Ultra Precision Reference LTZ1000

[TiN]

Or my ftp (link in sub), I'll repost

[Do Ma]

Here more pictures

[TiN]

Try to resize photos less, details are barely visible.
Regarding the schematics, what is the purpose of R12 (3.3K to ground, parallel to 1K of oven setpoint network)?.
Resistor network at buffer amplifier can be a problem as well. What was the idea to have it there?

[Echo88]

For isolated supply simple DC-DC-converters + LT3042 should do the trick: eliminating ground loops and providing low noise supplies for the modules.

[Do Ma]

With the potentionmeter at the buffer we wanted to trim the single voltages at the same level. It only was a test and yes it is possible to trim them at for example 7.5V
I will change the pcb layout and check grounding again. Also I'll check isolated supply for the modules and test it. I will give feedback again.

Thanks a lot!

[Kleinstein]

There is no need to trim / scale the reference modules to about the same voltage. With larger resistors for averaging there is not that much current flow, even if they are some 100 mV apart. It is only that the resistors stability gets more important when the differences get larger. However an extra scaling stage would likely add the same error of even more. Just for averaging there is no need to have an extra buffer - the output of the standard LTZ1000 circuit is already capable of driving several mA - so well enough  for something like a 1-10 K resistor.

Combining several LTC2057 OPs in one circuit could also cause trouble with one OP demodulating the noise produced by other OPs. The more reliable way is to stay with the LT1013, even in the SO8 version if needed.

The LTC2057 might be suitable to buffer the voltages behind the combining resistors.
Using combining resistors (and another buffer) at the ground side too could be an alternative to have isolated supplies. Buffering the negative side might need a kind of negative supply, though usually not -15 V, - 5 V or -3 V would be enough.

[chuckb]

In the corner of the LTZ1000 chip you will find the initials
MG
CN

CN is for Carl Nelson. He worked with Dobkin on the LM399 and he designed the LTZ1000.

He talks about some of that work here -
http://www.electronicdesign.com/analog/interview-analog-guru-carl-nelson

Two interesting sections -
"Ten years later (after the LM399 - cb), we began thinking about how to take the next step toward “God’s own reference.”  We made the decision to design a heated chip that had only the basic reference parts on it, comprising a Zener diode, a temperature-compensating transistor, a temperature-sensing transistor, and a heater. That way, we could optimize the layout to be sure all the devices were at exactly the same temperature, even with the heater running at full power. The user could crank up the Zener diode current to get extremely low noise. That became the LTZ1000, and it was the only game in town for very high-end DVMs (digital voltmeters) and NIST (National Institute of Standards and Technology) transfer standards. I have heard scientists used it to calibrate the Large Hadron Collider. Yet even that part has drift and noise residues. The goal is to make a part so good that you can’t measure it, so the work goes on."

"One place that Spice breaks down is thermal interactions on the chip. I have designed quite a few high-current regulators, so I tend to think of any current below 100 mA as being thermally benign. Yet one of my recent designs had a thermal runaway problem in a 20-mA bias loop because it was all jammed into a tiny space. Spice never saw it coming.
I am now developing a method to do thermal modeling in Spice that takes into account each transistor’s temperature. It’s rudimentary, but it works, and lets you see die temperature unfold in front of you as time proceeds. I plan to include it in a macromodel of the LTZ1000 heated voltage reference, so customers can do thermal modeling as well as electrical. The Spice folks are choking on their lunch when they see this, but I think it’s the wave of the future."

Another comment by Mr Nelson -
http://www.electronicdesign.com/analog/what-s-all-ltz1000-stuff-anyway

"Brisebois explains, “The simple loop circuits will work at low noise, but not low tempco. By far the most popular circuit is the full ‘7V Positive Reference Circuit’ (Fig 3). One useful thing to understand (it took me 15 years) about the LTZ1000 is that pin 4 is the p-substrate. All the other pins operate above that (which is why you want the heater a diode-drop up) ... except for the mysterious case of pins 6 and 7, which operate 0.6V below the substrate.” Carl Nelson, the LTZ1000 designer, notes he did this to allow for a true subsurface Kelvin connection to the bottom of the Zener, reducing its resistance."

[TiN]

Very thank you, tear to the eye are that interview and these people contributions.   

[MK]

Does anyone know when that spice model for the LTZ1000 might become available? is there a partial model anywhere yet?

I have used a 2N5089 in my simple model, possibly still too large an assumed size, and hfe much too high, anyone think of a smaller, lower hfe transistor I could use instead?

