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

[2N3055]

Yea, correct.
I'll have to try pin 1-3 too

It should be the same...

[Gyro]

First candy test. Nothing fancy, just K2400 run with help of little snake.



Blue line - pin2-pin3 junction
Orange line - pin1-pin2 junction

Voltage sourced -10 to +10V, current compliance 20mADC

DATA 1 (XLS)
DATA 2 (XLS)

How long did this trace take and in what order did you measure the parameters (back to back zeners vs the single zener trace)? Unfortunately device (junction) temperature is the important factor rather than ambient temperature. If the device temperature changed at all during the sweep then the characteristics will have been changing too. There is little data in the critical conduction areas, where Temperature coefficients of the forward and reverse biased zeners are so critical. If you read back through the posts you will see that Alex found a TC turnover point at 1.4mA for the Zener alone and I found another at 20mA for series Zener and forward biased Zener (diode) at 20mA. I can't see either showing up on the sweep (although at -20mA it may just be visible at the botton left of the trace), also the slopes on the left and right of the trace are different too (right hand one is shallower with no hint of turnover visible), when I would have expected them to be near identical this makes me think that device temperature was changing,

EDIT: Sorry, I think I'm getting tangled up between I/V curves with TC.  Something still doesn't look right in comparison between the bottom left and top right slopes though (back to back Zeners).

[SvanGool]

@TiN
Could it be that either 1-3 or 2-3 is inverted (otherwise you have two zeners in the same direction, which is not the case according 1-2)?

[VintageNut]

Be aware that when you sweep across a sign change, the instrument is on two completely different ranges.

If you want to rule out different range offsets and/or different range linearities, only sweep from 0V to 10V then reverse the banana plugs at the SMU and then run the 0 to 10V sweep again. Now you have the full characterization and only use one range of the SMU.

Also, when you sweep from -10V to +10V continuously, the device is decreasing voltage (and heat) in the first half of the sweep and increasing voltage (and heat) in the second half of the sweep.

Sweeping from 0V to 10V for both halves of the curve may reduce the difference in heating as well.

These devices are very sensitive to change in temperature.

[Cerebus]

Be aware that when you sweep across a sign change, the instrument is on two completely different ranges.

Why do you think that?

Unlike some of the other 2xxx range of instruments, I don't have schematics for the Keithley 2400 but I'd be surprised if it doesn't use the same current balancing integrating ADC that the other 2xxx series instruments do and the 'source' side of things is explicitly called out as 4 quadrant in the specifications.

[VintageNut]

Be aware that when you sweep across a sign change, the instrument is on two completely different ranges.

Why do you think that?

Unlike some of the other 2xxx range of instruments, I don't have schematics for the Keithley 2400 but I'd be surprised if it doesn't use the same current balancing integrating ADC that the other 2xxx series instruments do and the 'source' side of things is explicitly called out as 4 quadrant in the specifications.

The calibration report with data would be the definitive document.

TiN, did you have your K2400 calibrated? Do you have the data report?

Attached is the pertinent info from a 2450 that I used. Source +10V actual is 10.00022. Source -10V actual is -9.99997

[Cerebus]

Be aware that when you sweep across a sign change, the instrument is on two completely different ranges.

Why do you think that?

Unlike some of the other 2xxx range of instruments, I don't have schematics for the Keithley 2400 but I'd be surprised if it doesn't use the same current balancing integrating ADC that the other 2xxx series instruments do and the 'source' side of things is explicitly called out as 4 quadrant in the specifications.

The calibration report with data would be the definitive document.

TiN, did you have your K2400 calibrated? Do you have the data report?

Attached is the pertinent info from a 2450 that I used. Source +10V actual is 10.00022. Source -10V actual is -9.99997

You said "two completely different ranges" and I want to understand why you think that and if indeed that is the case. The calibration points aren't evidence of that, just of one range with 30uV offset and 19 ppm gain errors (and doubtless some linearity error too) unless there's something about the way the K2400 works that you know and that I don't.

[TiN]

I'll calibrate and generate report for you once get back, no problem. I calibrated it some year ago, and two years before it was calibrated by Tek Taiwan. Here's that old report.

