The LTFLU (aka SZA263) reference zener diode circuit

[Kleinstein]

The tighter specs could be simply because of stricter criteria for acceptance, that may be needed to get the zero TC point under reasonable (e.g. not to low transistor area, to low or high a current). I would not take the span for min/ max values to serious - those specs may not reflect the actual scattering seen. This are more like test limits that may not be updated if the process technology improves.

The more odd entry in the table is the LTZ1000 with a rather high voltage and AFAIK consistent positive TC of around 50 ppm/K without the heater.

The adjustment range for the TC via current density is not that small. They may have to reject some chips (possibly whole wafers) at the extremes, but that is part of the reason why high performance chips have there price.

[Micke]

I had a FLUKE 8842A mainboard that was only good for parts, so I decided to cut out the RefAmp assembly (SZA263 Datecode 8430) and make a 10V DC reference (in addition to my previous 2x LTZ1000A, 1x LTZ1000 and 1x FLUKE 731B)
The 7->10V booster was in turn cut out from my own designed LTZ1000 PCB  Again recycling! (Have just assembled LTZ #4, currently under burn in with elevated heater temperature, parallelled 1k with 4.7k giving ratio 15.77)
Voltage from RefAmp board was 7.00090V, so I ordered custom precision wire wound Ultrohm Plus resistors (10k and 4.284k 0,01% 3ppm/C) from Edwin Pettis, and without trimming the voltage out is 10.00006 on my FLUKE 8846A (which I know show 3ppm high on 10VDC), so only +3ppm without any trimming, impressive! 
In the box I put a linear +/- 15V supply, LTC2057 chopper amp, 7->10V boost resistors with copper tape for thermal coupling, and a NPN darlington output buffer stage (inspired from KJ7E, thanks!). I really like those hermetic resistor networks on the 8842A RefAmp assembly, the 8846A is also full of them 

[Noopy]

Btw, one unit was marked as LTFLU-1CH while the other was marked as LTFLU-1ACH. I leave it up to you to find out what's the difference between both.

-branadic-

I took a better (cleaner) picture of the second LTFLU. But nothing special to see. Found no difference between the LTFLU-1CH and the LTFLU-1ACH.

https://www.richis-lab.de/REF04.htm

The surface and the colours are looking different. Perhaps the manufacturing process has changed. Perhaps it´s coincidence… 

[Conrad Hoffman]

Is there any reasonable source of the LTFLU parts to buy? I found some for 3/4 the price of an LTZ1000 and I'd only pay that much if I had to repair something.

[BU508A]

Is there any reasonable source of the LTFLU parts to buy? I found some for 3/4 the price of an LTZ1000 and I'd only pay that much if I had to repair something.

I've bought some LTFLU-1 from Walton electronics in Shenzhen (Alibaba). They look very genuine.
The pictures above from Noopy came from 2 LTFLUs which I'd sent Branadic for taking some HiRes pictures under the microscope.

https://www.eevblog.com/forum/metrology/the-ltflu-(aka-sza263)-reference-zener-diode-circuit/msg1282144/#msg1282144

[dietert1]

Wrote a little report on some nice results with the LTFLU references i showed above.

http://www.cadt.de/metrology/2020-05-18 Vref_Ovens.pdf

Regards, Dieter

[notfaded1]

Wrote a little report on some nice results with the LTFLU references i showed above.

http://www.cadt.de/metrology/2020-05-18 Vref_Ovens.pdf

Regards, Dieter

This looks pretty promising Dieter.  I wonder if I can build a nice ovenized 10V reference with my LTFLU in similar way to what Micke did but by putting it in an oven?  I'm not sure I fully understand how the NPN darlington output buffer stage works though... I'm not an EE?

Regards,

Bill

[Micke]

This looks pretty promising Dieter.  I wonder if I can build a nice ovenized 10V reference with my LTFLU in similar way to what Micke did but by putting it in an oven?  I'm not sure I fully understand how the NPN darlington output buffer stage works though... I'm not an EE?

Regards,

Bill

I did not add an oven, just reference circuit "as-is" from the FLUKE 8842A DMM...
The Darlington output buffer is to protect the SZA263 from short circuit on the output, and increase the output capacity, can´t find my notes right now, but loading output with 100mA was no problem, with not that many PPM´s Voltage drop    (adding heat sink on transistor could be good though)

[dietert1]

You need two high quality references to do that fine-tuning. I mean unless you want to spend thousands on a HP 3458A or the like.
Later i want to repeat the same for the other LTFLU reference, the one i made in January. That one was tuned using our old Geller AD587 as reference.