[Vtile]

Hello it is me again, asking silly questions. Why all the LTZ1000s are seated down to PCB and not top of nice teflon(ish) standoff like used in many some hi-performance OPs with TO-99 cans.

Edit. PS. Another silly question (I have plenty of them) why LTZ1000s are seated to PCB at all and not thermally sealed container with only as non-thermally-conducting leads as possible.

[MK]

To keep the pads and leads at the same temperature you need some insulation on the solder side of the PCB as well as some insulation to keep out draughts on the top. Too much insulation and the control loop cannot control the die temperature though.

[Vtile]

Hmm.... Too much insulation.. So what is the optimal amount of insulation? ... I suppose the amount that correspond to positive slope of the optimal heating power. Too weird, time to get some sleep.  I also have silly question as how there is no convection introduced drift inside the enormous free space inside of the can, are those vacuum?

[kj7e]

Hmm.... Too much insulation.. So what is the optimal amount of insulation? ... I suppose the amount that correspond to positive slope of the optimal heating power. Too weird, time to get some sleep.  I also have silly question as how there is no convection introduced drift inside the enormous free space inside of the can, are those vacuum?

They are filled with dried air.  There is some thought the chip can maintain better thermal stability if allowed to draw more heater power. However, air currents around the leads, top and bottom cause thermal fluctuations.  Maybe some open cell foam to dampen air currents yet allow it to "breath" is a good compromise.

I did a video showing just a bit of air around the legs will affect the output;

[Vtile]

Interesting so there is better thermal stability when the can is upside down? As there is microclimate inside the can and when the system is not in equilibrium (oven driver is heating or cooling to compensate) there should be convection as there is molecules.

[3roomlab]

hmmm after re-reading the vicinity of this post

The most thorough analysis of LTZ1000: http://bbs.38hot.net/forum.php?mod=viewthread&tid=88278 (use google translate)

and also parts of the 38hot.net, it appears that many of the DIY or prototype pics i saw in many threads do not use the "pre-compensation" resistor which is indicated for unstabilized application (extra resistor above pin3), but since it does reduce TC, why dont heated builds include them too?

[martinr33]

That's an interesting question. As I read it, the goal is to select that resistor such that the zener diode has a zero temperature coefficient. Of course, that is a bit redundant with the heated substrate. However  I wonder if that would eliminate some  of the variability associated with things like orientation, or temperature differentials associated with the legs. 

[MK]

As the legs are KOVAR alloy to seal properly to the glass inserts around each pin you are stuck with the 8 thermocouples to contend with, the bias 400K resistor and or the series compensation resistor "above" the zener can compensate for temperature variations of the die, but cannot compensate any thermocouple effects.

[Kleinstein]

hmmm after re-reading the vicinity of this post

The most thorough analysis of LTZ1000: http://bbs.38hot.net/forum.php?mod=viewthread&tid=88278 (use google translate)

and also parts of the 38hot.net, it appears that many of the DIY or prototype pics i saw in many threads do not use the "pre-compensation" resistor which is indicated for unstabilized application (extra resistor above pin3), but since it does reduce TC, why dont heated builds include them too?

The resistor above to adjust the TC to zero has the big disadvantage that it effectively increases the zener resistance and this way makes the circuit more sensitive to resistor changes, e.g. with the 120 Ohms, or the 70 K from the voltage loop. So it is usually making things worse in a heated application. The LTZ goes the way of having a reasonably low unheated TC (e.g. 50 ppm/K), but having a super stable temperature (like 0.001 K).

Too much insulation can reduce the effectivity of the temperature control, as the usual circuit is made for higher power and does not correct for the square voltage - power relation at the heater. So loop gain (and thus the ability control the temperature) goes down as the heater power goes down. If used with very good thermal insulation, a modified temperature control part might be helpful to keep the temperature as stable as with higher power. The heater loop still running with the nonlinear heater indicates that the temperature control loop is still not that much optimized.

Keeping the leads short helps to reduce temperature dependent resistance of the wires. For keeping the solder joints at the same temperature and thus avoiding thermal EMF longer leads would be more of an advantage if isolation against air currents is used. So the best length for the wires is not an easy question - it more like experience showed that shorter leads usually work better.

[martinr33]

That raises two questions.

1) Would a small (3mm) insulating standoff be a good idea? This standoff would eliminate any cooling due to air flow. The longer leads would, as noted, reduce heat losses through the legs.

2) Was balancing the zener current part of the original design idea, eliminated when LT found that it did not help once the diode temperature was stabilized? Maybe they thought   that an oven was not necessary in some applications. The Geller reference is an unheated design that is set to the zero on the temperature curve.

[MisterDiodes]

That raises two questions.

