Maxim 'precision' voltage dividers MAX5420A/21A

[splin]

In quotes because - a) it's Maxim and b) the specs are a somewhat lightweight on detail.

Has anyone played with or evaluated these? They are intended to provide the gain feedback resistors and switches for a PGA, with divider ratios of 1,2 4 or 8. The main 'hard' specification is accuracy of .025% which is pretty good (2ppm/K using the box method over -40C to 85C), if not Metrology standard. Naturally there are no long term drift or hysteresis specifications or sample to sample variance data.

However the typical graphs on page 3 and 4 are amazing - eyeballing the maximum slopes, dV/dT for the 1:1, 1:2, 1:4 and 1:8 ratios are .024, .13, .36 and .28ppm/K respectively!

Using the more usual box method, the TCs would be much better.

Does this pass anyones' smell test? The graphs for the lower grade B part are quite a bit different; since you'd expect the A and B grade parts to be selected from the same production lines this probably tells us, not surprisingly, that there is a lot of variation between individual parts. Nevertheless, the TCs shown for the B and C (same as B) grades are still low -  .04, .18, .2 and .8ppm/K for the 1:8 ratios.

[EmmanuelFaure]

It smells fishy, but... That being said, high resolution DACs perform at this level of acuracy. Example, the 16 bits DAC MAX541, gain drift typ 0.1ppm/°C, offset drift typ 0.05ppm/°C :
https://datasheets.maximintegrated.com/en/ds/MAX541-MAX542.pdf

[TheUnnamedNewbie]

Given how closely matched the resistors can be on-chip, I think this is easier to do on chip compared to with discrete. The components can be extremely well matched in terms of both temperature and impurities so it makes sense to me it's possible, esp. when you add in some calibration/trimming at the factory (which they will have to do in order to get absolute accuracy. As mentioned before, DACs that pull of this off are not uncommon - for example, the AD5790 does even better, with 0.018ppm/degree gain+offset drift which I believe is the best on the market right now? (well, it also costs 60 bucks in one-of...)

[splin]

It smells fishy, but... That being said, high resolution DACs perform at this level of acuracy. Example, the 16 bits DAC MAX541, gain drift typ 0.1ppm/°C, offset drift typ 0.05ppm/°C :
https://datasheets.maximintegrated.com/en/ds/MAX541-MAX542.pdf

The MAX5420 typical graph shows a 1:2 ratio TC of .032ppm/K using the box method. Even the LT5400 has a typical matching TC of 'only' .2ppm/K, which would equate to .1ppm/K for a 1:2 divider.

I'm not sure that gain and offset drift specs of a DAC reflect resistor matching tempcos though. The mid-scale stability is more relevant but many DAC datasheets don't show this in which case the INL v temperature specs are the most relevant. The AD5790 datasheet does show midscale drift v temperature showing a worse case slope of .1ppm/K with a 5V span, so again very comparable to the much cheaper MAX5421.

Yes of course it's possible to get excellent temperature matching of resistors on silicon, but the question was if anybody had actually tried these out to see if they perform anywhere near these typical specs, especially for a $1.6 part (C grade)?

It's hardly unknown to find that real parts perform much worse than the typical spec (but still within limits). Obviously this can happen due to perfectly legitimate reasons including a mid-life process change where the datasheet doesn't get updated.

Manufacturers have differing reputations in this area. I don't know how good or bad Maxim are but I have seen a few people complaining about parts that don't meet specs.

[splin]

Given how closely matched the resistors can be on-chip, I think this is easier to do on chip compared to with discrete. The components can be extremely well matched in terms of both temperature and impurities so it makes sense to me it's possible, esp. when you add in some calibration/trimming at the factory (which they will have to do in order to get absolute accuracy. As mentioned before, DACs that pull of this off are not uncommon - for example, the AD5790 does even better, with 0.018ppm/degree gain+offset drift which I believe is the best on the market right now? (well, it also costs 60 bucks in one-of...)

I does have some impressive specs - but the 16bit AD5760 has virtually identical specifications apart from resolution and INL, at a third of the cost but is still not dirt cheap. As I said in my previous reply the midscale TC is .1ppm/K max or .032ppm/K average so very similar to the MAX5421.

The tempco is only one factor though - long term stability is even more important as you can compensate the TC. Unfortunately the datasheet has nothing to say about this hence my original question.