Regards, Dieter

[Kleinstein]

Usually an oven around the reference can keep the temperature stable to better than 1 K , often more  like 0.1 K.
So there is no real need to get the TC trimming so extremely good.

[dietert1]

Sometimes i am wondering whether you believe your own statements here. When the LTZ1000 has a typical TC of 50 ppm/K at its operating temperature and is used for an 8 digit multimeter (resolution 0.01 ppm), i know that the temperature stability of the LT1000 oven is about 0.2 mK. Your opinion is welcome, yet irrelevant in this case.

Regards, Dieter

[SilverSolder]

Sometimes i am wondering whether you believe your own statements here. When the LTZ1000 has a typical TC of 50 ppm/K at its operating temperature and is used for an 8 digit multimeter (resolution 0.01 ppm), i know that the temperature stability of the LT1000 oven is about 0.2 mK. Your opinion is welcome, yet irrelevant in this case.

Regards, Dieter

What about all the resistors and other components around the LTZ itself?  Ovenizing all that external stuff seems a good idea, on the face of it?

[Kleinstein]

With the LTZ circuit the effect of the other resistors is attenuated quite a bit (e.g. a factor of 80 or more). For ultimate TC performance there usually is an adjusted R9 to compensate residual TC and this would naturally also include the small effect of the resistors.
The more critical point for the resistors is usually long term drift, not the TC. Here an oven does not help much.
One still looks for low TC resistors as the long term stable resistors usually also have a low TC. If there were good data on long time drift one would look for these.

For the SZA/LTFLU circuit the attenuation effect is similar, possibly even a little better. Here it is more convenient to have more/all of the circuit also in the oven, as there is no internal heater.

[BU508A]

For the SZA/LTFLU circuit the attenuation effect is similar, possibly even a little better. Here it is more convenient to have more/all of the circuit also in the oven, as there is no internal heater.

That's probably why Fluke put them in an oven as well in the 732B.
Here is a teardown of the 732B with some nice pictures:

https://www.eevblog.com/forum/testgear/fluke-732b-dc-standard-teardown/

[SilverSolder]

For the SZA/LTFLU circuit the attenuation effect is similar, possibly even a little better. Here it is more convenient to have more/all of the circuit also in the oven, as there is no internal heater.

That's probably why Fluke put them in an oven as well in the 732B.
Here is a teardown of the 732B with some nice pictures:

https://www.eevblog.com/forum/testgear/fluke-732b-dc-standard-teardown/

It doesn't look like Fluke heated the resistors just out of convenience (i.e. because the resistors were already on the board so might as well bake the whole thing), they actually went to the trouble of making an extra cavity in the oven assembly just to enclose the resistors.  Even if suppressed by a factor 80, it might buy us a wider environmental operating temperature range, compared to leaving the critical resistors in free air?

[Kleinstein]

In the 10 V reference there are also resistors for the 7 V to 10 V step. These resistors are the really critical ones. With the SZA263 and similar this gain stage is often part of the reference directly.

[nnills]

In the 10 V reference there are also resistors for the 7 V to 10 V step. These resistors are the really critical ones. With the SZA263 and similar this gain stage is often part of the reference directly.

Looking at the schematics in the F732B manual, this seems to be the only mode of operation of this IC. There must necessarily be a voltage differential across a resistor(R3) somewhere. As I see this these gain resistors(R2/3) are always present. And their TC is only attenuated by a factor of 1.4, necessitating ovenising or good TC matching. If this matching can be achieved, no oven is needed as the IC with the proper biasing already has no TC(although in a small range).

[dietert1]

I measured the series of the outer two resistors (bottom and top) today (connected to second and third pad from the top row on the left). The sum of them is 200ohms, so each one is 100ohms.



-branadic-

Today i also measured the LTFLU on chip heater, with the result of 180.5 Ohm, so that's a confirmation. Wasn't able yet to probe the 4 separate zener voltages. Need to get some kind of micro manipulator.
That LTFLU is one i pulled because its transistor had a low hfe. Under the microscope it appears like none of the fuses was blown. Maybe they had to reject LTFLUs with low hfe, since they did not know which one of the eight transistors was bad.