I'm out from lab next 5 days. I don't think this basic test suffer much from SMU specs. Sweep was rather fast,  ~2minutes total, with 5mV increments. I can run sweeps with 3458 as sampler, that shall fix linearity issue?

My plan for next step is to mount diode to copper slug, put YSI 44008 thermistor on it and attach to TEC for controlling temperature. Then I can run much slower sweeps with focus just in zener regions with temperature points from +15 to +65 in 1C step. Pretty sure can do 0.005C stability with the setup.

[Gyro]

Has anyone tried duplicating the voltage divider TC compensation adjustment that I posted?

Given that the connection between the zener and the forward biased diode/zener is brought out (unlike a 1N82x for instance) then I can only assume that this is how TC adjustment would normally be implemented (unless every application runs it at 50'C). Whilst running with an individual zener biased at 1.4mA appears workable, this doesn't tie up with any of the listed datasheet parameters.

[VintageNut]

Be aware that when you sweep across a sign change, the instrument is on two completely different ranges.

Why do you think that?

Unlike some of the other 2xxx range of instruments, I don't have schematics for the Keithley 2400 but I'd be surprised if it doesn't use the same current balancing integrating ADC that the other 2xxx series instruments do and the 'source' side of things is explicitly called out as 4 quadrant in the specifications.

The calibration report with data would be the definitive document.

TiN, did you have your K2400 calibrated? Do you have the data report?

Attached is the pertinent info from a 2450 that I used. Source +10V actual is 10.00022. Source -10V actual is -9.99997

You said "two completely different ranges" and I want to understand why you think that and if indeed that is the case. The calibration points aren't evidence of that, just of one range with 30uV offset and 19 ppm gain errors (and doubtless some linearity error too) unless there's something about the way the K2400 works that you know and that I don't.

No, I don't have special knowledge. I look at the calibration report and see that the + and - source have a different measured value absolute value. So, even if a device being measured were perfectly symmetric in its behavior for + and - voltage, the 2400 would not necessarily plot a symmetric curve.

The best thing to do would be to plot curves that are the measured values with the uncertainties added and subtracted. You have three curves. The measured curve and the curves that represent the most negative uncertainty and the most positive uncertainty. You cannot know symmetry any better than the min and max uncertainty curves.

In practice, most of the keithley gear is far better than the worst-case specifications. To know if and how much better, you have to perform some tests to quantify the offset and noise of the range being used.

[TiN]

I'd expect changes from selfheating in diodes would be magnitude more than this error from SMUs ADC (which is almost a clone of K2000). Anyway, running same sweep with 3458 in parallel is no difficulty.

[VintageNut]

I'd expect changes from selfheating in diodes would be magnitude more than this error from SMUs ADC (which is almost a clone of K2000). Anyway, running same sweep with 3458 in parallel is no difficulty.

Excellent. For symmetry of self heating I suggest starting at 0V and sweeping to 10V, maybe pause for some recovery and then sweep from 0 to -10V.

[plesa]

I'd expect changes from selfheating in diodes would be magnitude more than this error from SMUs ADC (which is almost a clone of K2000). Anyway, running same sweep with 3458 in parallel is no difficulty.

Excellent. For symmetry of self heating I suggest starting at 0V and sweeping to 10V, maybe pause for some recovery and then sweep from 0 to -10V.

Keithley 2400 has pulse option, which will eliminate self heating. Similar way I measured parameters of LTZ1000 on 2636B.
If we agree in meantime what and how measure, I can program it and measure all zeners which I will be distributing in EU next week when I will be back home

[TiN]

I was thinking of pulsed sweeps too. I believe 4142B SMU has pulsed modes natively.

[VintageNut]

I was thinking of pulsed sweeps too. I believe 4142B SMU has pulsed modes natively.

Pulse mode measurement usually is much smaller than 1 NPLC giving lower resolution. No free lunch. It will be instructive to see the results.

[Galaxyrise]

I took my "heated" 2dw232 and breadboarded it into the following circuit, just using cheap resistors.  The objective was to see if it would start, and to tweak the heater parameters. the 4.4mA zener current put the 0TC point of the zener is at a good place to heat the zener to.