Interestingly the AD5790 spec does (uniqually for a DAC?) specify long term stability of the INL at .1ppm typ after 750hours at 135C - far better than any discrete resistor spec that I've seen. Even the LT5400 only specifies 2ppm (2000h @ 35C) or 4ppm (2000h @ 7C).

By contrast a Vishay hermetic foil divider, VHD200 specifies 10ppm (2000h @ 70), but being Vishay it doesn't say if that is maximum, typical or 'typical limit' (whatever that means - they use the term in some other foil resistor datasheets).

I wonder why we don't see more IC resistor networks? There are lots of applications for precision dividers and they can outperform pretty much any alternative for stability and should be pretty cheap to manufacture as absolute tolerance is rarely important so trimming or testing to close limits aren't required. Granted they have their disadvantages such as higher noiser and limited resistance values but these aren't a problem for many usage cases.

[MisterDiodes]

Haven't used these per se, but if they're anything like LT5400 - lots of red flags for any metrology application.  Namely any diffused resistor on silicon, even though it's going to have a close TC match to it's neighbor - will have generally bad absolute value tolerance / stability; and -always- more noise than good PWW or Foil resistors (although AD9571's DACs do work well with pretty good noise  - for a chip resistor array - but you pay for that feature).

In terms of the glorious divider ratio tolerance they claim, usually you don't give a whit if your system has a software compensation somewhere.  What you really want most times is rock solid -stability- of the divider ratio over time and temp.; you don't care if it's a touch high or low because you've got system gain + offset compensation anyway.

For me, what seems to remove these from a serious metrology application is the noticeable lack of noise specs on the datasheet - that would be the first thing I'd investigate.  Second thing to look at is what is the ratio drift per 1000 hrs or drift /yr.  At least I didn't see that when I glanced at the datasheet.

Voltage standoff / self heating issues will also be another consideration - these won't be as rugged as good resistors, if that matters to you.  What happens with heating and cooling cycles?  Is there hysteresis? Humidity effects? What happens to the ratio when the board is flexed?  How do these tolerate voltage spikes & glitches?

One last thing to consider is it's Maxim...Generally a poor track record for keeping parts in stock if you have a project with an expected long lifetime.  That's just been our experience, YMMV.

This doesn't mean it's a bad part, just giving you some ideas where any chip-scale resistor array tends to really fall apart for precision application.  For serious applications generally you'll have some good, real resistors selected into the circuit via mosfet switch / relay, etc.

The advantage to these parts is they are small, and sometimes your application needs that more than high performance.

[MisterDiodes]

Another thought:

That part looks to be from 2005 and earlier, and back then they didn't know how to put it into a substrate heat spreader package like LT5400 (which is bad enough for metrology) - you'd probably want that feature for a precision divider for good thermal response, if that matters.  Without that your ratio can get wobbly -during- warm up / cool down.   Even though the chip is small you've got resistors on there with fairly large absolute-value TC (even though they are match close for ratio TC) and so you're not going to get a close ratio match unless the entire chip is at same temp.

Just something to consider.

Also doesn't look like its it's been a fast seller according to FindChips, and that should tell you something.  I'd worry about the Maxim-only supply chain if you were doing a big project design.

[splin]

Haven't used these per se, but if they're anything like LT5400 - lots of red flags for any metrology application.  Namely any diffused resistor on silicon, even though it's going to have a close TC match to it's neighbor - will have generally bad absolute value tolerance / stability; and -always- more noise than good PWW or Foil resistors (although AD9571's DACs do work well with pretty good noise  - for a chip resistor array - but you pay for that feature).

IC resistors clearly have their disadvantages including 1/f noise and ESD susceptibility but as far as I can see there are no other resistor technolgies (that aren't extremely esoteric/expensive such as JJAs) that come anywhere close as regards ratio tracking and ratio stability - though I'd be happy to be proven wrong.

Could you make even an 18 bit DAC out of any other type of resistor and still guarantee it to be monotonic, INL < 1LSB and claim typical INL drift of .11ppm after 1000 hours at 100C (AD5781)? The best Vishay hermetic foil divider that I can find is the VHD200/HD144 which is 100x worse at < 10ppm (2000h @ 70C) - the typical performance *might* be rather better, but it could be worse - this is Vishay after all. I'm pretty sure that PWWs are even further off the mark.

