Ultra Precision Reference LTZ1000

[EC8010]

BF245C is no longer made and probably unobtainable, but J112 (a switching transistor) can be used in its place. The J112 is not as linear as a BF245C, but that won't matter in the proposed use.

[Andreas]

Hello,

if you want to use the FET the BF256B would fit better (and is still available).
Take care of different pin-out.
Other possibility would be MMBF5486.

with best regards

Andreas

[grx]

Thank you very much for you design wisdom. Avoiding self heating of the opamp is an interesting idea and since all these components can be bypassed I will add them.

I wonder about the R13 resistor to ensure startup when there is no FET, isnt that role already assumed by the usual D2 series diode? This seems redundant. The CERN reference has this diode between the fet and the 70k resistors, and no pull-up. I wonder if that pull-up is a regulation risk, as it is directly coupled to the positive supply. You have a very good regulator on your board, so that should be okay, but in my case, the regulator is not on the same board to avoid thermal influences, so I dont like the presence of R13.

Is the presence of the FET entirely replacing D2? since the gate-source/drain junction is a diode itself?

Farnell only has the J109/110/111/112/113 in through hole components. I will test that. I dont want SMD on this board except the unavoidable LTC2057 output buffer.

Also, could a NPN BJT be used here instead? Like, a BC108 in metal package, as I am planning to use for the heater transistor.

Finally, for the output buffer, I notice that you have a 7.14v->10V boost stage via the 25k and 10k resistors. This is another source for drift. If I want an unity buffer, can I just short the 10k resistor and omit the 25k one?

I have attached my current revision of the schematic.

[iMo]

There are nice ready boards for example (older kicad versions):

Several versions of the board by Andreas (with 7V buffer):
https://github.com/cellularmitosis/px-ref/tree/master
Frank's version with 7V->10V buffer:
https://github.com/cellularmitosis/dr-frank-ltz1000/tree/master
Simple version without buffer:
https://github.com/evetere/eV-LTZ1000

PS: any idea how to migrate the libs used above into the KiCad 10.0.1?

[grx]

Haha... For sure there aren't thousands of ways to execute this design, mine is quite close to the px ref in fact. And the point is making one myself, otherwise I would not have bothered with my own design.

Thanks for the reminder of the proper references!

Looking at this,

* I have now confirmed that I will use the same unity buffer as CellularMitosis PX design, it's just a protection measure. I'll build a 10v buffer on another board.
* still no LDO on this board, again I dont want to heat the reference board for no reason so I'll add that outside
* I will experiment with the FET current driver, while having the option to get rid of it.

One of the last questions now is about the criticality of kelvin sensing the top of the zener. I have added a footprint for a low value cathode tc resistor, so the zener voltage to the buffer is kelvin sensed by design, but the question is rather to determine if the kelvin connection is also critical for the heater divider and the 70k resistors. Currents in these branches should be stable, but I guess I will do it (I fear the pain) at least for the intellectual satisfaction. edit: I did that and that was okay.

Design is almost final now. I still wonder about that bootstrap resistor to the zener voltage. Oh well I have placed it and I will experiment.

[iMo]

You are missing a capacitor from pin 3 of the 2057 to ground (an RC filter)..

Also why to wire the protection diode in series with Vcc? That creates a drop with -2mV/C (not sure what is the impact here, though). I would put it in parallel with Vcc/GND and I would use a PSU with a current limit, like , say, 50mA max..
A TVS diode there might be handy (like say an 18V one with 15V Vcc). Another TVS at the 10V output (like 12-13V one).

And you may place there an optional LT5400 resistor pack with the 9k and 1k resistors inside (<1ppm/C ratio TC), and do wire it like a 2k/4k5 divider in the 2057's feedback (LTZ voltage x1.444) and you will get the Vref close to the 10V (you could hardly get a better divider cheap)..

PS: and frankly, I would not mess with the jfet there..

[grx]

Hello,

Thank you for these interesting remarks.

I usually put protection diodes in series to avoid shorts when they are misconnected, but in that case you are right, the temperature dependent drop is not the brightest decision, I will update that for a shorting protection and I'll put a current limit in the psu.

I have added a 100n film cap to the LTC2057 non inverting input after I posted this schematic, thanks for this.