It does start. The integrator gain was reduced 100x from the LTZ1000 circuit to get it to oscillate less. The result was much worse than I expected; almost 100ppm/C. 

[Kleinstein]

At 4 mA the TC of the zener section alone should be rather high (and the zener + diode) is likely not yet really low.

Even with external temperature control one should start with an current that produces a rather low TC. The main thing the temperature control is effective with is avoiding the second order TC - here even a rough regulation helps a lot. So a reasonable aim would be more like a current set to have a linear TC somewhere in the < 10 ppm/C range and than have temperature control down to the 0.1 K range. It might be possible to use the oven (e.g. use modulated temperature for adjustment) also to do in circuit adjustment of the current to get the TC really down (e.g. < 1 ppm/K).

There is nothing special with the 50 C setting : the curve from the DS is just a typical one, with a lot of possible variations between samples (e.g. current to get a similar curve might be anywhere from about 3 mA to 20 mA, or at a 5 mA current the maximum in voltage could be at a totally different temperature. So even with temperature control one should have an individually set current.

If one only has cheap resistors available one should use the unscaled output from the reference, the scaled up version can be still Ok to set the current.

For temperature control the internal sensor is rather slow - thus not that much gain possible in a simple configuration. To get a really good regulation one might need a second, faster sensor. One could still use the ref. internal sensor for the integral part. So the internal sensor still sets the final temperature, but the external sensor allows faster settling and reaction.

Using a resistor as a heat has the difficulty that this is square law heater - this would need a rather conservative / slow loop tuning. It might be better to use a transistor as a heater as this allows an nearly linear heater response.

[VintageNut]

I have been setting up tests and allowing the device to tell me what it wants to do.

The device is mounted to a banana plug and to a solid copper bar and wrapped in tissues. It is at ambient but is shielded from direct drafts.

Here is 6mA of current over a long time. You can see the HVAC system of my house trying to control temperature.

The thing to note here is that when first turned on, throwing away the first 50 to 100 samples, the 2DW232 voltage rises. So, the 6mA current is below zero TC for ambient temperature in my office/lab.

 

[VintageNut]

Next, I  forced 30mA, the max datasheet current, and watched. The voltage decreases indicating that 30mA is above zero TC for my office/lab.

Next, I performed a manual decade current search and determined that zero TC is between 10mA and 20mA ambient operation. Then I performed a binary search between 10mA and 20mA. The first good current is 16.25mA. The sequence was 10mA -> 20mA -> 15mA -> 17.5mA -> 16.25mA. 16.25mA is probably not optimal but it is very good for short term.

For 200 or so samples the standard deviation is less than 3uV. 

[Galaxyrise]

At 4 mA the TC of the zener section alone should be rather high (and the zener + diode) is likely not yet really low.

I don't think one can ever say the TC of the zener section alone is high without stating the temperature.  I went with 4mA to have a zero TC temperature that was sufficiently elevated from room temperature. I had started by heating the unit while measuring the zener and zener+diode voltages.  I chose the point where the zener voltage had zero slope as the heater set-point.  However, it certainly didn't work out this way, so something is wrong with my understanding or my technique.

If one only has cheap resistors available one should use the unscaled output from the reference, the scaled up version can be still Ok to set the current.

The 100ppm/C number was just the zener voltage; I wasn't expecting the scaled up number to be stable at all in this prototype. So as a result, the diode current was also fairly unstable, but I expected this to have a very attenuated result on th zener voltage.

For temperature control the internal sensor is rather slow

So perhaps my zero TC discovery was done too quickly.  I will try the temperature ramp again but much slower and see if the zero TC values move.

Using a resistor as a heat has the difficulty that this is square law heater... It might be better to use a transistor as a heater as this allows an nearly linear heater response.

I think the resistor's response is linear enough in the 20mA region that it should be stabilizing at.  I could mount the transistor on the case as well.  Hmm... I wonder if I could mount a SOT23 transistor to the bottom of the case...