And if you could make one I reckon I'm on fairly safe ground that you couldn't match the 20 bit version. It may be that they do have serious problems with hysteresis, package stresses (temperature and humidity) etc. - but I'm sure that there are several people here, including you, that have some experience of these 18 and 20 bit DACs who could tell us how well these DACs maintain there specs over time.

I wouldn't expect these high performance resistor networks to have to be expensive either - the DACs are expensive partly because of marketing/competition but mostly I assume because of the necessity for high precision trimming and lengthy, high precision testing. For ratiometric applications these are hardly necessary because calibration is usually required anyway.

So why are there so few such parts available? Are there many others beyond the LT5400 and these Maxim parts?

In terms of the glorious divider ratio tolerance they claim, usually you don't give a whit if your system has a software compensation somewhere.  What you really want most times is rock solid -stability- of the divider ratio over time and temp.; you don't care if it's a touch high or low because you've got system gain + offset compensation anyway.

For me, what seems to remove these from a serious metrology application is the noticeable lack of noise specs on the datasheet - that would be the first thing I'd investigate.  Second thing to look at is what is the ratio drift per 1000 hrs or drift /yr.  At least I didn't see that when I glanced at the datasheet.

The Maxim parts don't specify these at all and they might not be very good. But the AD DACs seem to demonstrate that IC based resistors have the capability to be unsurpassed. As for noise, the AD5780 spec is 7.5nV/rt(Hz) which is the expected thermal noise from its 3400 ohm output resistance. The 1/f noise is quite high at 1.1uV p-p 0.1 to 10Hz, but it isn't clear if this is due to the resistor chain or pickup from elsewhere on the chip. The 16 bit AD5541A spec is 11.8nV/rt(Hz) (6250 ohms), 134nV p-p 0.1 to 10Hz 134nV p-p which suggests that the resistor network noise doesn't have to be much of a problem - though its possible that the process or design required for the ultra-stable 18 & 20 bit DACs have inherently noisy resistors.

Voltage standoff / self heating issues will also be another consideration - these won't be as rugged as good resistors, if that matters to you.  What happens with heating and cooling cycles?  Is there hysteresis? Humidity effects? What happens to the ratio when the board is flexed?  How do these tolerate voltage spikes & glitches?

One last thing to consider is it's Maxim...Generally a poor track record for keeping parts in stock if you have a project with an expected long lifetime.  That's just been our experience, YMMV.

This doesn't mean it's a bad part, just giving you some ideas where any chip-scale resistor array tends to really fall apart for precision application.  For serious applications generally you'll have some good, real resistors selected into the circuit via mosfet switch / relay, etc.

The advantage to these parts is they are small, and sometimes your application needs that more than high performance.

I hadn't realized that the Maxim part was relatively old so thanks for pointing that out. Chances are its performance may be quite respectable but I'm not going to invest any time evaluating them - unless perhaps someone can vouch for their stability?

[MisterDiodes]

IC resistors clearly have their disadvantages including 1/f noise and ESD susceptibility but as far as I can see there are no other resistor technolgies (that aren't extremely esoteric/expensive such as JJAs) that come anywhere close as regards ratio tracking and ratio stability - though I'd be happy to be proven wrong.

Well, you did ask the question in the Metrology section, and this is where we're always after top end performance.

For a simple divider, we do this all the time with quality PWW at DC, the specs on that divider part are not that hard to achieve with good PWW, even if they are real for that IC (doubtful). Get a proper matched set of ratio resistors made by Edwin or GR, get them properly thermal coupled and you'll be down to real, stable < 0.2ppm RATIO TC, and lowest possible noise.  We do that all the time for good inst. amps and find them -much- more stable and reliable than LT5400-type devices, especially for a real world signal conditioner that might be exposed to voltage glitches and "whatever".  I really, really doubt those other TC specs. on that Maxim part - but try one and see if it fits your application.

 If you need to do AC then a Caddock or Vishay foil divider will perform very well also.  Caddock is the absolute champ for rugged, stable dividers on a lot of Fluke DMMs, and they will build exactly what you need that will be very stable for decades.  Again, it's not the exact ratio that's important, its stability.  You'll calibrate out to the exact ratio needed.

Watch out for the inst. amps with integrated resistors also - There is the datasheet and then real world TC - sometimes you wind up with a much larger drift than you anticipated.  The other problem is those IC resistors are extremely stress-sensitive, head's up on that if you have to deal with vibration.

