, diodes recovering after harsh forces ... 

( i notices with LM399 , same picture , after 6 month under high temp, they drifted a lot ... then stopped and became rock solid ... will see )

[branadic]

As announced I've build the ovenized 2DW233 reference. It took a while for the boards to arrive at my place. However, here is how the ovenized reference looks like now. I first powered it up and it shows a 9.97V output by now, which is almost what I expected and planed to use a precision multiturn pot to set it exactly to 10V anyway. The ovenized reference is still small enough to fit into the flask shown a few posts before together with some foam. So it's now time to perform some more measurements on this particular device.

-branadic-

EDIT: The complete ovenized reference inside the flask with some additional thermal insulation draws about 19mA @ 12V. A LT1763 based voltage regulator is sitting inside the flask too.

[Andreas]

Hmm,

still no real "zero T.C." results? or did I overlook something?

I went a different way:
I bought cheap SMD (SOT-23) double zeners (ESD-protection diodes AZ23C5V6)
https://de.rs-online.com/web/p/zener-dioden/7081857/?sra=pstk

Ok they are not really low noise (around 2uVpp at 7.5 mA) but two T.C. measurements with my standard voltage regulators of 9.35V and 14.0V and a simple 1K 1% series resistor with both diodes in series look promising that the zero T.C. will be somewhere between 3 and 7.5 mA.

At 9.35V I got: -20.2 ppm/K in average
At 14.0V I got: +20.1 ppm/K average T.C.

Of course that does not tell anything about long term stability etc.
But +/- 20 ppm/K on first try is rather good compared to some other cheap references.

with best regards

Andreas

Update: The best result I got:
at 10.6V: +0.23 ppm/K average T.C. (green curve)
but the diagram shows that the temperature range for "zero T.C." is very limited
some degress left or right from the optimum it rises quickly above 1 deg/K

[branadic]

still no real "zero T.C." results? or did I overlook something?

Nope, I couldn't perform this measurements yet as I was of the bench because of illness. But it's on my to do list. Back on the bench since monday I finallized my 9x LTZ1000 references and performed first measurements.

-branadic-

[janaf]

Thanks for sharing. Even if these are not time/temperature reference stable, they could be used in very low noise power supplies...

[branadic]

Even if these are not time/temperature reference stable

Well, I'm not sure about that. I currently have one 2DW233 with an oven running and it looks like it starts to settle down, but to be sure about that it needs a bit of time to get a statement on that.

Re: vintage ltz1000 from 1986 NIB !.

-branadic-

[janaf]

A batch of 2DW233 from taobao just arrived.

https://item.taobao.com/item.htm?spm=a1z09.2.0.0.gpf2Zk&id=35815633601

They have the 'right' symbol, the factory symbol is also on the bags.
Mine have house marks/bin color dots? on them, ie may have been subject to selection.

Some stats on 25pcs at 7.5mA (per diode) fresh out of the bags:

- Average voltage 5.65V, +/-0.05V
- The pairs all matched within +/-12mV with a standard deviation 8.3mV.
- For roughly one in 5 device pairs, the two diodes where within 2mV from each other.

EDIT: Updated stats for 'back-to-back' zener+diode:
- Average voltage 6.414, 15 devices within +/-60mV

 

[Kleinstein]

25 Ohms is quite a high resistance.
The real test would be if the noise is low - the "wrong" diodes tend to have a much higher noise.
If low noise the other question is if a low TC can be obtained at a reasonable current. There seem to be some batches with low noise, but too high a current for low TC. 

[janaf]

The 25 ohm was a single quick and dirty measurement... will do some better ones.

Will do noise measurements soon.
The noise presented by the OP, was that per single diode or a pair?

[janaf]

Did some slightly better measurements of the impedance. It's a bit hard as there's a lot of temperature drift  .....

In the end, I let each point stabilize for a couple of minutes. Then, the impedance was around 6.5-ish ohms at 7- mA.

[Kleinstein]

The relevant noise if for a series connection of the diodes, so one as a zener and the other as a forward diode. Ideally this would be at a current where the TC is low.  AFAIR the current for a low TC can vary quite a lot - the 7.5 mA are more like an ideal value, but others have found some 5 mA to > 30 mA (thus no practical zero TC).
To get a really good noise test it would need a suitable current for low TC first so that temperature fluctuations would not give so much contribution.

The difference between a good low noise zener and a different more normal zener is huge - like a factor of 10 or 100. So a first test could use a fixed e.g. 7.5 mA current.

For the differential resistance one could use an AC current super imposed, so that the temperature would not change that fast.

