Author Topic: Want a 5ppm/C (max) reference? Then don't buy a voltage reference...  (Read 5352 times)

0 Members and 1 Guest are viewing this topic.

Offline splinTopic starter

  • Frequent Contributor
  • **
  • Posts: 999
  • Country: gb
... buy a DAC!

The cheapest TI 2.5V reference better than 5ppm/C (max) is the 3ppm/C LM4140ACM-2.5 costing $2.39 @ 1K from TI. The B-grade, LM4140BCM-2.5, 6ppm/C max is $2.0.

Or, for only $1.14 you could buy a DAC80501 DAC60501- a 3ppm typical, 5ppm/C max reference with a 12-bit DAC thrown in for free! Ok, it's a bit noisier with 14uVp-p (0.1 to 10Hz) v 5.4uVp-p for the LM4140 - but you get an unbelievably good drift specification of only 12uV, or 4.8ppm @ 35C over 1900 hours compared to 60ppm over 1000 hours for the LM4140!!

As I said, unbelievable. Unless they have step-change technology here, it probably has similar drift to other plastic encapsulated band gap references, so nothing to see here folks.

The DAC isn't too shabby either with +/- 1 LSB INL, but if you are really prepared to push the boat out and splurge another $0.86 you get a 16 bit, +/- 1LSB INL DAC for $2.86.

You've got to love pricing anomolies - but it does make component selection a right pain in the  great deal of fun. The flip side is that you are always wondering if some uber low cost camera imaging IC or the like could have provided a much cheaper ADC + data encoder solution (that would need less processing to re-encode into your required form than processing from scratch) for your latest wonder product.

And lest you think this post is some sort of attempt at TI viral marketing (it isn't), the AD5310R @ $1.29 is a 10 bit DAC with a 2ppm/C typical, 5ppm/C max reference. It's a bit noisier though @ 16.5uVp-p.

[EDIT] Typo - the DAC80501 is the 16 bit DAC, the $1.14 12 bit DAC is the DAC60501 - thanks to SiliconWizard for pointing it out.
« Last Edit: July 09, 2019, 02:03:46 pm by splin »
 

Offline krayvonk

  • Regular Contributor
  • *
  • !
  • Posts: 144
  • Country: au
Re: Want a 5ppm/C (max) reference? Then don't buy a voltage reference...
« Reply #1 on: July 09, 2019, 02:50:16 am »
How much does it cost per bit of significance?
 

Offline capt bullshot

  • Super Contributor
  • ***
  • Posts: 3033
  • Country: de
    • Mostly useless stuff, but nice to have: wunderkis.de
Re: Want a 5ppm/C (max) reference? Then don't buy a voltage reference...
« Reply #2 on: July 09, 2019, 06:01:42 am »
Yes, not that surprising.
If you need a simpe 12Bit ADC with 8 or more input channel multiplexer, it's cheapest to use a small uC for that.
Safety devices hinder evolution
 

Offline exe

  • Supporter
  • ****
  • Posts: 2647
  • Country: nl
  • self-educated hobbyist
Re: Want a 5ppm/C (max) reference? Then don't buy a voltage reference...
« Reply #3 on: July 09, 2019, 08:34:00 am »
But is DAC80501 on sale yet? I don't see it in stock... anywhere.
 

Online tszaboo

  • Super Contributor
  • ***
  • Posts: 8218
  • Country: nl
  • Current job: ATEX product design
Re: Want a 5ppm/C (max) reference? Then don't buy a voltage reference...
« Reply #4 on: July 09, 2019, 08:58:37 am »
The LM4140 is old silicon, and doesn't represent the best technology.
The DAC80501 specified temperature range is 0-70C, while most of the other references that are about 5-8ppm/C from TI are specified from -40-125C. They use the box method to give you the tempco, and on the lower and higher extremes, the tempco is a lot worse than next to 0 degrees. That means, that a REF3425 with 6ppm specification will behave much better on the 0-70 C region than that DAC. Just look at Figure 3 in SBAS804B.
Not to mention the specification for the DAC's reference seems to be... missing important info. Like the drift spec is for 1900 hours 35C, instead of first 1000 hour, second 1000 hours. And the value there... Let's just say I dont believe it. They claim 12uV, which is less than 5ppm drift for almost 2000 hours. It is better that this is pre-production datasheet.
 