The truth is, diffused IC resistors are no match for real resistors for ruggedness and lowest possible noise. PWW is always lowest noise, then slightly up from that would be foil...and up from there is a chip-scale diffused resistor - those will always have more noise and relatively loose tolerance and large absolute TC compared to other choices...BUT if you can use them only in ratio mode on a proper design you can get smaller footprints and reasonable performance.

Build an 18-bit DAC from scratch? Sure, what's the budget? <Laughing>  Of course most of the time something like that will be on an IC, because there you're dealing with a controlled bias voltage and don't need that ruggedness like you might on a front end signal conditioner - but believe me the design you're talking about been done before and more in the lab, long ago.  The IC way is easier and more practical for a DAC of course, and you have the advantage of automated laser trimming.

For a simple divider (/2, /4, /8) like you originally mentioned, we normally use real resistors since we're typically after performance and stability most of the time.  Not all applications need that.

[MisterDiodes]

Just another quick head's up on AD5781 / 91 and that type of 18~20 DAC:

Splin, you touched on this briefly, but we're always very suspicious of datasheets changing units when the want to hide something:  For instance AD quotes that these parts having  7.5nV/rtHz noise density over and over again...but notice that's only at the higher frequency, say above 100Hz.  Get down to .1 to 10Hz range and suddenly they spec 1.1uV peak noise...which is more like 200+nV /rt Hz density.  Suddenly this is looking not quite so good.  It is still good for precision DAC's on a single chip, I'm not saying it's a bad part...but look out for those things in the datasheet when designing.

Now if you want to see how really crappy chip-scale resistors are for certain applications, start trying to design a test jig with one of these where the customer is looking for a DC spec that is more like 1mHz (or 100 uHz) to 10Hz, and under some uV noise peak at whatever signal voltage over whatever temp range and whatever drift rate over time.  When you get closer to DC that resistor (and whatever else) 1/f noise problem gets to be a much bigger problem, and now a solution that doesn't add 1/f noise starts looking more attractive.  Like -quality- PWW resistor dividers, if that's practical.

These type of parts are made with more resistors and switches than you'd find in a regular R-2R ladder - for instance these AD parts have 63 switches and resistor sets inside (at least that's what they admit to)...the LSB's are part of a standard R-2R ladder and then the upper MSB resistors are made up of groups of matched resistors.  In a nutshell, during manufacturing test those resistors are measured & trimmed to a certain set of ratios, and the logic section is programmed with a lookup table solution to convert your requested 18~20 bit DAC setting to the resistor array switch output combination that results in a low INL output.  That's a technique you can use with a chip-scale array of not-the-best resistors and wind up with a pretty good final result for INL.

More interesting items in those parts & datasheets, just some head's up items to be aware of when looking at precision DAC's:

1.  Notice that the INL changes with your reference voltage span - they give you a register to adjust that, but notice even with that the INL gets worse at lower spans.  Keep that in mind when designing.

2.  Notice how they give you specs in terms of LSB's?  That means the relative error might be better or worse depending on Ref span.  It kind of nice when LT comes right out and tells you "4 ppm" or whatever as max error, and that covers the whole range of what the part can do.  Notice the INL is plotted with more error at 5V ref span, I wonder what happens at lower Ref span input?

3.  The noise spec changes to "uV peak" from nV/rt Hz noise density at low freq.  Changing to better-sounding spec units = Red flag.  That means you have to check this yourself for your application.  If you don't need 20-bit adjustment range and more of a simple divider, then this is why you might look at discrete resistors to get you lower added noise at low freq.

4.  Notice the specs they give cover only 5V to 10V Vref Ref span, you probably want to stay in that same region for your design. 

5.  This is a good one:  The datasheets always seem to gloss over exactly what Vref chip is used.  It is always just shown as "5V Ref" or "10V Ref".  The DAC output noise at FS output is always going to be no better than the Vref used.

6.  Do you notice that the demo boards for these parts come with the layout for an LTZ1000a Vref?  Do you notice you get really best specs when your Vref span is right around 7VDC? Hmmmmm...   That should tell you something of what you might use for a good low-noise Vref for these DACs.  AD at the time couldn't come out and tell you that in the datasheet since LT was their competitor...I wonder what will happen now that it's one company?

Anyway, just some food for thought at what to look for on so called "Precision" datasheets.  Be careful of what is "by accident kinda sorta" being covered up.

Again, these aren't bad parts, just watch out for that 1/f noise problem at low freq.