I do not wish to add a 10V stage on this board, because I want this module to be as stable as possible, without additional drift from the divider resistors. If I learnt one thing from this very long topic, is that the output stage will never be as stable as the reference itself, so I want to isolate that. The output buffer on this board is as simple as possible, with the only goal to avoid killing the LTZ if I do a mistake. The slight offset it might add will be part of the reference voltage itself.

I consider this board as a black box with no tunable components. It will live in a steel shielded box with thermal insulation inside and outside.

The continuation of my personal volt nut plan is to have a separate buffer board connected at the output of this one, with a more elaborate 10v stage that would include a LT1206 (with a low offset opamp pre-driver). And maybe a kelvin connection with sense wires, so I can work around voltage drops in connected wires. Not sure about that, and thats outside of the scope of this reference module. I can not plan everything at once.

Another board will have the regulator and battery charging stuff.

Attached is the latest schematic and layout.

I'll be getting the initial version of the pcb end of this week, and I'll populate it with jellybeans components and a ltz simulation to validate the circuit.
I'll reserve precision components for this second pcb revision, which I will be able to order soon.

Unimaginable progress was done between the two board revisions, thank you everyone for your feedback.

[iMo]

.. and there is a lot of space at the 7Vref output to put there a capacitor (like 100nF ceramic) in parallel with, say, a 10V TVS diode..
 

[grx]

Indeed!

Ceramic? While I'm at it, it will be film everywhere! TBH I'm feeling lazy to determine which ones of these caps absolutely require film, and which ones could be ceramic.

Is the TVS absolutely required? that would be a protection for the protection. mmmmh okay, I'll add a SMAJ7.5A

edit: given that opamps dont like driving capacitances I wonder if preloading the output with 100nF is wise.

edit: both done, and now I wonder if I should replace the input protection diode by a TVS diode! It would have both roles I guess, simultaneously protecting from reverse polarity and perturbations!

[iMo]

TVS diodes on the input and output, sure:

A TVS diode there might be handy (like say an 18V one with 15V Vcc). Another TVS at the 10V output (like 12-13V one).

Mind there are uni- and bi-directional TVS diodes.

Also do not forget to limit the Vcc current somehow.

If you look carefully at the schematics above (ie the PX one) there is the capacitor at the output of the 2057.
Opamps do not like them unless there is an isolation resistor in between (you have it there).

Ceramics vs. foil - foil one where the capacitor's current leakage is critical or where the precise capacitance or its stability is of importance, otherwise ceramics, imho.. Smaller value ceramic is usually better than a larger value foil..

The first thing people do is they do polarise the Vcc wrongly, use a power source which produces high voltage glitches, short the outputs of the Vref or feed some bad voltage to the output of the Vref.. Not talking ESD..

PS: the max short circuit output current of the 2057 is 30mA typ, so it hopefully survives a short..

[grx]

I need to learn how to read lol

Thanks for the remarks on the capacitors.

Of course we are talking about unipolar TVS. Negative voltages are not legal here

Since a tvs is a zener it is my understanding that it will clamp above its rated reverse voltage, but also if the forward voltage is exceeded. The datasheet does not show great values of forward voltages: 3.5v is not a great protection for reverse voltage. I will need a schottky in parallel for reverse voltage protection.

see page 3 https://www.littelfuse.com/assetdocs/tvs-diodes-smaj-datasheet?assetguid=13c2a823-03b8-4d1f-9ddc-9b44670aed9d

Since TVS do not start clamping before they're well above their rated voltage, I selected a 7.5V one for the reference output, which will start conducting at 8.3V, and will limit the voltage at 12-18V (according to the pulse conditions)

There are several usual misunderstandings about TVS:
* they are made to protect against pulses only, like the static electricity shocks produced by clothes, or fast surges induced by nearby power devices (but they are exceedingly efficient at it, I still have a very stressful EMC testing session very alive in my memory, everything had to be tested again after adding a set of TVS we forgot about, it was night and day on the test results)
* they will not protect from a DC overvoltage for very long (they will burn, max 1W continuous dissipation in the SMA package, and possibly in a safe way, eg short)
* they do not clamp at their rated voltage, far from it.

