ADR1399 reference

[branadic]

Is that the latest schematics and firmware? I think there were some changes after the evaluation here by Andreas and Kleinstein..

I have a newer, slightly modified board version that also covers the ADR1399, so with the additional required filter.

-branadic-

[Conrad Hoffman]

Continuing with the measurements, I painted the top of the package black so I could get a good temperature reading. Seems to sit about 150-155F. With a small plastic cup to shield it, the heater current drops to 9.4 mA. I wonder if the die has some thermal insulation between it and the package.

@branadic is that board available to purchase, or Gerbers?

[branadic]

Attached is the latest version of that board.

-branadic-

[miro123]

R11 can be ten.. twenty times higher. 10K is good starting point.
On such way you are avoiding the the cross couple with the noisy +15V line. For more info look at HP34401 schematics

[DavidKo]

I have tried to design the PCB for LCC version, but I'm not sure which way to go. The schematic is based on Analog design, I have only tried to add the NTC and have as much variability as possible (1 or 2 separate power supplies, with or without protection diode, added measurement/power supply point). But where I'm not sure is PCB design. I have prepared "neck" design, which is based on Analog PCB and later I have completely redrawn it to spider design, but I'm not sure which way to go.

Additionally if you have the idea to make it even more universal you are welcome. At the end I want to have 2 PCB on 100x100mm plate with power supply in between and maybe also the amplifier to have 10V output.

[branadic]

My recommendation for LCC package still is to put it on a flex board or use a flex-rigid board with the LCC package located on the flex part.

-branadic-

[Conrad Hoffman]

No expert by any means but the neck design looks more straightforward. Not sure the spider buys much. I'm convinced that lowering the conduction out of the LCC package is important. Hollow an area under it- have to run traces outside only if you do that. I don't see any reason for the NIC pads to be there at all. I really want a 10V output on my board and it will almost certainly use thru-hole resistors in some pad-able/trimmable arrangement. On a small board I don't know if the sense connections buy anything if the traces have decent width. Caveat- I haven't sat down with PCB software and tried to do this yet!

[DavidKo]

@branadic I have checked the flex board pricing and for the first experiments it does not make too much sense. At least I have found prizes > $100. I will rather use 0.6mm pcb (prize is the same as for standard one $2 + shipping). But maybe it make sense to create on thin PCB only the "adapter" with pin header  and create separate power and amplifier PCB from standard 1.6mm FR4. The idea with header have also advantage, that later the reference can be used in different design without need to desolder it.

@Conrad Good point, I will remove the NIC pads


The enclosure around LCC package should be 3D printed or foam to hold the reference package in "stable" temperature generated by the inner oven.

[cellularmitosis]

I have tried to design the PCB

Hi David,

Are you familiar with the "bootstrapped zener" circuit?

[Kleinstein]

IN principle the kelvin connections with seprate sense and drive make some sense. However with the ADR1399 there is quite some current (2.2  mA) through the sense pins and this makes the separate sense pins essentially useless to compensate for trace and bond wire resistance, as most of the current still flows through the sense side. The separate force and sense pins still make sense to provider a lower output resistance, so to replace an extra buffer. However this is a relatively rare use case.

Another point may be a few more options for the compensation, though the externa 1 µF + 5 ohms does not look so bad. For the shown impedance curves it looks like 1 µF is more on the low side and if a MLCC is used one needs to take into account the reduction in capacitance with bias. So I would more like plan for 2 x 1 µF or 2.2 µF or even more if this is a X7R or similar capacitor.
 

To get a 10 V reference, the classical bootstrapped supply to the zener should be the best solution. There is than no real need to serparate the drive and sense side and one could join them right at the chip.
One could still seprate the current and voltage path after connecting sense and drive.

Even if mainly used as a 10 V reference it would make sense to also have the option to measure the raw 7 V. This would allow to see if drift is from the reference itself or the resistors to set the gain.

[DavidKo]

@shodan@micron Kelvin point should be the two "crosses" close to the output. But I have messed that out during my rework of the PCB (the crosses were right to the reference without cut out). The output should be directly connected the REFS not from current crosses. It will be easy modification.

@cellularmitosis Such an approach can be used since the the heat power and reference power are separated. I think that the bootstrapped circuit is used in LM399 portable calibrator mentioned in datasheet. The bootstrapped circuit can be part of the free space behind the reference.

I'm aware of quite high current to the sense part, that is why I have implemented transistor between power and REFS as close as possible to decrease the current as in Analog proto board. The board itself is intended to be experimental to try everything what came to your mind. This is also the reason for jumpers. For example I want to try to drive the reference sense with 2.2 mA from external current source (bootstrapped circuit will be good for it). Right now I have 3rd iteration and still I'm finding things to improve and every additional idea counts. Firstly I had planned to be able to use also TO version for ADR/LM 399 reference, but I had left that idea since there are better designs already available.

Final design should be on 100x100mm board with 2 references, stabilized power supply and amplifier to 10V (ideal will be 2, but with option to use 1 amplifier and paralleling the references to experiment with noise). Thank you for your remarks, I want to remove big flaws before I publish the design.

[Conrad Hoffman]

I've used the bootstrap circuit with good success when using 1N829 type diodes. It's what I had in mind for this, as I really want something just a few PPM under 10 VDC. Classic opamps might be OP-07, but what makes sense in 2022?