Online tggzzz

  • Super Contributor
  • ***
  • Posts: 21226
  • Country: gb
  • Numbers, not adjectives
    • Having fun doing more, with less
Re: Want a 5ppm/C (max) reference? Then don't buy a voltage reference...
« Reply #5 on: July 09, 2019, 08:59:18 am »
And be aware of the other specs and factors, e.g.
  • initial accuracy 0.1% (i.e. 1000ppm)
  • 60ppm drift after 1000 hours (10% of devices 140ppm)
  • 20ppm thermal hysteresis
  • 9uV/sqrt(Hz) noise at/below 0.1Hz
  • PCB/mounting stresses, especially for SMD devices
  • humidity
There are lies, damned lies, statistics - and ADC/DAC specs.
Glider pilot's aphorism: "there is no substitute for span". Retort: "There is a substitute: skill+imagination. But you can buy span".
Having fun doing more, with less
 
The following users thanked this post: BrianHG

Offline SiliconWizard

  • Super Contributor
  • ***
  • Posts: 15797
  • Country: fr
Re: Want a 5ppm/C (max) reference? Then don't buy a voltage reference...
« Reply #6 on: July 09, 2019, 01:43:14 pm »
The DAC80501 is an interesting product indeed, still not available (preproduction stage).
http://www.ti.com/product/DAC80501

Just for clarification, the DAC80501 is a 16-bit DAC (not 12-bit) and right now the unit cost per 1k is given at $2.86.
The 12-bit version is the DAC60501, displayed at $1.14 per 1k indeed. As it's in preproduction, the unit cost may not quite be definitive...

The initial accuracy of the internal ref is +/-0.1%, same as the LM4140. Not bad. As said above, its specs are not spectacular either and some info that is usually given for voltage references are not given here. Also compare the load regulation between the two... At 2.5V, it's almost 10x worse for the DACx0501. So it would all depend on how you're going to use it.

When I need something better, I go for something like the MAX6126. About twice the cost of the LM4140. But that is what I'd call a voltage reference.
« Last Edit: July 09, 2019, 01:57:21 pm by SiliconWizard »
 

Online tggzzz

  • Super Contributor
  • ***
  • Posts: 21226
  • Country: gb
  • Numbers, not adjectives
    • Having fun doing more, with less
Re: Want a 5ppm/C (max) reference? Then don't buy a voltage reference...
« Reply #7 on: July 09, 2019, 03:27:11 pm »
When I need something better, I go for something like the MAX6126. About twice the cost of the LM4140. But that is what I'd call a voltage reference.

Still 20ppm or 100ppm per 1000hours. Voltage references usually stabilise over time, and the spec is ppm/sqrt(hours).

Besides, "friends don't let friends use Maxim".
There are lies, damned lies, statistics - and ADC/DAC specs.
Glider pilot's aphorism: "there is no substitute for span". Retort: "There is a substitute: skill+imagination. But you can buy span".
Having fun doing more, with less
 

Offline SiliconWizard

  • Super Contributor
  • ***
  • Posts: 15797
  • Country: fr
Re: Want a 5ppm/C (max) reference? Then don't buy a voltage reference...
« Reply #8 on: July 09, 2019, 03:40:54 pm »
When I need something better, I go for something like the MAX6126. About twice the cost of the LM4140. But that is what I'd call a voltage reference.

Still 20ppm or 100ppm per 1000hours. Voltage references usually stabilise over time, and the spec is ppm/sqrt(hours).

Yeah. It's just a couple dollars IC. It's good and has served me well on occasion, but don't expect it to beat a precision, temperature-stabilized voltage reference in an oven. ;D

Besides, "friends don't let friends use Maxim".

Yeah, this is your opinion though. Maxim makes some pretty good products (I tend to favor Linear/AD for various reasons, but still). I admit they are not always the best with support and availability (especially for their MCUs as far as I'm concerned), but to knock them off altogether... sounds a bit excessive.
 

