Ultra Precision Reference LTZ1000

[retrolefty]

I found 0.3-0.8ppm/°C Precision zener-based references today. Maybe of interest for someone?

Hello Branadic - for some reason your link is broken ...

BR
gazelle

Just remove the trailing / on the URL

[branadic]

I found 0.3-0.8ppm/°C Precision zener-based references today. Maybe of interest for someone?

Hello Branadic - for some reason your link is broken ...

BR
gazelle

It should work now.

[paulie]

Recently playing around with back to back low cost glass zeners observed virtually zero tc at certain current and wondered about fitness for precision voltage references. How they compare (noise, hysteresis, etc) with the "buried" type like found in LTZ1000. Discussing with a client who is a semi product engineer he stated the difference was minor compared to other issues but wasn't able to direct me to any documentation on comparisons. No luck with net search either.

This thread seems to have top experts in this area so I was wondering if anybody had experience with discrete glass zeners vs buried type or know of any links to comparisons.

[paulie]

Thank you for that summary. I've read this thread twice and your post saved me from going through again. In addition to making those discussions clearer did bring up a couple points I don't recall mentioned. Greatly appreciated.

In my case, not owning real precision gear, a $50 dollar part like LTZ1000 not in consideration. Because  of a self imposed challenge even a $10 LM399 out ATM. Mostly this is for low cost DIY projects like the 5 digit $5 Ebay meter and some 24 bit ADC chips in my junk box which need a reference. Right now my best hope is bandgap or homegrown zener circuits and the best resolution possible. Long term stability is not really a big concern right now either so just trying to stay in the "pennies not dollars" range for parts. No plans to build 8 1/2 digit voltmeter but hopefully at least a couple orders of magnitude less.

Your description certainly gave more info/details and bigger picture than Wikipedia which was my best source so far. Thanks again.

[Vgkid]

I would say that the most stable discrete zener reference I have heard of is Microsemi's USR series.
http://www.microsemi.com/document-portal/doc_view/10943-sa6-35-pdf
top spec 934

[janaf]

I didn't know about the URS931! The 1N829 is also rated 5ppm/C, I have not seen any drift data. I think additionally compensated and presaged 1N829 where used in the Solartron 8.5digit DMMs.

URS931 are obsolete? while you can still buy 1N829.

http://www.microsemi.com/document-portal/doc_download/125462-lds-0220

[Vgkid]

The 1n939 was used in the Solartron 7055
USR932 was used in the 7065, crudely ovened.
ovenized selected USR932 was used in the 7075.
All Solartron models btw.

[paulie]

Mmmm... looks like I came to the right place for information on this.

0.0005%/C is way more than needed for my purpose and  temperature range -55deg to 100deg means the "S curve" is nowhere near as short as hinted earlier. At least in the ball park with initial tests (0.1mv) on some homegrown zener parts I've been tracking for last couple weeks.

Prices for USR, 1n829, and 1n939 are more than an LM399 so just as out of reach as the LTX1000 for my budget so it looks like the DIY parts at 2 cents each are for me. I hope performance of cheapies is close enough. The circuit looks a lot more manageable to compared to LTZ too which is another plus. Even simpler because I don't need negative ref. I got OP07s but maybe not even any op amp will be needed if ADC ref input impedance is high and I use software calibration constants.

[splin]

Recently playing around with back to back low cost glass zeners observed virtually zero tc at certain current and wondered about fitness for precision voltage references. How they compare (noise, hysteresis, etc) with the "buried" type like found in LTZ1000. Discussing with a client who is a semi product engineer he stated the difference was minor compared to other issues but wasn't able to direct me to any documentation on comparisons. No luck with net search either.

This thread seems to have top experts in this area so I was wondering if anybody had experience with discrete glass zeners vs buried type or know of any links to comparisons.

You'll find it hard to beat WarrenS's experience from his volt-nuts postings. See:

https://www.febo.com/pipermail/volt-nuts/2013-January/002362.html

https://www.febo.com/pipermail/volt-nuts/2013-January/002372.html

So it's quite straightforward to make a good reference so long as you're not in a hurry:

https://www.febo.com/pipermail/volt-nuts/2011-September/001046.html

Please note that not all 1N825, 6.2 volt reference diodes are created equal.
This one is cherry picked from my special supply of ones which I've:

1) Aged at 50 MA for a year,
2) Run in at 7.5 MA for another year,
3) Then No power Aged for another 5 years,
4) Tested for stable and repeatability for better than 1PPM,
5) Tested for low noise for better than 0.1PPM,
6) Has a room temperature 'Zero TC' under of 0.1PPM
7) and the Zero TC current is a under 5 ma.
 

You didn't need it before 2022 did you? 