[Kleinstein]

For the amplifer on could consider an auto zero type like LTC2057,  MCP6V51 or OPA189.  There are a few slightly more modern alternatives to the OP07, like OPA207, OPA202 or ADA4077. The drift and noise from an OP07 or similar would still be less than the reference. So there is no absolute need for an auto zero OP-amp, but it would not really hurt either. Most of the modern ones come in SMD only.

[iMo]

..I will rather use 0.6mm pcb..

You may create "microslots around the pads" (no idea what is the thinnest slot they can mill), say 0.3mm thick slots - that gives some flexibility too..

[Conrad Hoffman]

For the amplifer on could consider an auto zero type like LTC2057,  MCP6V51 or OPA189.  There are a few slightly more modern alternatives to the OP07, like OPA207, OPA202 or ADA4077. The drift and noise from an OP07 or similar would still be less than the reference. So there is no absolute need for an auto zero OP-amp, but it would not really hurt either. Most of the modern ones come in SMD only.

I used an auto-zero in the Mini-Metrology reference and it worked OK. It didn't have much drive current, which messed me up the first time I tried to use it with low value feedback resistors. It was also easy to damage, which is OK if in a socket, but not so much for SMT parts. I'm surprised the OP-07 is still competitive! I'll look at the other choices.
 

[Kleinstein]

To get a little more drive power, one can add an NPN transistor as emitter follower in the loop. The gain setting divider already provides in minimum load.

The OP07 is not really that competative - it is just one that is available in DIP and can be relatively cheap (some versions). Especially the power consumption is a bit on the high side compared to modern alternatives.

[Extrasolar]

My Take to the 8LCC package, I opted for symmetric copper polygons or traces.

[DavidKo]

@Extrasolar: My solution was symmetrical till I have added the transistor between REFF and REFS.

@Kleinstein: I want to have DIP and SOIC option on one PCB, idea is to have "integrated" SOIC adapter.

@imo: I can try that if there will be space on PCB. Additional transistor and NTC added additional connections.

[DavidKo]

I have simulated the bootstrap circuit.  I have added TDP1603 in second iteration (LT5400-1 is used since it is in LTspice database). 200k start-up resistor is used since it is in original LM399 datasheet. I had played with "tuning" resistor placement and the 4R7 at the top was the winner. When I had placed the resistor in the bottom, the resistor was quite big in comparison with 1k - 1k resistor had tuned the resistor placed there.

[iMo]

The problem with your LT5400 design is you are using 2 packages.
The beauty of the LT5400 is its "ratio tempco" lower than 1ppm/C (ADI told me). That is achieved by placing all 4 resistors on the same chip. When you combine two packages you break that principle.
I think there is no way to achieve <1ppm/C other than with a single LT5400 package.
I did with single package w/ 10k resistors and my several measurements resulted in 0.5-0.6ppm/C (LM399), but you would get 10.57V or so. Any other discrete resistor added to the 5400 will kill that results, sure.

PS: LT5400's resistor tempco is typ 8ppm/C and max 25ppm/C. Only the "ratio tempco" between the resistors (like tempco of  R1/R2, or (R1+R2)/R4, etc.) on its chip is <1ppm/C.

[DavidKo]

The problem with your LT5400 design is you are using 2 packages.

Probably it was not clear from my statement. LTspice does not have TDP1603, which I want to use (8 resistors in one package). Thus I have used 2x LT5400-1. I can have 8 standalone resistors in schematic instead, but this was simpler.

[iMo]

Ok, when you want to simulate tempco you may simply add to the resistor value the tempco, like  "10k tc=10e-6" for +10ppm/C and thus simulate N resistors with the same/similar tempco when in the same package.
".step TEMP 20 50 1" will then step your simulation from 20 to 50C step 1C.

[Kleinstein]

The TC specs are usually for a relatively large range, considerably larger than relevant to a precision measurement. Much of the TC extremes usually happen at the temperature extremes. This especially applies to the better grade resistors, where it is not just the linear TV but also contributions from higher order terms. For the reduced temperature interval alone one can expect resistors to behave quite a bit better than the limits for the larger range.
I have done some tests with NOMCA and ORN resistor arrays. In the tests involving different combinations the relative TCR was usually below 1 ppm/K, at least near room temperature. This is despite having much more tolerance in the specs.
I would anyway less worry about the TC. The critical point is more the hard to predict long term dirft.

From the excess noise the NOMCA type did not perform that great (barely meeting the specs), so I would not really recommend them. The noise of the LM1399 should still be a bit higher, though not much.

In some circuits one can still combine the resistors from 2 chips, if they are mixed equally between the chips. Even if not perfect some interleaving can help quite a bit.
A gain of 1.5  (e.g. use 3 equal resistors) combined with a -5% drop would be a real option, with the resistors for the 5% being about 20 times less important.

[iMo]

I burned X megawatts of electricity while simulating all possible variants of the divider, but at the end of the day I always finished with a single LT5400 and 10.57V as the simplest and cheapest solution with lowest achievable tempco..    Of course the long term stability of the divider is even more important (as you can measure the temperature pretty well)..

[DavidKo]

@imo: Mouser pricing TDP16031001BUF 6.62EUR (500ppm@2000h) vs. LT5400BCMS8E-4#PBF 10.03EUR (not in stock, <2ppm@2000h, cheapest version with worse matching that is not important in our case). For the testing purposes the Vishay is at least 3 times cheaper.