5V precision referene

[kcs]

I want to build a simple yet relatively precise 5V reference. I have searched DigiKey for suitable IC and found some that could fit my needs. Is ADR4550BRZ a good choice from this list? Seems to be a good value for money, but maybe I am missing something?
Another question: how should I protect this IC from ESD?

[ttp]

Voltagestandard.com (http://www.voltagestandard.com/DMMCheck.html) use TI REF5050, another candidate.

[kcs]

REF5050 is in that list. It costs about the same but has a slightly worse characteristics. I am wondering why they have decided to go with REF5050. It is also interesting how they manage to assure 0.01% accuracy when max accuracy for this IC, based on datasheet, is 0.05%.

I have found that www.voltagestandard.com is offering Vref5-002, a 5.000V, 0.0025% accurate precision voltage reference device. They use VRE305A as a main chip. It looks impressive on datasheet, but it costs $93 on DigiKey. I can get it for under $30 from eBay but I am not sure whether it is some sort of scam or not.

Too many choices... 

[thmjpr]

REF5050 is in that list. It costs about the same but has a slightly worse characteristics. I am wondering why they have decided to go with REF5050. It is also interesting how they manage to assure 0.01% accuracy when max accuracy for this IC, based on datasheet, is 0.05%.

I have found that www.voltagestandard.com is offering Vref5-002, a 5.000V, 0.0025% accurate precision voltage reference device. They use VRE305A as a main chip. It looks impressive on datasheet, but it costs $93 on DigiKey. I can get it for under $30 from eBay but I am not sure whether it is some sort of scam or not.

Too many choices... 

Note the datasheet value is tolerance for initial value when it starts being used. If your design is to drop in the part, then yes you will expect 0.05%. However if you plan to calibrate your circuit the initial value is not important (within reason). It's OK if its out by 10%, because you can use your 3458A (capable of <0.001% accuracy) to measure it and adjust/record appropriately. You have a calibrated 3458A right?

Then the important part after calibration is stability + temperature coefficient. Dave has mentioned this in some videos.

Quick comparison:
ADR4550BRZ 0.02%, 2.0 ppm/°C, 25ppm/1khrs
MAX6250ACSA 0.02%, 2.0 ppm/°C, 20ppm/1khrs
REF5050IDGKT 0.05%, 3.0 ppm/°C, 50ppm/1khrs

I personally would not use the ebay part unless you have access suitable equipment to check at least initial accuracy.

[aghp]

With this,  it is also quite easy to make good 5V

TI REF102C
5ppm/khr typ
2.5ppm/Celsius.   

(but look datasheets more deep and you can see there are temp areas where it is nearly zero, I have used this for my "transfer standard" where pre aged and temperature shoked Ref102C is ovenized to 42 celsius with extremely tight temperature control an dwith good thermal insulation. But this is only one my own "volt-nut"experiment .  I have developed improved version using around 30 parallel "averaged" (noise reduction)  Ref102 but this experiment is not ready due to difficulties with low frequency noise handling... 0.01Hz to 10Hz  area and it IS difficult to reduce this noise residues. There in this experiment can also easy find that example 20 mm long thick copper wire is too high value resistor and then also knock head to "big" thermal voltage problems). It is more easy (and cheaper) to buy ready made Fluke transfer reference but experiment is good way to learn...least that it need respect HP and  Fluke and Keithley higher end items etc.))

[Fox]

The folowing ones would be good choices (also the Analog Devices one you mentioned)
LTC6655B 0.025%, 2ppm/°C max and 1ppm/°C typical
or
ISL21009 0.01% 3ppm/°C

It depend how do you want to use it, if want to use batteries to power the reference i would recommend the ISL21009 or the ADR4550.
If you want low noise performace the LTC6655 is the one to go.
Altough the  MAX6250 and the REF5050 are good references too.

As for ESD protection the LTC6655 has internal protection diodes and the ISL21009 is rated up to 6kV (Human Body Model), so you don't necessarily
need external ESD protection.
If you haven't already i would recommend reading the Appnote AN82 from Linear Technology "Understanding and applying Voltage References" and
Design Note DN229 "Don't be fooled by Voltage Reference Long-Therm Drift and Hysteresis" also from Linear Technology

[kcs]

TI REF102C
5ppm/khr typ
2.5ppm/Celsius.

5ppm is only for REF102CM (TO-99) which is obsolete and hard to get. For other packages it is 20 ppm. REF102 outputs 10V and requires expensive INA105 to get 5V.

It seems that HP 3458A uses LTZ1000 for reference voltage. Typ =7.2V, Min=7.0V, Max=7.5V. Is that precision?

I have made a comparison table of some reference ICs. Based on my findings, I believe I will go with MAX6250ACSA+. It seems to the best value for money.

[Bored@Work]

It seems that HP 3458A uses LTZ1000 for reference voltage. Typ =7.2V, Min=7.0V, Max=7.5V. Is that precision?

The absolute value doesn't matter. It is supposed to be calibrated. What matters is short and long-term drift, including tempco. And there that reference shines.

[Neganur]

It seems that HP 3458A uses LTZ1000 for reference voltage. Typ =7.2V, Min=7.0V, Max=7.5V. Is that precision?

Compare its tempco to the ones in your table

0.05ppm/°C

[kcs]

Compare its tempco to the ones in your table

0.05ppm/°C

Yeah, really not bad for $50.

[nukie]

LTZ1000 is as good as it can get and it is proven to work well in one of the best multimeter.

I have three sets of reference built around the LTZ1000 and LT1013 burn in and aged for two years. Readings taken from the lab, I couldn't read any drift with the Keithley 2000. I don't power it all the time, I just hook it jobs warm up for 2 hours before I take readings. Drift is too small for any 6 1/2 digit multimeter.