And if you wondered how he could afford to test such large batches of 1n829s which cost $10 or so apiece:

If you're interested, I'll send you the zener, because I got a lot of them
on EBay a long time ago when they where only 1 cent each.

Splin

[paulie]

Wow, just when I think I got all the leads I need Splin comes along with those volt-nuts links. I've not really checked out that site because not being a real volt-nut didn't think there was benefit and also I thought it was German only. How wrong I was.

So 1 cent to 10 dollars. Makes me wonder just what the real difference is between 1n829 and the ones that are still cheap. Maybe like Iphone it's just a matter of fashion. People talk, price skyrockets.

Yes the main difference is the zero TC current, with some parts there is no zero TC current.

Using my patented "hot finger" method I was able to check 200 pcs (thats right... whole $2 worth) in little over an hour but didn't find any that failed to zero. So looks like I'll have to do testing to see just how the S curve varies from one to the other. That will probably take more than an hour.

Besides trying to match the LTZ1000 7v I also tested many with much lower voltage. All the way down to 2v thinking it would be nice to get away with 5v only instead of needing 15v plus multiple regulators. They all had a 0 TC but the current was surprisingly high. Up to 10x more than the few milliamps mentioned previously. That needs looking into.

[paulie]

YES! That is exactly the circuit I've been looking for. Except for the $20 op amp we do seem to be on the same poor man's wavelength now. OP07 will have to do instead of chopper. At 10 cents ea I did pick up a bunch of LT1034 but to be honest the penny zeners thrill me more so probably try it with those first.

I do have experience with TC ovens. See my "Worlds Smallest and Cheapest" below. I found bipolar gets ±2deg but FET like 2N7000 as you might make out in the closeup gets ±0.5deg. Super thin 40awg wire and the 1" foam block not only help with that but keeps power consumption down to about 4-5ma. That an order of magnitude better than equivalent commercial units. The 2 wire device is NTC thermistor for independent verification.

ps. I have even smaller using tiny (aka expensive) NTC with 2N7002 SOT-23 but no photos. I also have units big enough to keep a loaf of bread at "just the right temperature". Of course current consumption a lot more than few milliamps.

[paulie]

The LTC2057 is around US$2.50 in small quantities

Strangely almost 10x that on Ebay:

http://www.ebay.com/sch/i.html?_from=R40&_sacat=0&_nkw=LTC2057&_sop=15

I don't know where you are getting LT1034 at 10-cents, can you share where you bought those?  I can only find them at US$2.50

More accurately they were closer to 20 cents shipped (5/$2.99 30% sale) on Aliexpress. Inconvenient for me to dig up the order link from my current location and searching is always a tedious process there but here's a USA Ebay seller I paid 40 cents to get them quicker. Too bad because they ended up on a shelf. 4040a were cheaper and better for my application.

http://www.ebay.com/itm/381138383313?_trksid=p2055119.m1438.l2649&ssPageName=STRK%3AMEBIDX%3AIT

At this time and probably never any intention to match LTZ performance but I would like to improve on the Ebay 5 digit meter hence current interest in better references. I will probably not be getting involved with DIY wirewound or $40 foil resistors either. However I do have rather large selection and quantities of 10ppm Vishay 0805 left over from a day job client. I have experimented with "adjusting" values using dremel disk and fine tune with nail file. The only issue seems to be sealing them for long term stabilty. It would be wonderful if they had Teflon paint or 2 part glue. For now I'm using epoxy and polyurethane hoping to resist oxygen and water.

ps. BTW I don't see any need in my applications for R3 at all. I need stable but not exact voltage as I plan to calibrate the meter in EEPROM.

[paulie]

Always go with through-hole due to board stress [PC material TCE, and humidity] causes changes in surface mount resistors-- they act like strain gauges.

At 100x the cost it's unlikely though-hole are in the cards for me ATM. For now the Vishay 10ppm will have to do. Specially considering they cost me zilch. However IMO even new 25ppm SMD at ten cents or so are a good deal for those on a budget.

BTW note that in terms of mechanical stress SMD can become through-hole by tacking on a couple lengths of #30 ww wire. Using appropriate heat sinks of course. I did try it with a few of of the 10ppm 0805 then realized for my current project such extreme measures not required.

I think you're advice is sound though and really appreciate you taking time. Specially like the idea about double wall shrink. Already ordered couple meters.

[paulie]

With careful adjustment of R3, you can get the TempCo to be very close to zero,

DM, I wired up your circuit and it might be working but one thing bothers me. R3 seems to have no effect on TC. It appears to be voltage trim which I have no need for since my application requires stable but not exact voltage. Did you mean adjust R1 or R2 instead? Can you explain the circuit operation a little more?