[janaf]

Here's the local impedance for a fast sweep, marginal self heating, 1mS per data point.

EDIT New graph with back-to-back diode-zener connection.

[janaf]

Ah, yes of course, back to back would give temp compensation. I wasn't thinking much of that as I'll use mine for quiet power supplies, not primarily as a reference. 

[IconicPCB]

Consider using one an d a half diodes i series.

By that i mean two zeners plus a normal diode.
The net effect isdoubling of reference voltage with lower temperature zero tc point.

[Kleinstein]

Consider using one an d a half diodes i series.

By that i mean two zeners plus a normal diode.
The net effect isdoubling of reference voltage with lower temperature zero tc point.

2 Zener +  1 Diode in forward direction is an option, if one can ensure good thermal contact.  The temperature where you get zero TC depends on the current, and it does not need much change in current to shift the temperature. So it's more about lowering the current, than getting a lower temperature. It could be especially an option for those diodes that would need more than 25 mA to get zero TC, when used as 1 zener +  1 forward diode.

[IconicPCB]

Yes.. depends on Your perspective.

I was looking to get  the zero temp coefficient point at room temperature and could not reach it within reasonable ( op amp output ) current level.

It struck me that I could achieve zero tempco setting by combining the diodes in above fashion .

[branadic]

So I have my ovenized 2DW233 running for a while now, with the resistors being inside the oven (see posts before). The reference is still decreasing in output voltage but shows indication of getting stable in average output voltage soon.
However, the short term readings themself are quite unstable. I found, that low frequency noise (0.1 - 10Hz) has increased to about 10µVpp. I'm not sure if this is due to the reference itself or the fact that the resistors (PTF56) are ovensized at a temperature of about 40°C producing excessive noise.
Anyone out there with experience in PTF56 resistor noise at higher temperature or long term noise behavior of 2DW233?

-branadic-

[Edwin G. Pettis]

As far as the resistors go, any significant increase in noise over and above the normal thermal noise would be caused by faulty weld joints.  In the case of the diodes, most likely source of increasing noise above the expected thermal limit would be caused by additional contamination of the semiconductor, another quality control issue.

[Kleinstein]

The DW232 circuit may be more sensitive to resistor effects than the LTZ1000 circuit. However the resistors should not contribute much to the noise, at least not the direct way.  I see more a chance that the resistors may be part of the voltage drift, though it would need quite some change to explain the observed effect.

There may be an indirect effect on noise: the circuit should ideally be initially adjusted to get a low TC of the reference at the operating point and set temperature. This would keep the effect of temperature variations low. However if things drift chances are the TC could also change and the reference can become more sensitive to temperature variations.

Of cause one can never exclude a possibly really bad resistor, e.g. with a poor contact of the end cap.

[Edwin G. Pettis]

To be clear, a poor weld joint particularly with the sandwich type of joint can indeed produce microvolts of noise and offset, PTF resistors utilize sandwich welds.  Further more, there are no end caps on precision resistors they cause way too high TCR, those are almost always relegated to power resistors.  With a sandwich weld, it can initially appear good but deteriorate with time and heat, while 40°C isn't particularly hot it is certainly hotter than room temperature.  I have seem many a sandwich weld joint pass all kinds of testing including high temperature stress and still fail later on, on the other hand there are sandwich joints which can pass all testing and remain good for the rest of the component's life.

It was said earlier that the circuitry was in an 40°C oven so unless his oven is fluctuating, temperature is not the primary suspect, although he did not mention how long he had been running the circuit in the oven.  As I mentioned the diode's quality could also be suspect for unexpected noise.

[Gyro]

I still can't work out why they use such thick and incredibly stiff steel wires on these devices... both from the perspective of thermal emfs and cracking of the glass seals, visible in janaf's second photo (#602).

[SilverSolder]

The steel isn't plain soft iron or anything, it is selected to match the thermal coefficient of glass (to prevent it cracking due to temperature swings).

[babysitter]

The steel isn't plain soft iron or anything, it is selected to match the thermal coefficient of glass (to prevent it cracking due to temperature swings).

Its possibly an alloy named KOVAR or a cousin of it, whose temperature coefficient of expansion matches that of (borosilicate) glass. Also there is invar, which has close to zero tce.

[Gyro]

Hmmm, maybe it is Kovar, it is highly magnetic / ferrous. The trouble is that it is so thick and stiff compared to the seals that, at best, you get surface chipping of the glass, as seen in the photo. The leads are certainly much thicker / stiffer than anything I've ever come across on a TO-5 can. I can see no reason for it.