ADR1399 reference

[iMo]

..
I have also considered retrofitting an ADR1399 into my DMM6500 to lower the noise, but that's a project for when my warranty expires...

I would not do it, it will not help at all in your case.. In the 34401A with 399 the new 1399 moved the noise around the "plateau" in the ADEV from say 450nV down to 320nV. It is very little on the graph. And the 34401A is way much less noisier than the DMM6500 when looking at the graph in that area.

Moreover, I soon hit the hidden obstacle built in - while lowering the internal Vref noise (I also replaced the opamp in the integrator) now I can see buried "anomalies" creating a noise which cannot be decreased any more [easily].

Your DMM6500 is showing its limits for a noise decrease attempt already with the stock 399..

As Kleinstein wrote many times already - "..easier is to build a new ADC.." - or something like that.. 

PS: what could be an interesting measurement with the DMM6500 is to keep sampling for a longer duration - like a week or more - and look at the ADEV - how deep it follows the red line below (the line Keithley traded for the higher noise at lower taus).

[laichh]

Anyone here happen to have a thermal image of a ADR1399KEZ (8-Terminal LCC Package) in operation?

[Alex Nikitin]

My 2 cents. HIOKI DM7275 (LTZ1000) , Keysight 34465AM (ADR1399) (Left vertical scale) and Agilent 34970A (LM399) (Right vertical scale, before I've calibrated the 10V range)  , sampling at 10 NPLC (0.5s intervals) 10V from Fluke 731B for 2 hours. The 34970A was not fully warmed up from the look of it. I will repeat this measurement again for longer.

Cheers

Alex

[exe]

I'll post mine because why else did I buy dmm7510 and ten399 reference? I post two plots: one is last 24 hours, and another one when there was some unknown anomaly. Occasionally something happens and measurements are a bit off. Has anyone seen anything like that?

Measurement settings: 5 npls, impedance auto, average 20 samples per measurement. Additionally, there is a sliding window average of 200 samples to make graph smoother. The whole setup is on for last 113 days.

It looks like my setup is not far off to what other people see? With temperature correction it looks even better)

[iMo]

..the 1399 jumps too, from time to time.. 

PS: A high time ADI, TI, or whoever finally starts producing buried references with N zeners in parallel (like the LTFLU was)..

PPS: perhaps the REF80 does it with 4?

[Alex Nikitin]

..the 1399 jumps too, from time to time.. 

I have two units upgraded with ADR1399, the HP3245AM does occasionally jump. Not often, but documented (see attached) . The Keysight 34465AM works OK, been calibrated by Keysight twice, looks stable and so far I didn't see any jumps. I also have several units with ancient LM399s as references and some jump, some don't. My newly acquired 34970A occasionally jumps (I have seen it) . My HP3456A jumps though it is only visible if I average NPLC1 results.

Cheers

Alex

[Kleinstein]

Jumps of some 5 µV ( 7 µV after scaling to 10 V) is quite a lot - comparable and maybe even a bit worse than a LM399.

In my limited tests I saw some jumps, but it is hard to tell if it is from the reference, the DVM part or an EMI issue.
I remember from the LM399, that it reacts to the plastic cap moving side to side. The ADR1399 has essentially the same cap.

[iMo]

With a good 1399 you may wait for a week or more to see a jump. I have several 399s (out of perhaps 15) which "do not jump" simply because I have not watch them for long enough time yet.
I have one which never jumped and after 2 days in the DMM I got jumps.

Also it is important you compare the 399/1399 with a reference type which does not jump - like the battery, or perhaps the jfet based one, or bandgap one (do they jump?).
For example the DUT is the not-jumping reference, in the DMM you have the 399 or 1399 or other_zener_type_ref.

And you have to watch it for weeks, 100PLC is ok as the jumps are usually longer than X minutes.
Shorter jumps and more frequent one are not jumps but telegraphing, or popcorn, imho..

Anyhow, my bet all 399/1399 jump. Is that even mentioned in the DS? There is a nice noise p-p graph, but have you ever seen a graph with the jumps there in the DS?
The only way how to fix it is to wire say 4-8 onchip zeners in parallel (via summing resistors or something like that) such the jumps average out.

The LTFLU has got 4 zeners:
https://www.richis-lab.de/REF04.htm

PS: in my 34401P I put a small foam cap on the 1399/399, from the top side fixed by the aluminum shielding, so random movements of the cap are not possible..