1) Would a small (3mm) insulating standoff be a good idea? This standoff would eliminate any cooling due to air flow. The longer leads would, as noted, reduce heat losses through the legs.

2) Was balancing the zener current part of the original design idea, eliminated when LT found that it did not help once the diode temperature was stabilized? Maybe they thought   that an oven was not necessary in some applications. The Geller reference is an unheated design that is set to the zero on the temperature curve.

The LTZ is a power in / power out device first.  A stable voltage output is the by-product.  You must have -some- thermal heat flow out of the LTZ otherwise the heater won't be able to regulate.  So you protect the LTZ from air drafts, but not necessarily insulate it.  You want some heat flow down those leads, normally.  By the same token you don't normally ovenize the LTZ itself unless you're heating the die externally - but really what we've seen in every case we tried: the on-die heater does a (much) better job.

If you over insulate the LTZ you get a wobbly output and see the heater go out of regulation - you don't ever want to see that heater current dropping low or close to zero.  The voodoo slots you see on some designs sometimes allow an un-balanced air draft to develop and you never want that either.  We don't use slots or crop circles in our designs; we've never had the need for that.  Some specific application might want that if there is an external heat source issue that needs to be mitigated, but it won't be required from the LTZ itself.

Not all applications require the manufacturing expense or extra power required for a heater.  Not a lot of real applications even need an LTZ.  If you're needing an LTZ in some design, by definition everything downstream is probably going to be very, very expensive anyway.  Just changing the LTZ Vref voltage to another absolute value in a very stable, low noise way is usually much harder to achieve than the basic LTZ circuit.

[Kennylxix]

Hi. I was wondering if anyone has any spare pcbs for the ltz1000 reference board?  I've got the ltz1000 on the way but can't afford to have 3 boards made (I take care of my disabled mother full time so money is perpetually tight)
I'm happy to pay a reasonable amount plus shipping.
Thanks!

[kj7e]

I have two extra extra of the latest B03 version.  Send me a PM. 

[3roomlab]

this post is stemming from the "new adventures" found in the new thread by manateemafia MX reference

due to a lack of an actual ability to see inside an enclosure, i resort to femm4.2 to try out some disspation simulation. it appears copper clad PCB may leak alot more heat than needed?
in my crude simulation, the LTZ sit inside a 30mm foam ball, which is sliced thru by a PCB. the left n right half are different PCB. right half is full thick 1.6mm PCB 2oz, left is suppose to mimic slotting and 1oz copper. and for the fun of it, the edge is mated to the outer box. a funny carbon box inside a foam skin.

(since femm4.2 does not have all the materials, most of the chosen materials nearly matches PCB, silicon, copper, ABS, etc.)

the femm have a plot function, which seem to describe a steeper temp isolation (lower plot) for the modified left side (pic T-grad.gif) vs the right side (upper plot).

while it is true everything inside a project box will be heat up, however if in the case of a LTZ (non-A) if we assume it can be foam isolated to increase the thermal resistance to 400C/W just like an A version, wont it be a boon?

but as i try to math it out, a KOVAR pin/leg about 2mm long x 21mil, can have its thermal resistance estimated at approx 520C/W. for 8pins = about 65C/W. so for the stated stable thermal resistance inside LTZ pdf (=80C/W), the pins alone is the single most leaky "heat channel" (for all the pins to get 80C/W, they need to be at least 96mil long). this imply the LTZ heater need to work alot more harder pushing out heat to short pins, so in theory if the pin landing no longer need to be heated as vigorously by ramping up thermal resistance, should the zener voltage become more stable?
mass of LTZ1000 ~ 1g
mass of 3cm foam ~ 0.02g
mass of 1.6mm FR4 400mil diam ~ 1.6g ??

due to my above assumptions, since the 8 pins carry a heavy thermal influence, wont this require more attention to remove more thermal mass around the pin landing? so that the LTZ only try to adjust/compensate its own temperature and not the entire PCB?
(rename **.txt to **.FEH to use in femm)

[d-smes]

@3roomlab-  Very interesting model results!  Thanks for sharing.

Do you have individual copper traces connected to each pin in model (if so, what width per pin?) or do all pins connect to solid 1 oz or 2 oz copper plane on both sides of PCB?  Or only one side?   Also, when you say "edge is mated to the outer box", do you mean the top/bottom copper planes are thermally connected to box?  Or just the PCB substrate?

[MK]

2.gif does confuse me, I am not sure what has been plotted there, it looks like the insulation keeps cooler than the surrounding air to me, so it must be plotting something different to what I expect? also any copper layers will not be 100% fill.