So maybe you might want more challenge building a 7v reference.

[alm]

The LTZ1000 is also used in commercial voltage references (eg. Fluke 732B). They key to precision references is usually the word 'selected'. Fluke and Agilent burn in the references for many hours and choose the one with the best drift/noise/tempco characteristics.

As was already stated multiple times in this thread, you should value stability over initial accuracy. You can measure a stable reference once and it will hold this value for years. A reference with good initial accuracy put poor tempco/drift will require much more frequent calibration, so what's the point of the initial accuracy?

[Hypernova]

Is there any study how how long a reference will age for, theses things are like wine but is there a point where they stop drifting in a practical sense?

[alm]

It will depend on the reference, some might never stop drifting. Agilent, Fluke and Keithley will have tons of data on the aging of the references they use (eg. LM399, LTZ1000), but I doubt that they ever published any of it.

In this article by Bob Pease he mentionts a testing program where they age LM199 references for six weeks. Note that this was only to select them. The LM199 runs at 85°C, not sure what this means for how fast the aging plateaus (high temperature will increase aging, but may also allow it to reach a plateau sooner). Some manufacturers specify a much lower aging rate after the first 1000h of operation, but this data may be based on accelerated life time studies (see the LT note that someone posted about what's wrong with that). Good references get much more stable over time, this is why many of those old bench DMMs are still in spec many years after their last calibration.

Their is a graph in the LT LM399 datasheet which indicates that the aging slows down after 4 months. LT DN229 also shows some data for much shorter time scales. I'm not sure how applicable this information is for other references, but a time scale in the order of months seems to be a reasonable estimate. Note that aging will just slow down, it won't stop completely.

[kcs]

I have understood that when you care about precision voltage reference you should be mainly interested in short and long period voltage shifts. Here are some of my thoughts:
First. There are different levels of precision, eg. 5.000V and 5.000000V. I don't aim for the latter one.
Second. VRE3050 datasheet says that "Output Voltage: +4.9995 (Min), +5.0000 (Typ), +5.0005 (Max)". Correct me if I am wrong, but I understand that no matter what (even if temperature during measurement is -40°C or +85°C; or IC is turned on for 1 min or 1M hours) it should never output voltage above 5.0005 or below 4.9995. For example, LTZ1000 is not good for my needs because it could output 7.123456789V and this value could be very stable but I don't have accurate enough device to get this reading. So I will never know that mine LTZ1000 outputs 7.123456789V. And no, finding a company or a person who can do that precise measurement is not a option in my case.

Interesting observation. Some manufacturers use "Output Voltage" statement and some use "Initial Output Voltage". As I understand, the first should guarantee that output will not exceed declared values no matter what.

[alm]

Output voltage for that device is specified for T=25C, not over the temp range. Note that the blurb on the first page talks about initial accuracy. Maybe I missed it, but what's wrong with the Geller labs and Voltagestandard products? You pay them the extra money to calibrate the reference with an expensive DMM.

[kcs]

Ok, you got me with T=25C. But it does not say anything about long-term stability. So at 25C over 10 years of continuous use output voltage should be within 4.9995-5.0005 range. No?

There is nothing wrong with Geller labs and Voltagestandard products, I just originally wanted to do it myself.

[poodyp]

But it does not say anything about long-term stability.

It says long term stability is 6ppm/1000hrs. After 10 years it could be off worst case by 525ppm, or ±2.625mV, but you'll never know.

[LaurenceW]

KCS, you state you want "easy" and "relatively precise", but I guess it depends on the degree of those terms!

I have used a couple of MAX6225 devices, and right out of the box WITHOUT any adjustment, each gave 2.5000V on my 50,000 count Fluke 289. (5V devices also available). I cannot get them to budge from this figure over a reasonable range of temperatures or durations, so suspect that these devices (and maybe others) already perform much better that the spec sheets, and certainly good enough for my application.

One thing, however. Make sure you stick to the spec sheets in terms of input and output capacitive decoupling, and take the voltage reference close to the pins of the device. Otherwise, you start to gain or lose even fractions of a mV in the voltage drop in your current-carrying circuit tracks/wires.

I got my devices FREE from Maxim, who will supply engineering samples. You have to give them some reasonable expectation that you might buy more of their devices in the future! I said I was a designer (well, I am and do).

[free_electron]

The LTZ1000 is also used in commercial voltage references (eg. Fluke 732B).

No it is not !. i got proof !. keep your eyes peeled....

[amspire]

The LTZ1000 is also used in commercial voltage references (eg. Fluke 732B).

No it is not !. i got proof !. keep your eyes peeled....

Interesting - just had a peek at the manual. There is hope for us all.

Teardown photos would be really interesting particularly on how they have built the oven.

I love the spec that says if the reference has been off for over 1 hour, you have to turn it on for 24 hours to settle.

[free_electron]

and that is what i got ... It was time to replace the battery so i tore it completely down

[fmaimon]

Not needed anymore --> If you want to wait for free_electron's teardown, don't try to read the spoiler below...



From the schematic on page 4-6 (60) of the 732B datasheet I would say that it uses the SZA263 / LTFLU-1, as it is the only reference I know that has this kind of connection...

[amspire]

From the schematic on page 4-6 (60) of the 732B datasheet I would say that it uses the motorola SZA263 or similar, as it is the only reference I know that has this kind of connection...

Looks like you are right. Is there any data anywhere for the SZA263?

[free_electron]

LTFLU-1 In mine combined with op97 to make 10 volt