[EC8010]

Shorter jumps and more frequent one are not jumps but telegraphing, or popcorn, imho..

Whether they're infrequent or frequent, they're popcorn noise. The defining characteristic is that any given jump is followed by another of equal amplitude but opposite polarity. I believe that all LTZ1000, LTFLU, LM399, REF102, etc voltage references have popcorn noise - the defining characteristic being the number of jumps in a given period.

I am quite prepared to believe that it's possible to improve upon the supplied plastic cap of a 399/1399; I wouldn't like to state that the supplied loose plastic cap isn't tribolelectric and incapable of producing a jump. Thermal effects and vibration are much underrated.

[Kleinstein]

The effect of the cap moving is likely thermal. Triboelectric effects are more a thing with high impedance. I don't think this is what makes popcorn steps - I just got comparable high effects.

I have also seen popcorn noise jumps also with 4391 type JFETs as a reference, just rather small size and more then 2 levels. With many test one would just not see them. It took looking at the difference to really see them.

So far with the LM399 I have seen a rather fixed amplitude ( 3-4 µV) but different rates for the jumps (from rough 1 per minute to 1 per hour). So far for the LM399 I have only seen 2 levels - though with many measurements one has 2 references and thus may not see if 3 levels  come from 1 reference.

[iMo]

One thing worth investigating might be how fast the edges of the jumps are, i.e. whether this is a fast or a slow event. I doubt thermal propagation could be that fast, but who knows - perhaps phonons could affect the unstable charge centres inside.

When you look at the shots of the refrence chips made by @Noopy, you can see nice strings of bright spots throughout the zener junctions. They are randomly distributed, and their number and intensity depend on Iz, as Noopy demonstrated several times.
So the material there is non-uniform, with random localized regions of high fields.

I think the jumps could come from the ion implantation process, where the crystal lattice is quite damaged by dopants being shot into it at high energies. The annealing may not completely iron out all of the damage. There could therefore still be some centres in unusual lattice positions, where charges jump between them from time to time. The constant amplitude of the jumps we see in 399/1399 may indicate what kind of localised traps or generation/recombination centres are actually involved in the process..

[miro123]

Based on my limited experience with three ADR1399 devices, I would say that the voltage jumps are very difficult to distinguish from the overall broadband and 1/f noise.

In my observation, they behave more like my two LTZ1000 references than the much noisier and more "jumpy" LM399.

[iMo]

..so when you see this - that is the "jump"..
This 1399 does it perhaps 1-2x per week..

PS: the leading edge of the jump is not steep with 100PLC made of 10x10PLC, of course.
Some 399/1399s make always positive jumps, some always negative jumps (vs. their nominal voltage).
I've never seen one which made positive as well as negative jumps.
So it might be matter of a single trap or a gen/recomb centre, imho..

[miro123]

The jumps here are some 5..6uV . I have observer such  huge jumps only of one bad batch of 8 out of 10 pcs LM399.
Is this the graph with lm399 or adr1399?

[iMo]

The graph above is with my 34401P (1399+OPA140 inside) against 11.6V battery (the DUT).
It is a zoom-in from the graph in my post #454 above.

PS: the ADR1399 is not a new design, it is the old LM399 with minor changes in some resistor's values (ie. 8x higher Iz). The cap is made of a new material, as well as there is the new ceramic package option. That had been reported by the product manager of the chip years back during a Metrology meetup, as I can remember..

[bsw_m]

I would like to share a custom reference core topology designed specifically to isolate the ADR1399 subsurface Zener from external disturbances. This circuit serves as the core reference engine for a traveling voltage standard. Complete data, system schematics, and performance logs for the entire standard will be published later in a dedicated standalone thread.
This schematic diagram does not show power supplies, output buffers, etc.
The proposed analog core for the ADR1399 features a fully floating Zener topology. This design isolates the reference from ground loops and EMI

[miro123]

Interesting approach. I like the idea too.
I try to figure out how the circuit works but i did not succeeded
e.g U8A seems on input to be gnd and another tied to zenner positive side let say +7V . Is this the intention?
do you vant to have ADR1399 Pin 1 close to GND and ADR Pin2 close = -7V

[mawyatt]

I would like to share a custom reference core topology designed specifically to isolate the ADR1399 subsurface Zener from external disturbances. This circuit serves as the core reference engine for a traveling voltage standard. Complete data, system schematics, and performance logs for the entire standard will be published later in a dedicated standalone thread.
This schematic diagram does not show power supplies, output buffers, etc.
The proposed analog core for the ADR1399 features a fully floating Zener topology. This design isolates the reference from ground loops and EMI

What's the noise performance wrt to the "normal" grounded configuration?