The TVS on the input should not be tailored to the planned input voltage, because it's a recipe for unnecessary trouble if I change my opinion. The goal is to protect the LT1013, LTC2057 and LTZ1000.
Absolute maximum ratings are
44V for the LT1013
40V for the LTC2057
35V probably for the LTZ1000

The littelfuse datasheet shows that 35V is the maximum clamping voltage (advertised) of the SMAJ22A
Any TVS below this will clamp the voltage within safe limits for these 3 circuits.
However, this is optimistic. The Bourns datasheet (https://www.farnell.com/datasheets/3946914.pdf) indicates that these values are for a calibrated 10/1000us pulse, but the specs for the calibrated 8/20us pulse are worse, 35v limit on this kind of pulse mandates the use of a SMAJ16A
Which is fine, the breakdown voltage for this one is 17.8V so I can safely use 12 or 15V

But there are situations in which a TVS will not save you! In that case other protections are required.

edit: changes done

[grx]

I was having a look at the AD5791 eval board here https://www.analog.com/media/en/technical-documentation/user-guides/eval-ad5791sdz-ug-1152.pdf
And they have a schematic for the ltz1000 reference, which is very classic, but uses the ADA4077-2 op amp: https://www.analog.com/media/en/technical-documentation/data-sheets/ada4077-1_4077-2_4077-4.pdf

With the LT1013, this is one of their rare op amps to have an explicit advertised characteristic for "Long-term offset voltage drift (10,000 hours)" of 0.5 μV typical

I am always wary of typical specs, but that characteristic is quite important for stable voltage references.

For the LT1013, the stability figure is 0.5 uV per month, and one month is 30*24=720 hours, so the ADA4077 should be more stable than the LT1013.

Has anyone already evaluated this component? I do not have the proper hardware to test this.

OK, MisterDiodes had this opinion about this chip:

Even that is not the most perfect amp to use since it is not spec'd for low current noise at 10Hz.  The LT1013 is spec'd for low current AND low voltage noise @ <10Hz, AND comes in a DIP.  The LT1013 IS -the- amp to use with the LTZ because it was designed with that exact LTZ application in mind.  There has never been any true upgrade for it.

So I dont know.

Edit: this amp requires a dual supply, which is annoying for our voltage reference modules. So : Forget it. LT1013 it is.

[EC8010]

Since a TVS is a Zener it is my understanding that it will clamp above its rated reverse voltage, but also if the forward voltage is exceeded.

Since TVS do not start clamping before they're well above their rated voltage...

I used to think that, but I made some measurements using a Keithley 2400 SMU. What I found was that low voltage (10V etc) are a bit soggy, but higher voltage (200V) switch on very sharply indeed with little leakage current below. Also, they are much faster than Zeners. A microcontroller powered by a shunt regulated supply using a 5V1 Zener exploded each time on EMC tests (rail went to 11V), but a TVS solved the problem.

[Kleinstein]

The typical positv LTZ1000 circuit needs an OP-amp that works with a 0.5 V common mode voltage and this essentially a single supply capable amplifier.
The ADA4077 is not single supply and would need an extra negative supply and ideally an interlock for the time when the negative supply is not there.
The LT1013 10 Hz noise specs are only typical - so not much different from the typical curves for many other amplifiers.
The ADA4077 has typical current noise of some 0.2 pA and voltage noise of some 8 nV. This give a resistance for best noise figure of En/In = 40K. This is less than the typical 70 K at the collector, but not very much. The LT1013 is on the other side of optimal.

[iMo]

.. a Saturday's afternoon project..
Based on the PX board discussed here in past..
PS: Well, Sunday's as well..
 

[grx]

An interesting design!

I have specifically tried to avoid vias, but that was an arbitrary restriction. I ended up using one for the star ground in the end.

@Kleinstein: Thanks a lot, fully understood.

[iMo]

I've been still elaborating two things - how to organise the output from the buffer - like the sense/force stuff.
Any hint is welcomed.. 

And thinking about adding an option for mounting a small mezzanine board in the 7->10V divider part such I may experiment with the dividers.
So far there is the LT5400 pack (2k/4.5k) for around 10V output, or a simple resistor for the 7V out.
The current through the pack is rather high, like 10V/6.5k (15.3mW), however..

The 78L15 there is rather a placeholder there - not sure it is an optimal solution as it may get hot (like 3V x 30mA loss).
Also it requires a stable voltage at the input as otherwise will change its temperature.
I may not populate it and place there a choke such I get a CLC filter.
 