[paulie]

I'm not in the shop right now but IIRC adjusting R3 varied output by somewhat less than a volt so that's what should be across R1. I'm using back to back 6v2 zeners for D1 and they are thermally attached to the PNP so I was hoping this would work. The only other thing different is OP07 instead of the 2057 which I also thought was ok. Resistors are 1% with high ppm/C. I wasn't sure if by critical you meant value or TC.

ps. I notice it's a one transistor circuit like LTZ1000 but with PNP instead of NPN so not sure if it's really similar. And D2 is for startup?

Can you describe a brief theory of operation?

[paulie]

Maybe not hours but better part of one so far and I do like soldering.

OK, so apparently there is no attempt to adjust the zener itself for zero TC but just get close and compensate for the difference. When you said adjust R3 I thought only TC changes but didn't realize output does too so I'll take another look with that in mind. I'm not able to change the op amp but will try with a real 1034 when I get access early in the week. At this time I don't need a perfect  or long term reference so once I verify operation will continue to experiment with modifications.

ps. So basically this works by temperature COMPENSATION only like the D105 while the published LTZ circuit has the advantage of using HEATING and compensation. So things should get even better adding a temperature controlled oven to the mix. I think I'm getting the picture. Thanks for the jump start.

[paulie]

So taking another look with your comments in mind I see the circuit is actually working with penny zeners and 2n3906. The changing output voltage was confusing and R3 is uncomfortably close to the end with my components so not perfect. Also it is considerably more difficult to adjust than the simple zeners with resistor I've been playing with. Hot finger method does not work but as you hinted holding the whole thing over a heater and then fanning the whole thing does.

R1 is *way* too large in that circuit

Yours does not have an anode resistor. What effect does that have?  Will adding one alllow me to re-center R3?

If you tweak the TC to near zero, then add an oven, it will be very stable.

How to adjust zener current? Your 110R?

Or is it not as simple as that and they interact in a more complicated way? So far haven't played with those because they are soldered in. I've still only been at this minutes not "hours". It would be interesting to see how this compares with basic zero TC zener with resistor and the bandgaps. I know there are noise and other issues but my goal is only to get 5-6 digit reference not 8 1/2 and I want the easiest lowest cost solution.

[janaf]

I have been doing measurements on the LTZ1000ACH circuit to see how variations of the resistors changes the output voltage of the whole circuit. I have simply made small changes to the nominal resistor values and measured the change in output voltage.

It turns out that the output voltage changes are very predictable by size and sign.

This table summarizes the results and compare to those given in the datasheet.

Resistor	Datasheet	Measurements
R1	1/100	-1/770
R2	1/330	-1/250
R3	1/500	-1/1400
R4/R5	1/100	1/100

The biggest difference is that the R1 and R3 values do not seem as sensitive as the datasheet tells, while R2 is slightly more sensitive. And that the values are negative, ie increasing resistance lowers output voltage. Except for R4/R5 where an increased ratio increases output voltage.

I have posted these results earlier but think they are worth repeating. I have also corrected the sign, some where wrong earlier.

Actually the summary is based on hundreds of measurements, the matrix is as follows

Res	Range	dr
R1	80R-1K	1R2, 10R
R2&R3	47K-120K	100R,1K
R4&R5	1:12-1:15	500R

The results table is a summary only. The resuls are fairly constant over a wide range. No sweet spots.

For R1. Higher R1, i.e. lower zener current, lowers the sensitivity somewhat, (but at the price of higher noise). It can be noted that the zener current does not change the circuit total power as the sum of zener and heater is constant for a given temperature.

R2 tends to get better attenuated / less sensitive at higher values, it decreases by about 30% going from 47K to 120K.

The value of R3 does not change it's sensitivity / attenuation in the measured range.

For the most important value, R4/R5, a higher temperature i.e. higher R4/R5 ratio lowers the sensitivity of those resistors values somewhat, from 1/80 at 1:12, to 1:110 at 1:15 (but of course with the price of increased drift).

As a summary, The measured changes in sensitivity/attenuation are fairly marginal, not big enough to cause alarm, no sweet spots.

On the contrary, if the resistors are stable enough, without hesitation I'd use any of the values below, and expect at most, marginal changes in performance:

- R1: 80R to 150R
- R2 and R3: 50K to 100K
- R4: 10K to 30K with matching R5

[paulie]

If of interest I can post photos, data and diagrams...

I have been browsing this thread from the start but only recently become interested in specific LTZ circuits. I've collected dozens of diagrams but sure which are current or relevant. Could you please re-post the schematic for your data?

