REF70 new from TI

[KT88]

A lot of analog parts like references but also opamps, LDOs and others use some sort of trimming. Often times the parts feature DACs/digipots together with OTP memory. During testing more or less parameters will be adjusted and then stored in OTP. Especially precision parts like the REF70 won't be viable without that capability...
Lasertrimming is rarely used these days except for thin film resistors.

Cheers

Andreas

[2N3055]

For years digitally trimmed crystal oscillators are available. They are mostly lookup table with interpolation to trim tempco.
Same can be used with voltage reference, provided that tempco shape doesn't change as it ages.....
These trimming techniques usually are very useful to get decent accuracy across large temperature range, more than to provide highest possible sub ppm tempco, lab style..

[dietert1]

Similar methods are applicable for metrology setups as for integrated precision devices. One example is here: https://www.eevblog.com/forum/metrology/esi-resitance-standard-sr1010-and-standard-resistor-sr104/msg2719638/#msg2719638.
I determined a 1.84 ppm/K temperature coefficient using small ambient temperature changes. Later with the same setup i could determine drift/aging and even the quadratic term of TC.

Regards, Dieter

[Kleinstein]

I would expect the digital trimming to be static. That is some memory and DACs / digit pots to adjust the reference with constant values. So no change in the settings depending on temperature or similar. This way no real option to trim for long term drift, but an option to trim the voltage level and TC, possibly even higher order coefficients.

[Andreas]

I would expect the digital trimming to be static.

I would also expect some similar to the "SVR-T" reference by having 2 pots (DACs) one for the offset and the other for some temperature dependant influence from the PTAT voltage.

Otherwise you would need high ADC resolutions.
If having only 1 deg C resolution (~2mV of the PTAT output of the band gap reference) and a uncompensated TC of about 2 ppm/K each compensation step would give a 5 mV step on the output which is much larger than the noise of the 2.5 V reference.

So effectively for a pure digital system you need minimum a 0.1 deg C resolution over a 165 deg C temperature range. And even better if the uncompensated TC is more than 2 ppm/K.

with best regards

Andreas

[exe]

It seems ref70 is now called ref7025 and it is in production. The datasheet was updated. Some noticeable changes:
1. About long-term stability: "Corrected typical shift from 0.021% to 0.009%". That's good.
2. "Changed Figures 7-2 and 7-20, Long-Term Stability (First 1000 Hours), to longer duration (2000 hours)."
3. "Added clarification on how to connect OUTF and OUTS in specific load current condition". OUTS current is... wait for it... 4mA! (this is not part of main specification, they mention this in "10.2.1.2.2 Force and Sense Connection"). Can't make sense of it, this is comparable to maximum output current of 10mA. This also matches quiescent current of the reference. Does it mean they use OUTS as a supply pin? Or this 4mA are in addition to 4mA supply current?

The price from TI website is $12.696 for single qty + VAT. Quite expensive, I was hoping for better pricing. This matches LTC6655 price

PS LTC6655 also has quite big OUT_S current, which is 2mA.

[Kleinstein]

The sense current should be included in the supply current. Still the high sense current can cause a nasty supprise in some cases. An important point may be the question how stable this current is. Just a stable current would only give a little addition to the actual ref. votlage. A change in the current would compete with the change in the sense resistance.

They seem to quite proud of there figure 7-2, it comes up 3 times in the datasheet.

[3roomlab]

this is max6126 for comparison (and about 1/2 the price)



from this article
https://www.maximintegrated.com/en/design/technical-documents/app-notes/7/7009.html

the drift for this is interesting, some are +ve, some are -ve. with the right amt of luck, maybe 2 units will cancel the drift out.

[Kleinstein]

The drift curves look surprisingly good with the max6126, especially for a SO8 part. Looks like they somehow got lucky with the design to get low stress sensitivity.

However the noise is quite a bit higher than for the Ref70 / LTC6655. So it would need some 5 pieces max6126 combined to get a similar noise level.
It is still a good reference, but for different uses.

[exe]

The sense current should be included in the supply current. Still the high sense current can cause a nasty supprise in some cases. An important point may be the question how stable this current is. Just a stable current would only give a little addition to the actual ref. votlage. A change in the current would compete with the change in the sense resistance.

They seem to quite proud of there figure 7-2, it comes up 3 times in the datasheet.

On LTC6655, as I understand, the high current is due to low-value resistor divider (they also say that in the text). So, if we assume that REF70 works the same way, then the load resistance should be stable (but I cannot confirm this without a die shot). And this also means that the rest of the circuit doesn't consume much. This gives me a hope that MAX6226 with 0.38mA supply current is not much worse (but I don't have equipment to to verify this).

this is max6126 for comparison (and about 1/2 the price)

There's also MAX6226 in ceramic packaging and still 1/2 the price . And also MAX6079 with proper kelvin connection, not that high-side OUT_S rubbish .

[Kleinstein]

A badgap reference needs quite some current to get really low noise. So it is really odd the Ref70 uses so much current on the sense pin, unless they actually use it to power the actual reference curuit part.

A resistive divider would need some current, but this should not be so much, more like maybe 0.1 - 0.5 mA or so, but no real need for 2 or even 4 mA.

[Karel]

I recently used an ADR4540 in my DVM project. Looks like the REF70 isn't any better.

[exe]

I recently used an ADR4540 in my DVM project. Looks like the REF70 isn't any better.