[grx]

I also thought about that and I am also not sure.

In my board only, without a 10v stage, the circuit is not supposed to source any current, the buffer is just a protection for the ltz1000. So in theory I should not need a sense/force wiring at this point. I am also confused about how to wire that with a follower buffer as mine: positive sense should be the feedback, but where to connect the negative sense??? (see attachements)

However, I was planning to do that for the separate 10v stage, so I can use cables of reasonable length.

In a tentative schematic I did, I was surprised to see that only the feedback resistor divider was connected on the sense path. I found that quite unexpected and decided it was a problem for future me, which is... not yet

[iMo]

My current understanding (well pretty rusty, no warranty of any kind provided) is following:

1. with the simple buffer (none Rdiv) the Sense- is "-Vz Kelvin" - the BLUE wire  (direct wire to the pin 7 of the LTZ)
PS: the pin 7 of the LTZ is still wired to "-Vz Star", of course..
2. with the simple buffer (none Rdiv) the Force- is the "-Vz Star" the big via - the RED wire


3. with the 7->10V buffer (with Rdiv) the Sense- is the cold side of the Rdiv - the BLACK wire
4. with the 7->10V buffer (with Rdiv) the Force- is the "-Vz Star" - the big via - the RED wire

..or something like that.. 

[Jacques]

In my board only, without a 10v stage, the circuit is not supposed to source any current, the buffer is just a protection for the ltz1000. So in theory I should not need a sense/force wiring at this point. I am also confused about how to wire that with a follower buffer as mine: positive sense should be the feedback, but where to connect the negative sense??? (see attachements)

If you want to bring out the force+/sense+ (through binding posts, for example), you might want to add a resistor between sense+ and force+. This way the output op-amp feedback loop is always closed, even when nothing is connected.

[iMo]

1. with the simple buffer (none Rdiv) the Sense- is "-Vz Kelvin" - the BLUE wire  (direct wire to the pin 7 of the LTZ)
PS: the pin 7 of the LTZ is still wired to "-Vz Star", of course..
2. with the simple buffer (none Rdiv) the Force- is the "-Vz Star" the big via - the RED wire

Hmm, not sure the above is correct.. The sense- with the simple buffer is low impedance, so the current will create an additional voltage drop on the pin7-to-"-Vz Star" wire (the current from force+ splits between sense- and force- perhaps 50/50).. And it is not within a control loop in the simple buffer version.. So I would rather not connect sense- to pin7 in that scenario.. Looks like there is none "real" sense- with the simple buffer, imho..
Afaik we discussed the  "current cancellation" and force/sense with fluke stuff (was it in the LTFLU thread?) years back here..

[dietert1]

Yes, having high input impedance of the sense inputs requires buffer, that will in general introduce additional small errors (opamp input offset). So there is a price for supporting higher reference output currents by implementing a four wire scheme.

Regards, Dieter

[Andreas]

For sensing GND you will need one additional buffer
and of course a negative supply voltage like here:

https://xdevs.com/article/792x/

with best regards

Andreas

[iMo]

Yes, having high input impedance of the sense inputs requires buffer, that will in general introduce additional small errors (opamp input offset). So there is a price for supporting higher reference output currents by implementing a four wire scheme.

Regards, Dieter

We messed with that years back here.. I cannot find the posts, grrh (Sept 2019??). I can remember I even made ltspice sims on such sense inputs, etc. The search function here needs an improvement.. 

PS: https://www.eevblog.com/forum/metrology/fluke-732-10v-ref-module-current-cancellation-circuits/msg2689917/#msg2689917

[grx]

iMo: We have the same understanding of where the sense/force line should be in each case.

However, it is starting to become very complex for a "reference board"
The topic of output amplifiers looks as complex as the reference circuits by themselves.

I will not plan for a complete force/sense output on my board, specifically requiring a negative supply and another sense opamp is over my intended purpose.
I will do a separate output amp later.

Also,
I have received my "v1" boards. I will not waste valuable components on them now that the v2 is designed. So I will populate one with random crap from my drawers and build a "ltzish" reference with a zener, a resistor and 2 transistors I have lying around. That will help validate the circuit with zero risk.
And will also be the least stable ltz1000-like voltage reference of the world (reverse goals!)