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

[pigrew]

I ordered a set of 2DW232 from AliExpress, and they were delivered last week (no time for measurements, yet).

They ended up being "15-4" date code with a painted red dot on each. Inside, they have the smaller die (barely larger than two bond pads), covered in the white RTV-silicone-like substance.

The wire-bonds are not gold. I left a 2DW232 in salt water for the day, and the wirebond is still intact. I'll assume they are silver (which has a reasonably low thermal voltage with respect to copper and gold).

Why is the silicone used? One total guess is that it is to block the emitted light (mentioned earlier in this thread, I've not looked for it) in order to reduce the coupling between diodes?

Also, an interposer layer would be a good idea in order to reduce mechanical stress, though this would likely require a third bond pad and a more complicated silicon process to maybe do an extra implant for a top-side contact.

The leads are magnetic, and don't seem to easily rust, so I'll assume they are kovar.

[Cerebus]

The wire-bonds are not gold. I left a 2DW232 in salt water for the day, and the wirebond is still intact. I'll assume they are silver (which has a reasonably low thermal voltage with respect to copper and gold).

An odd assumption when aluminium bond wires are common and aluminium doesn't corrode in the presence of chloride ions (or all my aluminium cooking pans would have some big holes in the bottom).

[branadic]

I managed to put some parts together (2DW232 with both diodes paralleled, PTF56 resistors and LT1001) on a protoboard and to mount a selfmade aluminum heat sink (a rod with a flat and a crystal heater glued to the flat) to the zener diode. I didn't meet the 10V yet and the circuit needs some trimming, but it's ready to go for first measurements on the 3458A.

-branadic-

[Kleinstein]

The noise could very well be mainly the noise of the meter (e.g. LM399 reference). If the reference is as good as hoped for (the best reports so far)  the noise level of just the reference should be much lower.  To really get an idea of the noise it would take something like the difference of 2 zener diodes.

If you want to make a stable 10 V reference from 2DW232 is would be a good idea to start with 2 of the zeners in series (could also be the old style bridge circuit) to give some 10.6 V. The divider down to 10 V is than less sensitive to resistor drift than an amplifier starting from 5.3 V or 7 V.

[branadic]

The noise could very well be mainly the noise of the meter (e.g. LM399 reference).

That is exactly the case, that's why I measured with two different integration times to see this effect. I currently build a second board with exactly the same parts, still on protoboard level.

-branadic-

[branadic]

During the last days I measured the "10V"-output of the 2DW232 reference with our 3458A. Seems like I do have a negative T.C. problem. However, the measurement represents the extrem small noise of the reference. Hence, there is a lot of potential inside this small little puppy, so time to dig further into making everything a bit more stable.

-branadic-

[Kleinstein]

The TC of the reference depends on the zener current, with a more positive TC for a higher current. For the 10 V reference there is also the TC of the resistors to set the gain (or divider if 2 refs in series are used).  For those temperature compensated zeners (e.g. old time standard 1N829) it was common practice to adjust the TC close to zero by choosing the individually suitable current.

[Magnificent Bastard]

During the last days I measured the "10V"-ouput of the 2DW232 reference with our 3458A. Seems like I do have a negative T.C. problem. However, the measurement represents the extrem small noise of the reference. Hence, there is a lot of potential inside this small little puppy, so time to dig further into making everything a bit more stable.

-branadic-

Can you say what the horizontal units are?  (hours, or minutes, or..?)  It looks like that there is nearly one-division of time lag between a temperature change and the resulting voltage change...

[branadic]

The horizontal time scale is 342.290 * 4s / 3600 / 24 = 15,85 days.

-branadic-

[branadic]

After some analysis of the data I found a lag between voltage and temperature f ~570 * 4s = 38min. How? I used cross-correlation (Ocatve function xcov) and searched for the minimum. I than adjusted data by the found time lag and calculated correlation coefficient and the result is almost -1.
This is a good way to also evaluate humidtiy time constant.

However, circuit now needs some modification to get rid of it.

-branadic-

[Andreas]

Hello branadic,

38 minutes is quite large.
So either a large thermal mass is involved (3458A?) or the temperature sensor is not at the place of interest. (zener / DMM).
Perhaps you need more than one temperature sensor to decide between DMM or zener.

with best regards

Andreas

[branadic]

Hi Andreas,

38min is the change of voltage of the reference (inside a package of cotton covered by a small cookie box) by ambient temperature. So yes you are right, I need to dig into that and put a temperature sensor inside the cookie box, close to the reference.

-branadic-

[branadic]

To understand a device you need to understand the datasheet, here is part 1.

-branadic-

[branadic]

And the second part.