Offline Conrad Hoffman

  • Super Contributor
  • ***
  • Posts: 2077
  • Country: us
    • The Messy Basement
Re: Want a 5ppm/C (max) reference? Then don't buy a voltage reference...
« Reply #9 on: July 09, 2019, 04:20:40 pm »
No idea what they're like today but older engineers remember designing Maxim parts into new products, only to have the parts discontinued some months later. It's hard to forgive when that happens a couple times. Management doesn't remember the vendor as clearly as they remember the idiot who used them.
 

Offline SiliconWizard

  • Super Contributor
  • ***
  • Posts: 15797
  • Country: fr
Re: Want a 5ppm/C (max) reference? Then don't buy a voltage reference...
« Reply #10 on: July 09, 2019, 05:24:24 pm »
Well, out of curiosity, I just took a look again at the MAX6126 page: https://www.maximintegrated.com/en/products/analog/voltage-references/MAX6126.html
They state "But some versions of the family are Not Recommended for New Designs", which kinda sounds, knowing Maxim, like those versions are going to be discontinued any time now. You make a point. But it has lasted for a few years (I think I selected it like 4 years ago, that's admittedly not much).

Its successor doesn't look bad though: https://www.maximintegrated.com/en/products/analog/voltage-references/MAX6226.html
 
The following users thanked this post: exe

Offline bob91343

  • Super Contributor
  • ***
  • Posts: 2675
  • Country: us
Re: Want a 5ppm/C (max) reference? Then don't buy a voltage reference...
« Reply #11 on: July 09, 2019, 05:44:44 pm »
Initial accuracy of 0.1% isn't good enough.  I need something that is closer, right out of the box, to calibrate my instruments.  Even 0.01% is marginal.

Is there anything reasonable in price that is good enough?  Back in the day you could buy a Weston standard cell and get a pretty good reference but they were expensive.
 

Online tggzzz

  • Super Contributor
  • ***
  • Posts: 21226
  • Country: gb
  • Numbers, not adjectives
    • Having fun doing more, with less
Re: Want a 5ppm/C (max) reference? Then don't buy a voltage reference...
« Reply #12 on: July 09, 2019, 05:46:05 pm »
Besides, "friends don't let friends use Maxim".

Yeah, this is your opinion though. Maxim makes some pretty good products (I tend to favor Linear/AD for various reasons, but still). I admit they are not always the best with support and availability (especially for their MCUs as far as I'm concerned), but to knock them off altogether... sounds a bit excessive.

My introduction to Maxim was for a new interesting specialised device. performance was right, price was right, but there was never any availability. I had the feeling they "ran the device up the flagpole to see who saluted"; no salute => no production.

Only later did I come across "friends don't let friends use Maxim". It resonated with me, i.e. it confirmed my prejudice.
There are lies, damned lies, statistics - and ADC/DAC specs.
Glider pilot's aphorism: "there is no substitute for span". Retort: "There is a substitute: skill+imagination. But you can buy span".
Having fun doing more, with less
 

Offline SiliconWizard

  • Super Contributor
  • ***
  • Posts: 15797
  • Country: fr
Re: Want a 5ppm/C (max) reference? Then don't buy a voltage reference...
« Reply #13 on: July 09, 2019, 05:51:46 pm »
I understand. And now that I've just seen that the reference I used is going deprecated doesn't bode well either... this was only for a few units-project so not a huge issue here, but still. The "successor" I just saw, flagged as "new", looks pretty nice but... it costs almost twice as much!
 

Online tggzzz

  • Super Contributor
  • ***
  • Posts: 21226
  • Country: gb
  • Numbers, not adjectives
    • Having fun doing more, with less
Re: Want a 5ppm/C (max) reference? Then don't buy a voltage reference...
« Reply #14 on: July 09, 2019, 05:52:22 pm »
Initial accuracy of 0.1% isn't good enough.  I need something that is closer, right out of the box, to calibrate my instruments.  Even 0.01% is marginal.

Is there anything reasonable in price that is good enough?  Back in the day you could buy a Weston standard cell and get a pretty good reference but they were expensive.

Saturated Weston cells are 40uV/C, i.e. ~40ppm/C. Mine is still in spec after 70 years.

