T.C. + Hysteresis measurements on brand new LT1027DCLS8-5 voltage reference

[Andreas]

Hello,

I have long waiting for the "ideal" 5V voltage reference for my 24 bit ADCs.
- low T.C. < 1 ppm/K around 25 deg
- low hysteresis < 1 ppm for +/-15 degC
- power supply voltage below 12V (from around 10V on) for battery supplied devices.

After having measured  the LT1027CCN8-5 (DIP8 package)
against other references I always wanted to have this chip in a hermetically package.
The plastic package showed near ideal behaviour with respect to T.C. + hysteresis.
But humidity sensitivity with 0.5ppm/% rH makes the DIP8 package nearly useless.

Unfortunately the TO-5 package is no longer manufactured, so unobtanium.
But since a few weeks the LS8-package was introduced:
So I ordered some for some measurements...

Measurement setup:

The references are used for the VREF input of a 24 bit ADC which is connected with photocouplers to a PC for evaluation.
Near the GND pin of the LT1027 a NTC senses the temperature of the reference. The reference is wired into a DIP socket.
The input of the ADC (LT2400) is fed from a 7.2V stable voltage reference (LTZ#5)  by a stable 2:1 capacitive divider (LTC1043).
So the ADC is reading around 3600 mV. A positive drift of the VREF will give a negative drift on the ADC readings.
The ADC + 2:1 divider are put in a peltier cooler together with a small heater foil.
The LTZ#5 stable reference is outside the temperature box.

with best regards

Andreas

[Andreas]

First sample #1

Wires soldered at evening before measurement.
Measurement is started next morning without any pre-conditioning.
I generally start from around room temperature to the minimum temperature.
From there to maximum and back to room temperature.
My "lab" heats up in the evening so I have problems reaching low temperatures in the evening.

19.06.2016: (first run)
around 140 uV drift over 25 deg C
giving around 1.6 ppm/K with respect to the 3600 mV.
on the first temperature cycle also a large drift at the beginning of the measurement.
But all in all very low hysteresis below +/-2 uV including measurement noise.

20.06.2016: (2nd run)
again around 140uV drift over 25 deg C -> 1.6ppm/K
the ageing drift against the first run is around 7uV or 1.9ppm/day initial drift after soldering.
Hysteresis is much smaller than on the first run. (when neglecting noise).

So this explains why some manufacturers specify the hysteresis being measured on the 2nd temperature cycle.

with best regards

Andreas

[Andreas]

2nd sample #2

again no pre-conditioning after soldering.
This time I had the time for cleaning the reference after soldering.

21.06.2016 first run
TC is relative good on this device 48uV over 25 deg gives 0.53 ppm/K with box method.
There is also a "sweet spot" near room temperature with zero TC.
Especially when regarding the self heating of the reference of about 2-3 deg C.
Unfortunately the hysteresis is relative large (nearly +/-1 ppm) on the first run.

But this can be also partly due to the initial ageing drift.

22.06.2016 2nd run
noting to complain: the hysteresis has decreased to +/-0.5 ppm.
T.C. als slightly decreased to 40uV over 25 deg C = 0.44 ppm/K.
Initial drift around 1.7 ppm/day after soldering.
So I guess this will even be better after a longer run in phase.

This sample is the candidate that I am looking for my ADCs.

I guess that I will get a relative good yield as long as LT sells only the "D" grade specced with 3 ppm/K.

With best regards

Andreas

[branadic]

Wow, seems to be a really good candidate for the ADC.

I today received two pieces of LT1027DILS8-5. The question is how they will compare to LT1027DCLS8-5.

[branadic]

Sad they used a different pinout for LT1027LS8 compared to LT1236LS8, so a direct replacement is impossible and the ADC pcb needs a redesign. Same is true for LTC6655LS8 Why? I thought Linear Technology had some ingenieurs working for them. Unbelievable.

[Andreas]

Hello,

yes it is sad: every LS8 package with different pin-out.
Even worse if you put the 1027 in a 1236 design the output of the LT1027LS8 will short the battery 

There is only one advantage on the LT1027LS8: it is easier to make a cutout on the PCB since all relevant pins are on the
same side of the reference.

With best regards

Andreas

[branadic]

Andreas, you used the LT1027 only in 3 pin configuration, which means without noise reduction cap and without additional guard trace. So there is space for a bit improvement in noise as well. 

[Andreas]

Hello,

the NR pin reduces mainly wideband noise.
The 1/f noise is not affected much.