-branadic-

[branadic]

Looking at the given datasheet values results in an individual resistor between 800R (6V) to 700R (6.5V) for the circuit zlymex showed. So the parallel connection of the diodes I tryed is not the intended use. Need to solder a lil' bit and change my configuration.

-branadic-

[branadic]

What is interesting between two datasheets is the pin assignment:

Yongguang Elektronik Co., Ltd. states:
Output 1 indicates the negative terminal with color point, 2 is positive terminal, and 3 is empty.

While Shenzhen Shenxin Electronics Co., Ltd. states:
The color point of the pin "1". "2" is the negative end and the other end is the positive pole. At the end, the pin "3" is the spare pin.

So what is now correct for the Shanghai 17th Radio Factory diamond zeners that have been spread here?

-branadic-

[technix]

What is interesting between two datasheets is the pin assignment:

Yongguang Elektronik Co., Ltd. states:
Output 1 indicates the negative terminal with color point, 2 is positive terminal, and 3 is empty.

While Shenzhen Shenxin Electronics Co., Ltd. states:
The color point of the pin "1". "2" is the negative end and the other end is the positive pole. At the end, the pin "3" is the spare pin.

So what is now correct for the Shanghai 17th Radio Factory diamond zeners that have been spread here?

-branadic-

For the 17th Radio Factory, they make dual Zeners. One of the pin is shorted to the case, that pin is the common pin. It can be either common anode or common cathode, depending on what you ordered.

[GigaJoe]

was there any data about long term stability? or I missed it?
or it just something for a short term stability with very low noise ?
Does it make sense  to spare an effort for building something for a long them,  or rather move to LT1021,1027, ad587 ( I'm starting to hate all this low tempco precision resistors game)

[Magnificent Bastard]

was there any data about long term stability? or I missed it?
or it just something for a short term stability with very low noise ?
Does it make sense  to spare an effort for building something for a long them,  or rather move to LT1021,1027, ad587 ( I'm starting to hate all this low tempco precision resistors game)

These are buried (subsurface) Zener diodes, and so their LF noise time-drift should be inline with anything else you can buy with one of these type of diodes.  The operating current multiplied by the die temperature is what drives time drift rate with these.  The LF noise is related to 1/sqrt(Iz), so the operating current is a trade-off between time drift and LF noise.  At 5mA Iz, and a 45^oC die temperature, I would expect less than 1ppm/a time drift, and less than 1uVpp of LF noise.

[branadic]

The only data published can be found here

Partial translated:
"...Most devices have a 100-hour stability within 20 ppm. If the voltage change caused by the temperature is subtracted, then the true time drift may be smaller. Therefore, it can be said that there is no tendency to have any divergent drift. However, in view of the fact that there is still a lack of longer and more reliable data to prove, it is not yet possible to draw conclusions..."

-branadic-

[Kleinstein]

One problem with the 2DW232 refs seems to be rather wide scattering in properties, not just between sources, but also between the ones form a single batch. So the data-sheets are of a limited value - quite a few data shown in this thread show performance outside the specs. There is a tendency to need more current to get zero TC.

The reference itself is supposed to be a burried zener. However besides the intrinsic drift of the silicon, there is also a possible stress effect, e.g. from glue / solder to fix the chip to the case and a possible cover / case. So not all buried zeners are the same with drift. The quoted 20 ppm for 100 hours seem to be for a higher temperature (e.g. 100 C) - so this might be reasonable good, if used at a more moderate temperature. Given the large deviations from the DS so far, the drift data may not be that reliable. It takes years and a large number of refs tested to really know.

[branadic]

Don't call it problem, call it a challenge 

-branadic-

[GigaJoe]

buried (subsurface) Zener diodes, and so their LF noise time-drift should be inline with anything else you can buy with one of these type of diodes. 

I have a doubt about it, if the process would be so simple, then anyone can reproduce it.  But problems to grow a prefect semiconductor lay in  millions layers. starting with a regular process parameters, and quality of raw materials, and finish with fancy problems like unwanted injection of atoms that disrupting a structure, or metal atoms diffusion from conductor  that connecting atoms to output legs .... So it not soo simple ....

The variation of crystals may indirectly indicate an overall quality - that means a significant effort for selection, and even this doesn't guarantee a long term stability for a simply reason, during manufacturer process crystal was contaminated, and defects slowly spreading ....

[Andreas]

These are buried (subsurface) Zener diodes,

Hello,

I cannot find this term in the translated data sheet.
So my guess: they are (low noise) temperature compensated zeners like 1N821-1N829a

with best regards

Andreas

[GigaJoe]

it seems all long term  stable zeners in use technique called buried zener ... it a common name may not reflected in a documentation, if marketing dept. will not insist on ...