Newer references are based on zener diodes and are unsurprisingly expensive.
One option is to get a well-aged device and get it calibrated by a calibrated HP3458.
Alternatively new 6.5 digit bench multimeters are around £1000.
There are lies, damned lies, statistics - and ADC/DAC specs.
Glider pilot's aphorism: "there is no substitute for span". Retort: "There is a substitute: skill+imagination. But you can buy span".
Having fun doing more, with less
 

Online bdunham7

  • Super Contributor
  • ***
  • Posts: 8175
  • Country: us
Re: Want a 5ppm/C (max) reference? Then don't buy a voltage reference...
« Reply #15 on: July 09, 2019, 06:48:28 pm »
Initial accuracy of 0.1% isn't good enough.  I need something that is closer, right out of the box, to calibrate my instruments.  Even 0.01% is marginal.

Is there anything reasonable in price that is good enough?  Back in the day you could buy a Weston standard cell and get a pretty good reference but they were expensive.

I think this is as reasonable as you can get for a quasi-professional product.  By "quasi-professional" I mean the accuracy and service provided are top-notch, while the appearance and construction are hobbyist level.

http://www.voltagestandard.com/Home_Page.html
A 3.5 digit 4.5 digit 5 digit 5.5 digit 6.5 digit 7.5 digit DMM is good enough for most people.
 

Offline splinTopic starter

  • Frequent Contributor
  • **
  • Posts: 999
  • Country: gb
Re: Want a 5ppm/C (max) reference? Then don't buy a voltage reference...
« Reply #16 on: July 10, 2019, 12:51:27 am »
The LM4140 is old silicon, and doesn't represent the best technology.

True but not very relevant - the LTZ1000 is a lot older than the LM4140, but is still the best reference on the market (for TC, noise and stability - not price or power consumption or initial tolerance). This thread is not about references that 'represent the best technology' ( what even is that?) but the cheapest, or nearly so, 5ppm/C (max) reference available (from a 1st tier manufacturer).

There are many characteristics other than TC which need to be considered and the DAC60501's reference will not meet the needs of many applications - but it may will be good enough for others where it's attractive price compared to the alternatives could make it a good choice, and a free 12-bit DAC, or a 16 bit DAC for only $1.74 extra may make it very attractive.

Part of the reason for the OP was to point out that sometimes pricing oddities occur which mean that the 'obvious' solutions aren't always the best. Another example is that if you want an 80MSPS 12 bit ADC, by far the cheapest solution is to choose the $5.39 LPC4370 and get a free triple core, 208MHz ARM Cortex M4 micro thrown in for free. It obviously won't suit all applications, but for some it is outstanding in comparison to the free standing ADC and an FPGA and/or microcontroller solution you originally had in mind. If you were only searching for high speed ADCs you wouldn't find this option.

It would be great if anyone could come up with some other hidden gems of parts that you wouldn't expect or can be repurposed to good advantage. Parts used in very high volumes, such as toys, lighting, white goods etc. can be good candidates because the pricing can be very low and because of their specialist nature aren't well known outside of their direct application area - and because datasheets may not even be generally available.
 
Quote
The DAC80501 specified temperature range is 0-70C

It's TC is specified as:
  3ppm typ, 5ppm max 0 to 70C and
  5ppm typ, 8ppm max -40 to 125C
which is a much better way of specifying the TC than the single -45 to 125C figure given for most references - many applications aren't interested in temperatures outside the 0-70C range. Eg. a bench meter.

If 8ppm max over the industrial temperature range is not good enough for you but 6ppm is, then by all means choose the more expensive REF3425.

Quote
while most of the other references that are about 5-8ppm/C from TI are specified from -40-125C. They use the box method to give you the tempco, and on the lower and higher extremes, the tempco is a lot worse than next to 0 degrees. That means, that a REF3425 with 6ppm specification will behave much better on the 0-70 C region than that DAC. Just look at Figure 3 in SBAS804B.

You can't assume anything about the characteristics that aren't guaranteed limits, especially based on a graph of a single supposedly 'typical' unit in a datasheet. That graph may have been created when testing a few early pre-production samples or it could (unlikely) be an average of hundreds of production samples from many different batches - the datasheets rarely give such insights. Even then, production processes vary over time so using a typical graph of parts made years earlier when the DS was written is ill advised.