Since I use a integrating (sigma delta) converter
I do not worry much about wideband noise.
The LTC2400 has about factor 3 more 1/f noise than the reference.

With best regards

Andreas

[branadic]

Well, when reading this: "...This pin is the most sensitive pin on the device. For maximum protection a guard ring is recommended. The ring should be driven from a resistive divider from VOUT set to 4.4V (the opencircuit voltage on the NR pin)..." I would at least add the guard, for maximum protection on that pin. But how to drive this guard? Possibly 3k + 22k works fine?

[Andreas]

Hello,

Even a higher ohmic divider would do the job. (and save current + self heating of the reference).
2M2 + 300K would give for a 0.1uA (unrealistic high) leakage a shift of 30mV for the guard ring.
If you use a trim pot you could derrive the 0.6V at the upper end by adding a series resistor to the pot.

I for my part will isolate the NR pin by air. (or the ceramic package of the reference on the other side).
Using my in hole dead bug mounting method.

With best regards

Andreas

[Andreas]

Sample #3.

1 day preconditioned (running at room temperature).

T.C. around 0.8 ppm/K with sweet spot at 23 deg C (obviously for countrys north of my lab).
First I thougtht of large hysteresis.
But when looking closer to the diagrams it is in fact a ageing drift of around 2ppm/day of the regression curve.
The ageing drift obviously takes mostly place (as steps) at higher temperatures.
So I think sample #3 needs a relative long time to settle down with ageing drift.

With best regards

Andreas

[Andreas]

Hello,

I also did some 1/f (low frequency) noise (0.1-10Hz) measurements on the samples.
Typical measurements attached below.
I use 100 second measurement time.
So this will give usually somewhat larger peak-peak values than in datasheet (more outliers).
On the other side it is easier to detect popcorn noise with a longer measurement time.
Perhaps a better comparison for noise will be the AC rms value (corresponding to standard deviation).

LT1027DCLS8-5 #01    2.82uVpp or 354 nVrms 0.1-10 Hz noise
LT1027DCLS8-5 #02    2.23uVpp or 294 nVrms
LT1027DCLS8-5 #03    2.21uVpp or 312 nVrms

As comparison some values of other references in hermetically LS8-package:

LTC6655#1 (ADC20)   3.65uVpp or 430 nVrms
LT1236AILS8-5 #06    3.34uVpp or 414 nVrms

So all in all the LT1027 samples show slightly lower 1/f noise than the other parts.
The higer noise on the LT1236 can be explained by lower power consumption.
0.8mA for the LT1236 against 2.2mA for the LT1027 at the same technology (buried zener).

For the LTC6655 I always have larger noise values on my samples than the specced 0.25ppmpp (1.25uVpp) of the datasheet.
But the measurement cirquit in the datasheet/AN124 has the wrong (too low) bandwidth as we discussed already with Lymex and Alex.

https://www.eevblog.com/forum/metrology/diy-low-frenquency-noise-meter/msg939106/#msg939106
So all in all the noise performance / mA (5 mA typical) is not the best for the LTC6655.

With best regards

Andreas

Edit: corrected attachments

[Kleinstein]

For the power consumption, one also needs to take the supply voltage into account. The LTC6655 only needs about half the voltage compared to the buried zener references. Still the noise / power performance is not that good, especially for 5 V.
With a low (e.g. 3.3 V) supply for a 2.5 V reference performance is not that bad.

[Andreas]

Hello,

measured sample #4.
In my eyes it is defective ...

I had 3 days preconditioning at room temperature hoping to get initial drift somewhat away.
But T.C. and Hysteresis are relative high.

- Box TC 200uV over 26 deg C giving 2.1 ppm/K
- Hysteresis +/- 1.5 - 2 ppm
- And even on the large T.C. a visible "noise" in the (integrating) measurements.

The 1/f noise measurements show a catastrophic noise level of 23.1uVpp or 3709nV rms over 100 seconds.
So a factor 10 over the normal level.
The 10 second measurement shows something similar to popcorn noise.


With best regards

Andreas

[zlymex]

Hello Andreas,

Very nice setup and measurement. It seems to me that temperature curves of LT1027 have quadratic shape.
Have you measured the noise of that 3.6V(50% of LTZ1000 by LTC1043)?

[Andreas]

If you want you can email your results to your local LT sales person, and they might send out a free replacement.  I doubt that they would want the old one to be sent back.