[janaf]

Hi,

Plain vanilla circuit out of the datasheet except:

- I used two LT2057 opamps
- No diode out of the opamp that supplies the zener

[janaf]

A simple test anyone with a LTZ1000 circuit can do:

- Put a 10K resistor in parallel with the 120R. The resistance will then go down from 120 to 118R58, I.e. decrease by 1R42 which is 1.18%. The output voltage should then go UP by approximately 0.0118 * 1/770 * 7V = 0.108mV

If it does, it matches my measurements 

[paulie]

I'm not sure if 6 digits over months but not years is considered "stellar". This is what I'm trying to figure out. I'm definitely not interested in 8 1/2 digits that are dead on 20 yrs after last cal.

As far as resistors there are 10ppm Vishay SMD here with some selection and ability to trim. That's what I have to work with for the moment. No wirewound or foil for me yet.

Would 5000ppm/year time drift be 'OK' with you?

LOL. Well no because early indications are that the simple zener/resistor and bandgaps are showing one or two orders of magnitude better than that. I've been tracking a dozen or so of these cheap references for nearly 4 weeks now and they do appear time stable enough for my purpose. Mostly I want to know if I can send a reference out to have it checked and see if it still measures similar on my equipment as it did before I sent it. So 6 digits useable over a month or two would be great for now. I wouldn't mind having to re-calibrate a meter against my AD584 reference couple times a year.

The other question is short term temperature stability and noise. I'm looking at 3 basic categories each with cost/performance tradeoffs:

1. Simple zener with resistor trimmed for zero TC.
2. Bandgap parts obviously more convenient with no trim but noise and maybe long term issues.
3. More complicated transistor/opamp circuits which are not as easy to adjust but have the edge in performance.

All of these would benefit from an oven but that is a separate project for after I get a handle on the references.

I'm looking for the lowest cost version of each and potential for use with 6 digit maybe stable for months but not years. There is indication the glass zeners are useable as indicated by reviews of Solartron DMMs. So right now I'm trying to correlate similarities/differences of the one transistor circuits and greatly appreciate your input so far helping me understand them. Thanks.

[paulie]

How are you doing that?

Well I think by now you know I am not a metrology lab or anything close to a volt-nut. Penny-nut is more like it since cost it the top priority even if all else fails. This is direct opposite of everyone else in this thread and most on the site for that matter who behave like cost is no object. Or at least in their advice to others.

To answer your question my crude tests use the 5 digit Ebay meter ( https://www.eevblog.com/forum/projects/$5-voltmeter-with-5-digit-%280-1mv%29-resolution/ ). It is known to have reliable 0.1mv resolution and excellent accuracy from testing against 6 1/2 and 8 1/2  digit lab grade instruments. It uses a $2 Microchip ADC with built-in reference and little else. In my case I've attached a 16 channel 4051 based multiiplexer to read multiple channels and OP07 with gain of 10 and offset to boost effective resolution to 6 digits in the range of interest (2.5v).

From my first post here it was obvious you were very "diligent" and this definitely works in my favor. If I ignore 90% of your advice that I might consider overly cautious for my application that still puts me way ahead. Fortunately nearly everything you tell me rings true and I'm grateful for your help. Any differences we have are mostly due to disparate goals. You seek perfectionism and I'm after economy and minimalism.

But I do think your comments about 4 digit limit may be overly pessimistic since many existing devices, not just the $5 panel meter, but many expensive commercial instruments like Solartron, do not depend on LTZ1000 or LM399 or anything similar. So time will tell just how far off my pipe dreams are. For now probably better not to clutter up this thread because you have satisfied my needs for information ATM and I'm grateful. Thanks again.

[janaf]

Paulie,

There is a seller on Ebay selling LM399 at $4.60. I bought handful and they where good. No expensive external components needed, just a good stable regulated voltage supply and some resistors. The LM399 do not suffer much hysteresis so you can shut it off when not in use.

If that's too much, i can send you a dozen LM329CZ to tinker with. Just PM. When done with those, you may want to go to the LTZ1000. Or not.

[paulie]

Actually the cost of LM399 on Ebay has increased big time recently. Can't even even get one for 2x that now. I think the volt-nut craze caught on and there is a big rush for these parts.

When I first learned of LM399 and found the zener could be purchased separately I did a search but found they were even more expensive than the LM399 at that time. Cost of the LM329 has come down but if you can pop a couple in an envelop and send letter rate I might like to give one or two of those a try. Doubt it will be a "gateway drug" for LTZ100 but I'm very interested in how they really compare with the surface diodes. I'll have a chance to check out the 1034 too sometime next couple days which will be interesting.

DM, I'd have to agree on both counts. Today LTZ, being king of the references, probably would be most cost effective for a commercial 6 1/2 digit meter and definitely for 8 1/2. Even at thousands of dollars a unit those guy can't afford to fart around with selecting and fine tuning. I also know more than one person on this site has swapped out for an LTZ so right again.

In my "special" case not so anxious to pay 10x more for a reference than I did for the meter. I would like to see if I can match or beat the cost for a DIY version and if possible add another order of magnitude resolution.