Take a look the graph of the 1.8V LM4132 for example:



Three of the 5 samples actually have higher TCs over 0-70C than -40 to 125C. The graphs for other voltage LM4132s have similar shapes but with greater extremes at the higher temperature end but is that because there is something special about the 1.8V version or because they came from different batches?

In practice, the REF3425 may well behave better than the DAC's reference over 0-70C but you can't assume it from the DS information; if it's important then you need to characterise and qualify the parts yourself - and test every new batch you get until you are confident the batch to batch variation is small enough.

Quote
Not to mention the specification for the DAC's reference seems to be... missing important info. Like the drift spec is for 1900 hours 35C, instead of first 1000 hour, second 1000 hours. And the value there... Let's just say I dont believe it. They claim 12uV, which is less than 5ppm drift for almost 2000 hours. It is better that this is pre-production datasheet.

Yes, the drift spec is almost certainly hokum - which is why I went to the trouble to point it out in my OP:

Quote
but you get an unbelievably good drift specification of only 12uV, or 4.8ppm @ 35C over 1900 hours compared to 60ppm over 1000 hours for the LM4140!!

As I said, unbelievable. Unless they have step-change technology here, it probably has similar drift to other plastic encapsulated band gap references, so nothing to see here folks.
 
The following users thanked this post: thm_w

Offline splinTopic starter

  • Frequent Contributor
  • **
  • Posts: 999
  • Country: gb
Re: Want a 5ppm/C (max) reference? Then don't buy a voltage reference...
« Reply #17 on: July 10, 2019, 01:30:00 am »
When I need something better, I go for something like the MAX6126. About twice the cost of the LM4140. But that is what I'd call a voltage reference.

Still 20ppm or 100ppm per 1000hours. Voltage references usually stabilise over time, and the spec is ppm/sqrt(hours).

I'm very dubious about those stability specs - 20ppm is very good for a plastic encapsulated part. Other references available in metal or ceramic packages as well as plastic all show much worse drift for the plastic parts - usually 40 to 50ppm or more. Ie. it is the plastic which is the culprit, no matter how good the reference itself.

For a bit more evidence however, take a look at the graphs showing the stabiity over 1000 hours for the 2.5V uMax and SO parts - the two 'typical' uMax parts show 109ppm and 133ppm drift (max) compared to the spec of 100ppm. The SO parts show 29ppm and 35ppm maximum drift compared to the 20ppm spec.

A typical figure without any indication of the actual worst case spread or varience is almost useless in any case.
 

Offline bob91343

  • Super Contributor
  • ***
  • Posts: 2675
  • Country: us
Re: Want a 5ppm/C (max) reference? Then don't buy a voltage reference...
« Reply #18 on: July 10, 2019, 03:01:07 am »
I do have my HP 3456A that is my house standard for voltage.  But some day I may want to verify and/or calibrate it.  Short of an expensive deal with NIST or some local equivalent, I don't have any idea what I could do.

I also have an HP 3455A but it's not working.  If I can fix that I will be happy.  When it worked, it agreed spot on with the 3456A but then turned belly up.  If anyone has serious experience working on one of these I would love to get some advice.

I also have a few 10V IC chips that agree with one another remarkably well.  At one time I had them switch selectable so I could verify my voltmeters.  Once they have been characterized they will probably hold calibration very well.  I should reconnect that setup and take notes.

Now and then I ask myself why I need accurate voltage but have no satisfactory answer other than it must be a fetish.
 

Offline David Hess

  • Super Contributor
  • ***
  • Posts: 17427
  • Country: us
  • DavidH
Re: Want a 5ppm/C (max) reference? Then don't buy a voltage reference...
« Reply #19 on: July 10, 2019, 03:01:19 am »
I would be cautious about directly comparing references from some companies including Analog Devices and Texas Instruments.  In the past both have advertised noise and temperature coefficient specifications which were questionable or even impossible.  Pease and others called them out on this but I do not know that the datasheets were ever corrected.
 

Offline bob91343

  • Super Contributor
  • ***
  • Posts: 2675
  • Country: us
Re: Want a 5ppm/C (max) reference? Then don't buy a voltage reference...
« Reply #20 on: July 10, 2019, 03:04:26 am »
I do understand and appreciate your caution.  However, statistically speaking comparing 5 or 6 references over the long haul should show which if any are suspect.  I don't expect them all to be bad equally.