REF70 is a bit cheaper than ADR45* in LCC hermetic package: https://nl.mouser.com/Search/Refine?Keyword=ADR4525DEZ . But yeah, seems all parts cost approx the same, which is sad, not price competition

[chuckb]

I could not get a REF70 chip from the normal distributors yet, so I got a REF70 Demo PCB from the factory.

It supplies a solid 2.4998V (K2002) from 6V down to 2.7V input voltage. At 2.6V input (no load) to output voltage dropped 50uV. The supply current was 4.34ma.

I removed R2 on the demo pcb which shorts the output to the sense terminal and installed a 1 ohm resistor. There was 2.8mV drop across the 1.005 ohm resistor (0.5%, 25ppm) for 2.8ma through the sense pin. The installed resistance was verified with a Keithley 580 Micro-ohm meter.

Later checks will include noise and temperature stability.
At the end of the testing the lid will come off and we will gets some die pictures.

[chuckb]

The REF70 datasheet talks about 4ma for the sense pin. That may be for the 5V version of the chip. Just guessing.

[exe]

Thanks, chuckb! The datasheet says the dropout voltage is 250mV (but under load). So, if this 50uV sag happens due to insufficient voltage headroom, or it's caused by line regulation

• . If you have time, can you please check how much output voltage shifts when changing supply voltage from, say, from 2.9V to 12V? (Probably 12V is on the extreme side).
• The datasheet says that at 25C REF70 has 4ppm/V line regulation. For 6=>2.6V change in supply voltage this translates to, if I'm not mistaken, 34uV. It's not far off to what you measured, but it never hurts to verify datasheet figures :).

[chuckb]

Looks like the line regulation starts working well at 3.5V for the 2.5V reference.

I started at 2.6V input voltage and worked up to 15V. Then I returned to 3V input voltage and the output voltage had shifted less than 1 ppm. That's pretty good.

The Keithley 2002 has been turned on for the last few weeks for other testing.

[chuckb]

I did the first test of the REF7025 (2.5V) TempCo today with a 5V supply. From 25C to 61C the output voltage changed exactly -2ppm / deg C. It was right on the spec limit and very linear.

When the temperature was reduced back to 25C the hysteresis was +56ppm.

I will do several more temperature cycles over the next several days to see if the hysteresis reduces.

[Andreas]

When the temperature was reduced back to 25C the hysteresis was +56ppm.

Plastic package or LS8-Package?

with the plastic package it is likely that humidity makes the difference (drying of the package with heat).
If you repeat the test the package is already "dry" so less influence.

If you wait several weeks before repeating the test the package "soaks" humidity from the environment.
The game starts from new.

with best regards

Andreas

[chuckb]

It is the LS8 type ceramic package

[Andreas]

Hello,

for a LS8-package the change of your sample is rather high.
Eventually there is some influence from the (EPOXY) PCB where it is mounted.
(which humidity do you have in your area?)

compared to the LTC6655 its even worse than my first measurement of a LTC6655 in MSOP package (ADC14)
The pure LTC6655BLS8 (without influence from the PCB) I measured with +/-4 ppm hysteresis.
see:

https://www.eevblog.com/forum/metrology/ltc6655b-long-term-drift/msg2355873/#msg2355873

with best regards

Andreas

[exe]

The datasheet suggests that consecutive thermal cycling should have smaller impact on the voltage. Still, 56ppm shift is more than I'd expect.

I'm fighting my greediness and I'm willing to experiment on my max6226 for the sake of fun science. Stay tuned .

[chuckb]

My lab has a nice variable speed HVAC system with some humidity control. It is typically 23C +-0.2C. I have a number of circulation fans running so the temp control has been verified in the different corners of the lab. Most of the time it has a 40% RH +-2%. The last few days have been cool and raining so the humidity was 43%.

I have the REF70 demo PCB in a small heated chamber. For this test I use a SciGene Hybex Incubator ($50 'bay) with some 1/4" steel plates in the bottom to add thermal mass. There is no humidity seal on the chamber. I have not noticed any electrical interference between the heater controller and the DUT on previous sensitive electronics tests.

Attached is the plot from the last several days of testing. The chip temperature was monitored with a small, stick-on, OMEGA DIN 100 Pt Class B RTD. Datasheet attached. The resistance was converted to temperature by a 4-wire connected K2001 set for D100 RTD. A Fluke Model 52 with a thermocouple crosschecked the temperature. The two sensors agreed within 1 deg C.

Setup -
5V applied to the REF7025 demo PCB. The output voltage was monitored with a Keithley 2002.

Procedure -
I would change to a new temperature, wait about 2-3 hours, verify that the system had stable temp and voltage, and take a reading. The DUT was suspended in the air inside the chamber.

The testing started in the lower left corner of the graph at 2.4996V. As I raised the chamber temperature the output voltage decreased at 2ppm / deg C. I stopped testing at 61 deg C. I let it cool over night and the voltage had increased to 2.4997V.

As I increased the temperature in steps on the second day the voltage dropped in a similar manor of about -2ppm /C.

This time I let the PCB soak at 61C for 23 hours. The voltage increased quickly at first then it ended up with a +150uV total shift. The first 100uV of movement happened in 4 hours.
Over the next 8 hours I tested the chip as it cooled. With 1kg of steel in the chamber it does not cool off very quickly. It ended up 200uV above where it started 36 hours earlier.

Later, I will do a few more temperature cycles to 61C then a few going up to just 30C.

The first large hysteresis step is within the REF70 data sheet limits for a 0-70C temperature change. However the factory expects the voltage shift to reduce with cycles. I have not seen that yet. Linear Tech recommends stress relieving cuts in the PCB for their SMD Voltage references. That may help.