[Kleinstein]

I don't think a sot23 transistor would give enough heater power without getting rather hot locally and thus causing steep temperature gradients and thus possible thermal EMF problems.  AFAIK the LTZ1000 uses something like 100-200 mW, with the heater directly on the chip. So I would plan with a similar power rating, maybe even a little bit more. I would more think about gluing an TO126 to the top of the case. Heating only the ref chip may not be that good, as the pins seem to be some kind of steel and thus could show thermal EMF. 

From the voltage versus temperature curve (for one "typical" sample) form the DS the second order TC is about -0.25 ppm/K², the data from Gyro give something like -0.33ppm/K².  An 50 K higher temperature would give an additional TC of some -30 ppm/K. So adjusting the temperature is not an alternative to finding the right current. To get an Idea one how to use the chips it would be good if we had data for the TC of one diode for a few different currents.

The data from Gyro also suggest that his zener + diode at 5 mA would have and TC of about 1/3 of the normal diode which would be -600 µV/K or -100 ppm/K. So if at all zero TC would be at low temperature and 5 mA or more practical a much higher current (e.g. 15-20 mA). Getting zero TC for the zener alone somewhere at 1-1.5 mA suggest a TC of about -2 mV/K or -300 ppm/K for the series configuration at this current.

Unless the current gets too low, I would prefer the lower current operating point with only the zener anyway. For a heated version this would alow for more insulation and thus also less heating power.
Needing something like 16 mA to get zero TC is quite a lot - so this ref. might be more practical to use as zener only and than lower current. An important factor could also be how fast the TC changes with the current, so how accurate the current adjustment needs to be - chances are this is easier at high current.

[Alex Nikitin]

Here is the data I've collected over the weekend. The lab temperature varied from 24C to 20C. I've measure the zener only voltage with 1.397mA current through it (10V source with 3.401K resistor) using HP3458A and the diode only voltage with 0.65mA current through the diode using 34465A. The temperature of the case was measure with a thermocouple couple of times, so the diode voltage of 715mV corresponds to 22C. The resulting graph is attached. It looks like for this current the zener goes through a tempco sign change somewhere between 20C and 21C. The vertical scale for the zener voltage is 10uV or about 2ppm/div, for the diode voltage 2mV/div or ~1C/div. The zener noise is very low and there is no "pop-corn" noise visible so far.

Cheers

Alex

P.S. - added a zoomed-in graph for the "flat" part from 6am till 10pm on Saturday. The vertical scale for the zener voltage is 0.1ppm, for the diode voltage - 0.2mV or ~0.1C/div

[Kleinstein]

The stability / noise looks good. For the noise it is hard to tell how much comes from the meter and how much from the reference. So even the low current is sufficient for a low noise reference.

From the long run, it looks like much of the variations is still due to temperature variations. As the current is already adjusted for a small TC, one might need temperature stabilization to get a considerably more stable ref.. Current adjustment can only compensate the linear TC, but the second order effect still remains. Here temperature stabilization is likely more promising than even more adjustment / compensation.

Looking at the two points, where the diode shows 717 mV, there is a small but visible difference (about 2 ppm). This is more than the expected uncertainty of the meter. So this would be a first indication for aging or maybe hysteresis, the two still unknown parameters.

A phase of stable temperature gives a nice looking curve, but not much to learn from this. A similar measurement with more temperature variations would be more interesting.

[Alex Nikitin]

Looking at the two points, where the diode shows 717 mV, there is a small but visible difference (about 2 ppm). This is more than the expected uncertainty of the meter. So this would be a first indication for aging or maybe hysteresis, the two still unknown parameters.

Looking at these two points, the voltage difference is about 3uV or ~0.6ppm . As one of these points sits on a relatively fast temperature gradient it could be a number of variations, including thermal effects with connections. Right now the whole thing is rather messy built (literally in last few minutes before i've left the office on Friday  ), with some of the connections made with standard banana connectors and crocodile clips. I was not expecting to catch sub-ppm changes  .

Cheers

Alex

[Squantor]

Well, I bought some 2DW232's from aliexpress. This was before the group buy action. After almost two months they have arrived. Just wanted to do a quick lashup.

Whats this? Seems like a bit of a crack in between the bottom and lid of the TO-5 package? Pop, Oh  ... Well, lets wait until the real devices arrive .