Have another result:
I found the Vpp = 6 stdev repeated so often in this forum is a rule of thumb. Of course with constant noise level the maximum excursion will increase the longer you wait. The diagram shows some experimental numbers i got from my LTFLU logs. Error bars indicate error of expectation value, standard deviation is bigger. For example when looking at subsets of 100 samples, the ratio of Vpp over standard deviation has a standard deviation of 0.74 so likely ratios to be observed are between 4.6 and 6.1

Regards, Dieter

[dietert1]

I interpreted the structures this way:

From each zener there are four "resistor wires".
Every resistor is connected to the cathode.
Two parallel resistors are connected to the bondpad in the upper right corner.
One is connected to the emitter of the transistor batch. I assume it´s some current dividing.
The last one is a possible low impedance path instead of the resistor for dividing the current.
....

I can also confirm that interpretation. Measuring resistors and probing voltages from the chip i arrived at a schematic of the internals. I used noopys image to mark some structures. The path lengths were determined as 118 pixels for R7 and R8 and 317 pixels for R14. A nice agreement with the measured resistances. What noopy called "low impedance path" may be test points for each buried zener voltage, that isn't accessible otherwise.

Regards, Dieter

[Noopy]

What noopy called "low impedance path" may be test points for each buried zener voltage, that isn't accessible otherwise.

Hm... I don´t think that small metal squares are used for probing... They are veeeery small. 

[razvan784]

I think those are simply contacts. Because IC masks cost a lot, it was common practice for a design to allow for possible variants, where these variants have the same masks for all layers except the top metallization. You see this with the LTZ as well, where several heater resistors are fabricated, but only one is actually used. They probably tested all these variants and kept the best performing one. They kept those isolated contacts because the fabrication process requires them, you cannot have an isolated via.

[dietert1]

So by changing only the final metallization they could have connected the emitters to what i marked TP1 .. TP4 instead of using R13 .. R16. The alternative/optional averaging resistors would then be about 35 Ohms. Added those to my schematic.

Regards, Dieter

[branadic]

I meanwhile played a lot with the oven construction for my aluminum LTFLU board. Turned out to be a rabbit hole.

I bought one of those cheap W1209 thermostate modules and tried using it together with a 10k NTC (some 0402 NCP15XH103) on the LTFLU board and with the heater resistor BPR10J101 attached to its back. Turned out to produce large thermal gradients and large switching noise.

I then designed an aluminum case and had it fabricated at work, with the LTFLU board put upside down into it, so that the components are surrounded by the case. Again, this construction failed.
I then tried several positions of the heater resistor attached to aluminum case itself and even tried a seperate 10K NTC, insulating the room between case and components with styrofoam, also insulated the whole assembly, but whatever I tried, nothing worked. So I skipped the idea with the W1209. It is now used for pre-aging some LTZs, but that's a complete different story.

After some weeks of being disappointed I went back to my initial idea, having the heater resistor directly attached to the back of the board and using the onboard NTC. This setup was then put into a small styrofoam box together with some cotton ball, insulting the complete setup.
This time and since it currently wasn't in use I attached heater and NTC to my Arroyo 5305. After setting all parameters including the ones for the NTC (only Beta and temperature curve is given in the datasheet, so you have to reveal the coefficients for the Steinhart-Hart equation e.g. using the SRS Thermistor Calculator) I started the autotune function and indeed, PID parameters where found. Nice, that is something to work with.
Having the oven temperature set to a fixed value and running it all for 24h showed the temperature is rock stable. Even the last digit of "48.50°C" didn't change during that time. Wonderful.
Next thing to do, measure stability of LTFLU with DVM/DMM and the oven control running.

Lesson learned: Switched ovens do not work for precision stuff and you want linear regulation instead. I learned it the hard way and hope that helps someone at some point.

-branadic-

[dietert1]

Those Arroyo Instruments TEC controllers provide for a temperature resolution of about 0.02 to 0.05 mK. The instrument works like many DVMs: On power-on it defaults to a numerical resolution of 10 mK and to a resistance resolution of 1 Ohm for the 10K thermistor. With the command "SETEXTENDEDRES 3 \n" you can turn on the reserve digits and the instrument will deliver 1000 times more resolution. It incorporates an ADS1256 24-Bit ADC for temperature measurement. Don't know (yet) what is the typical TC of its reference resistor.

Regards, Dieter

[BU508A]

Regarding the Arroyo temperature controller, there is a nice video from Marco Reps about this unit.