Best

[d-el]

do you vant to have ADR1399 Pin 1 close to GND and ADR Pin2 close = -7V

U2 pin2 -7V
U4A and Q2 generate +3V (7V×(6.6k÷15k))

[Kleinstein]

The noise should be a little higher than in the standard configuration. Like 2 and a half OP-amp contributing instead of just 1.
The typical precision OP-amp, like OP07 and modern alternative should have significant lower noise than the reference. So the difference would not be large.
I see little advantage over the standard configuration, more like addional points to go wrong.

[Alex Nikitin]

The noise should be a little higher than in the standard configuration. Like 2 and a half OP-amp contributing instead of just 1.
The typical precision OP-amp, like OP07 and modern alternative should have significant lower noise than the reference. So the difference would not be large.
I see little advantage over the standard configuration, more like addional points to go wrong.

Also the offset drifts of opamps would add variations to the TC (and to the long term drift if you are not careful with the opamp choice) as a 0.7uV/C drift (reasonable value for a non-chopper precision opamp) would be 0.1ppm/C at 7V reference voltage.

Cheers

Alex

[bsw_m]

Indeed, the performance requirements for op-amps U4, U6, and U8 are quite stringent, and using something like the OP07 is not recommended here. The actual device utilizes TP27-SR  op-amps at all position.

[Alex Nikitin]

The actual device utilizes TP27-SR  op-amps at all position.

It is a chopper amp, isn't it?

Cheers

Alex

[bsw_m]

The actual device utilizes TP27-SR  op-amps at all position.

It is a chopper amp, isn't it?

Cheers

Alex

The TP27-SR datasheet does not explicitly mention 'chopper' or 'zero-drift' architecture, but its specifications clearly indicate that one of these topologies is being used. In any case, their real-world performance is an excellent fit for this circuit and will not become a bottleneck. Among the more common op-amps, the OPA189 would be my first choice to consider.

[bsw_m]

What's the noise performance wrt to the "normal" grounded configuration?

Hi mawyatt,

That is a crucial question. Mathematically, since we have multiple op-amps wrapped into interlocking feedback loops, their individual noise contributions add up orthogonally (as the root of the sum of squares).

In a standard single-ended configuration (Zener anode tied straight to hard ground), the output noise floor is determined by the Zener itself plus just one scaling/buffer op-amp.

In this floating core topology, the noise is slightly higher on paper because three op-amps (U4, U6, and U8) actively manage the biasing and potential translation with an error transfer coefficient of 1.44 each. The current-source op-amp (U3) is mathematically heavily attenuated (by a factor of ~0.00022 due to the R9/rz ratio), so its noise contribution is virtually zero.

However, in practical terms, this does not become a bottleneck at the 10V output terminals across the entire technological spread of the ADR1399 silicon.

The TP27-SR  op-amps used in this design with low noise floor of just 0.1 µV p-p in the 0.1–10 Hz band. When we calculate the cumulative noise contribution of all three active op-amps at the output, it scales to roughly 0.25 µV p-p.

Let's look at the final RSS-summed noise on the 10V terminals for different Zener selections:

1. For a hand-picked, "golden" Zener sample (1.0 µV p-p at the Zener level -> 1.44 µV p-p at 10V output):
V_total = sqrt(1.44² + 0.25²) ≈ 1.46 µV p-p  (Actual noise penalty: ~21 nV)

2. For a typical datasheet Zener (1.4 µV p-p at the Zener level -> 2.02 µV p-p at 10V output):
V_total = sqrt(2.02² + 0.25²) ≈ 2.03 µV p-p  (Actual noise penalty: ~15 nV)

3. For a worst-case datasheet Zener (2.2 µV p-p at the Zener level -> 3.17 µV p-p at 10V output):
V_total = sqrt(3.17² + 0.25²) ≈ 3.18 µV p-p  (Actual noise penalty: ~10 nV)

Summary:
The total noise penalty of this multi-loop architecture compared to a standard grounded configuration is roughly 30 nV at the 10V terminals. 

The baseline noise floor remains strictly limited by the physical properties of the ADR1399 silicon itself. We are essentially trading an invisible 30 nV worst-case noise penalty for improved immunity against EMI/RFI and ground loop currents.