This is a great opportunity to mill off the lid of the package, and make some high resolution photos of the die.

Hello,

if I were LT I would want back all defective items for examination especially on "new" parts.

But I will send #4 to Branadic, and hope he can do a photo.

Hello Andreas,

Very nice setup and measurement. It seems to me that temperature curves of LT1027 have quadratic shape.
Have you measured the noise of that 3.6V(50% of LTZ1000 by LTC1043)?

If you look at the old 1027 datasheet there are several local minimums and maximums over the temperature range due to the LT1027 internal temperature compensation.
So the quadratic shape is only true for a small (30 deg C) temperature range.

Up to now I did no noise measurement of the LTC1043 output.
The LTC2400 itself has large noise (1.5uVeff or 10uVpp) and I am usually measuring below that in a 10 second time window.
So I have to use a 1 minute integration time to get stable values below 1uVpp.

But this measurement is a interesting idea.

Edit: but I will have to change probably to a LTC2057 amplifier instead of a LTC1050 with 1.6uVpp noise at gain = 1 on my LTC1043 board.

With best regards

Andreas

[zlymex]

Hello Andreas,

What I'd like to know is whether LTC1043 add significant noise when divide voltage by half. 
The 7.2V output of LTZ1000 has 1.2uVpp noise nominal,  if step down to 3.6V by a pair of resistive divider, it will be around 0.6uVpp. If the performance of LTC1043 is ideal, we can expect similar result.

[Andreas]

Hello,

but with a LTC1043 you will need a buffer amplifier with low leakage on the input.
Otherwise you will see a 400 Hz ripple (switching frequency) caused by the leakage current on the output.
If you use a zero offset (drift) amplifier like LTC1050 as buffer this one will determine the noise level on the output.

I guess I will have to built one divider with a LTC2057 with 0.2uVpp so that the LTZ has the highest noise level.

with best regards

Andreas

[branadic]

But I will send #4 to Branadic, and hope he can do a photo.

I will take a photo of the die, that's for sure  It's up on the postman right now to bring the LT1027 to me.

I guess I will have to built one divider with a LTC2057 with 0.2uVpp so that the LTZ has the highest noise level.

I already designed a 2:1 divider board with LT5400 and LTC2057.

[Kleinstein]

The noise of the reference chips and AZ OPs have a different frequency dependence: at low frequencies, like measured the a LTC2400, the references have mainly 1/f type noise. The AZ OPs and the LTC2400 are mainly white noise and thus the peak-peak value really depends on the upper limit of the frequency band, not so much on the lower limit.

So one can not directly compare those numbers - it depends on the amount of averaging used.  With the more realistic 0.01 - 1 Hz band, the noise of the AZ OPs will be about 1/3 of the 0.1-10 Hz band usually used in the DS. For the references the 0.01-1 Hz and 0.1-10 Hz band have comparable noise, thus the lower bound is also important.

[Andreas]

I already designed a 2:1 divider board with LT5400 and LTC2057.

Sounds like "comparison"
What will the integrating behaviour of the capacitive divider (LT1043)
(ok it is "out of band" for the 0.1 - 10 Hz noise with 400 Hz chopping)
be worth against the pure Johnson noise of a LT5400 resistive divider.

With best regards

Andreas

[branadic]

So here we are. This is a very first impression of the LT1027 die.
It's pretty hugh for the former microscope with x5, so I had to use another one with x3.

[Edwin G. Pettis]

Andreas,

The LT5400 does indeed have 1/f noise like all other resistors with the exception of precision wire wounds, they are the only resistor type that do not have 1/f noise.

The LT5400 data sheet does not specify noise specifics, only an excess noise of <-55dB as a total noise within bandwidth, They do not specifically state that there is no 1/f noise in these resistors.

[Gyro]

So here we are. This is a very first impression of the LT1027 die.
It's pretty hugh for the former microscope with x5, so I had to use another one with x3.

An interesting die photo, thanks.

I see there are quite a number of trimming fuses and networks associated with the test pads on the left hand side. I assume the burried zener is the round structure at the top edge next to the top left bond wire? Although there seems to be another half way down on the left hand side.

It would be interesting to see what an AD587 (or 586) looks like in comparison.

[branadic]

I will do some further pictures within the next days. Also a comparision of LT1236 und LT1027 in matter of die size.

It would be interesting to see what an AD587 (or 586) looks like in comparison.

No problem, if you want to send me an exemplar preferable already opened I can take some pictures of it