As I said, it's a fetish so I will probably assume they are good and blindly go on with my life.
 

Offline SiliconWizard

  • Super Contributor
  • ***
  • Posts: 15797
  • Country: fr
Re: Want a 5ppm/C (max) reference? Then don't buy a voltage reference...
« Reply #21 on: July 10, 2019, 07:20:13 am »
I'm very dubious about those stability specs - 20ppm is very good for a plastic encapsulated part. Other references available in metal or ceramic packages as well as plastic all show much worse drift for the plastic parts - usually 40 to 50ppm or more. Ie. it is the plastic which is the culprit, no matter how good the reference itself.

Indeed. The new MAX6226, which has significantly better specs in terms of drift, only comes in a ceramic package.
« Last Edit: July 10, 2019, 07:23:31 am by SiliconWizard »
 

Offline exe

  • Supporter
  • ****
  • Posts: 2647
  • Country: nl
  • self-educated hobbyist
Re: Want a 5ppm/C (max) reference? Then don't buy a voltage reference...
« Reply #22 on: July 10, 2019, 09:48:03 am »
I'm very dubious about those stability specs - 20ppm is very good for a plastic encapsulated part. Other references available in metal or ceramic packages as well as plastic all show much worse drift for the plastic parts - usually 40 to 50ppm or more. Ie. it is the plastic which is the culprit, no matter how good the reference itself.

I'm no expert, but I agree. May be measurements done in a controlled environment with constant humidity, pressure etc.

I wonder if it's possible to apply some coating to make plastic not absorbing moist.
 

Online tszaboo

  • Super Contributor
  • ***
  • Posts: 8218
  • Country: nl
  • Current job: ATEX product design
Re: Want a 5ppm/C (max) reference? Then don't buy a voltage reference...
« Reply #23 on: July 10, 2019, 10:24:33 am »

Three of the 5 samples actually have higher TCs over 0-70C than -40 to 125C. The graphs for other voltage LM4132s have similar shapes but with greater extremes at the higher temperature end but is that because there is something special about the 1.8V version or because they came from different batches?
That's not how the box method works, and it is impossible to work with that graph for the box method. You have to take a bunch of samples (25-100), normalise it somewhere (for example 25 degrees) and measure the spread. Then divide it with the range. Most references will have some kind of a compensation built in, for example a zener with a regular diode in series, with different direction of the typical tempco.
But anyway, look at figure 1 of the REF3425, and tell me that the typical value is not good enough indication. I know when I need a voltage reference, which one I typically use, a dedicated and characterised one or something that comes as an extra.

Initial accuracy of 0.1% isn't good enough.  I need something that is closer, right out of the box, to calibrate my instruments.  Even 0.01% is marginal.

Is there anything reasonable in price that is good enough?  Back in the day you could buy a Weston standard cell and get a pretty good reference but they were expensive.
Let's say, I can sell you a 0.001% accurate reference, that changes it's output voltage by up to 0.02%, when you solder it to a PCB. Would you buy it? If yes, than I also have a bridge to sell.
 

Online tggzzz

  • Super Contributor
  • ***
  • Posts: 21226
  • Country: gb
  • Numbers, not adjectives
    • Having fun doing more, with less
Re: Want a 5ppm/C (max) reference? Then don't buy a voltage reference...
« Reply #24 on: July 10, 2019, 11:01:56 am »
I'm very dubious about those stability specs - 20ppm is very good for a plastic encapsulated part. Other references available in metal or ceramic packages as well as plastic all show much worse drift for the plastic parts - usually 40 to 50ppm or more. Ie. it is the plastic which is the culprit, no matter how good the reference itself.

I'm no expert, but I agree. May be measurements done in a controlled environment with constant humidity, pressure etc.

I wonder if it's possible to apply some coating to make plastic not absorbing moist.

"Conformal coatings" do not completely solve problems, and introduce new problems.
There are lies, damned lies, statistics - and ADC/DAC specs.
Glider pilot's aphorism: "there is no substitute for span". Retort: "There is a substitute: skill+imagination. But you can buy span".
Having fun doing more, with less
 


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf