Precision voltage references...?

[Boscoe]

Hi all,

Money no object, what’s the best precision voltage reference you can buy easily (from DigiKey perhaps) and solder onto a PCB? This means modules are allowed but separate units are not. All I can find is the VRE range and the AD58x range from Analog. I know there’s the LTZ references how I’d like absolute accuracy as well as low drift. I imagine it’ll be a case of using the LTZ and getting it calibrated are a local lab?

Thanks
George

[SiliconWizard]

If you were considering AD's offer with the AD58x, take a look at this: ADR4525
https://www.analog.com/en/products/adr4525.html
Not too bad.

[tggzzz]

And getting all the other voltnut PCB and case issues sorted, and having it powered up for 6months before getting it calibrated, and... But why not have a look in the metrology sub forum, or search for ltz1000; there's a lot of information available.

Of course your first task is to define what specification you actually need, and not grossly over engineering a circuit.

[Kleinstein]

The voltage needed can also make a difference - a stable 7 V from an LTZ1000 or LM399 are of limited use if stable 2.5 V are needed.
For the absolute accuracy a measurement (after soldering) with a good quality meter can be better than the usually specs.

Even for something simple as a voltage reference there are several parameters to look at: Voltage, TC, long term drift, hysteresis, noise and absolute accuracy. So it depends on the actual requirements.

[David Hess]

The LTZ1000 wins hands down with the LM399 a distant second for temperature coefficient, drift, and noise, but not initial accuracy.

There are several Maxim parts with an initial accuracy of 0.02% or 1 part in 5000 and 1ppm/C which would be suitable for 4 digit applications without calibration and perhaps 5-1/2 digit applications with calibration.  Similar parts from Analog Devices are 0.04% but at less than half the cost and probably represent the sweep spot.

Watch out for low frequency noise.  Except in ratiometric applications, the only way practical way to remove it is to use multiple references in parallel.  This gives an edge to the LTZ1000 and LM399 even though they cost more and are more difficult to use.

[Wolfgang]

Hi all,

Money no object, what’s the best precision voltage reference you can buy easily (from DigiKey perhaps) and solder onto a PCB? This means modules are allowed but separate units are not. All I can find is the VRE range and the AD58x range from Analog. I know there’s the LTZ references how I’d like absolute accuracy as well as low drift. I imagine it’ll be a case of using the LTZ and getting it calibrated are a local lab?

Thanks
George

You could try to buy an old HP3458A and use the LTZ1000 built in there. After 20 years of burn-in and a fresh cal this is probably the most stable part you can get.

[Andreas]

Hello,

as David already said: first LTZ1000 then LM399 for maximum stability.

The Maxim parts that I have (MAX6350, MAX6250A in PDIP8) have low T.C. and low initial tolerance but have large ageing (~10 ppm/year) against other buried zener references.

AD586 and AD587 in CERDIP8 generally behave much better in ageing (1-2 ppm/year after run in phase) but you will have to select them for low T.C. for highest demands. Otherwise you will need a oven for constant temperature or some temperature compensation to get below 1 ppm/K.

The 2 samples of VRE3050AS that I have examined had low T.C. but were "noisy as hell" like a cheap bandgap reference.
So I thougth it was not worth to make further calibration and ageing tests.
Also the influence of the PCB on the "SMD" device was rather large.

Unless you want to measure humidity you should stay away from all plastic packages and SMD devices. This is also true for the new "LS8" ceramic package. As soon as you solder it on a PCB the swelling of the PCB due to humidity can be measured. So the LS8 package is only useful if you solder it to a ceramic substrate with same TCE.

In a early "new product releases" LT specified around 10 ppm for a 25% rH change. When I asked them if the PCB could be the root cause they said "Yes". In the final datasheet this parameter is removed. Now they only give a hint that the PCB can have some influence.

with best regards

Andreas

[iMo]

In LT1027LS8 DS (2016) they spent entire page 10 with the elaboration of humidity influence via PCB stress

[Conrad Hoffman]

Unfortunately, the LT1027 hasn't been made in a metal package for years. That one was really great for the money.

[iMo]

They do LT1027 today in LS8 hermetic package.

[Conrad Hoffman]

Apparently it's still quite subject to PCB stress; I wonder if you could mount it on a few fine wires and eliminate that? At least that was the claim over in the metrology forum. It's way cheaper than the LTZ1000!

[iMo]

Sure, make a slot 7x7mm into the pcb (package is 5x5mm), put the package in and wire it with 3 (or 4) thin wires..
Still it is 5ppm/K max, not comparable with LTZ nor LM399..

Code: [Select]

Hermetic 5mm × 5mm LCC Leadless Chip Carrier Package:
Insensitive to Humidity
Thermal Hysteresis: 8ppm (0°C to 70°C)
Thermal Hysteresis: 12ppm (–40°C to 85°C)
Low Drift: 5ppm/°C Max
High Accuracy: ±0.10% Max
Low Noise: <1ppm Peak-to-Peak (0.1Hz to 10Hz)
Low Long Term Drift
12ppm at 1000Hr
18ppm at 3000Hr

[jbb]

Apparently it's still quite subject to PCB stress; I wonder if you could mount it on a few fine wires and eliminate that? At least that was the claim over in the metrology forum. It's way cheaper than the LTZ1000!

Maybe a flex PCB with some cunning slits.  You could even add an NTC or RTD next to it, and make a little flex lead pod to sit inside a tiny DIY oven...

[Andreas]

In LT1027LS8 DS (2016) they spent entire page 10 with the elaboration of humidity influence via PCB stress

3 slots, 4 slots
The sad truth is that 1 slot is better than 4.

Sure, make a slot 7x7mm into the pcb (package is 5x5mm), put the package in and wire it with 3 (or 4) thin wires..
Still it is 5ppm/K max, not comparable with LTZ nor LM399..

The T.C. isnt that bad (at least near room temperature).
But the hysteresis in the LS8-package is larger than in the DIP8 package.
So they messed up something with the die attach.

I am wondering if the (obsolete) metal can package shows that large hysteresis.

With best regards

Andreas

[iMo]

The LT1027 metal can H package - LT databook '92:
Hysteresis delta 25C  typ 10ppm
Long term stability    typ  20ppm/month

[Conrad Hoffman]

My calculator must be broken because it says I wrote the Mini-Metrology articles 24 years ago. Hard to believe. Anyway, the reference used the LT1027 in the metal can. I just tonight measured the original unit against a good standard. It was originally measured by a local cal lab and tagged at 9.999948 VDC. 24 years later it measures 10.000125 VDC. Maybe the chip drifted. Maybe the inexpensive metal film resistors drifted, but IMO, that's not bad. I have a second one built around the same time and it was tagged at 10 VDC even. It now measures 10.000334 so also not too terrible at 2+ decades. Interesting that both went up, or maybe chance?

[iMo]

All those LT10xx, LT12xx, AD5xx, REFxx were developed for 12-14bit ADC/DAC apps many decades back. Provided you do with 5V your +334uV off is still <1LSB with 14bit.
PS: I would add LM399 to the list too..

[magic]

Unfortunately, the LT1027 hasn't been made in a metal package for years. That one was really great for the money.

Did anyone try mounting the SO8 version inside a TO39 can and soldering it shut?

[Conrad Hoffman]

All those LT10xx, LT12xx, AD5xx, REFxx were developed for 12-14bit ADC/DAC apps many decades back. Provided you do with 5V your +334uV off is still <1LSB with 14bit.
PS: I would add LM399 to the list too..

True, but the numbers are for 10V, as the reference has a 2X buffer built in and nothing in the circuit is adjustable. OTOH, tech doesn't usually stand still for 24 years and today we want references for 18 bit systems or more. Best reference 24 years ago- LTZ1000. Best reference today? LTZ1000! Maybe we've hit the wall.

[iMo]

Ok, thus with 10V you are still <1LSB with 13bit ADC/DAC
Anyhow, all those references I listed above were not designed for something like >5 digits, imho. We use them as they are cheap and easy to apply.

[Conrad Hoffman]

I have an old Analogic/Data Precision AN3200 6-digit calibrator. It uses the LM399 as a reference and I've found it to be extremely stable over time, a couple ppm per year. It probably helps that everything I own is ancient and whatever drift was going to happen, already has.

[Andreas]

Ok, thus with 10V you are still <1LSB with 13bit ADC/DAC
Anyhow, all those references I listed above were not designed for something like >5 digits, imho. We use them as they are cheap and easy to apply.

The LT1027 metal can H package - LT databook '92:
Hysteresis delta 25C  typ 10ppm
Long term stability    typ  20ppm/month

You make them worse as they are when properly handled.
The 13 Bit are without further calibration.

the ageing in the data sheet is more or less for the first 1000 hrs after soldering stress.
After some time most of the buried zener references stabilize to ~1-2 ppm/year.
This is at least valid for the metal can housing and often for the CERDIP package.

Interesting that both went up, or maybe chance?

I think it is more the influence of the resistors so by chance. (I guess that the references were not active 24/7)

My 2 LT1027 (PDIP-based) ADCs (temperature compensated) are ageing in different direction compared to my LTZ#2 which itself drifts down 1-2 ppm/year.
When the reading goes up (thin lines) then the LT1027 goes down in reference voltage.
The 2:1 divider itself is LTC1043 based so is very stable over time.
The humidity (right axis 0.5 = 50%) is a sliding average over the last 5 days which gives the best fit for the LMS calculation of humidity (0.5 ppm/%rH) and ageing (ppb/year against LTZ#2) coefficients.
When correcting by these ccefficients you get the dotted resulting curves.

with best regards

Andreas

[branadic]

The problem of the different CTEs and thus the CTE mismatch, the ceramic package has a CTE of about 7ppm/K (https://global.kyocera.com/prdct/semicon/material/), while the copper and the FR4 board have about 17ppm/K and as a rule of thumbs a CTE mismatch of 10ppm/K will result in problems, might be overcome by using a flexible pcb, which is only 100µm in thickness and behaves not as a solid body compared to the FR4 circuit board or LS8 package. So putting the LS8 package on a flexboard, which itself is soldered to the FR4 board on one side only or on a rigid-flex board with the reference on the flex part might solve the issue using industrial standard technics and without using manual dead bug style technics.

-branadic-

[srb1954]

Is the Analogic/Data Precision AN3200 similar to the Data Precision AN8200 6-digit calibrator?

I acquired a faulty AN8200 really cheap and managed to repair it quite easily but I was wondering how good its accuracy and stability is. It was last cal'd in 2012.

I don't have ready access to any cal lab facilities to check this sort of thing.

Thanks