4

Can a low-cost, low-resolution RDAC and a precision voltage reference by used to make a coarsely adjustable high-precision DC voltage reference? Consider for example the AD5121 7-bit RDAC with ±5ppm/°C drift and 8% resistor accuracy driven by the ADR4525 reference with 1.25μVpp from 0.1 to 10Hz and 4pmm/°C drift. Assume that the output will be buffered with a zero-drift amplifier.

The initial resistor accuracy aside, will this circuit give me an adjustable voltage with DC performance nearly as good as the reference on its own? The reference will be used to periodically null the input of a 24-bit ADC for precision DC voltage measurements at the 5ppm level.

I ask this because only high resolution DACs and RDACs provide DC (1/f) noise specifications like the voltage references do. If the low-resolution DAC can really act as a resistor with 5ppm/°C TC then I would expect it to work fine.

On a related note, there are no SPICE models for these RDACs so do they really behave enough like a discrete resistor/pot to simulate as such?

Mike
  • 513
  • 2
  • 15
  • 1
    The unavoidable reference for a dac is the legendary Jim Williams’ application note 86. https://www.analog.com/media/en/technical-documentation/application-notes/an86f.pdf . The note has some years but the fundamentals remain. – Undertalk Oct 23 '18 at 02:07
  • This TI design looks relevant but the MDAC seems overkill to do pretty much what I describe above: http://www.ti.com/lit/ug/tidu964/tidu964.pdf – Mike Oct 23 '18 at 02:46

1 Answers1

6

The 5ppm is only a typical spec, and there is no guaranteed maximum. Also the 35ppm is a typical spec (of total resistance drift). Note that the 3-sigma element resistance drift over lifetime is in the hundreds of ppm so some parts can be pretty bad. You don't see those changes directly in your configuration but they can result in side effects if there are temperature gradients on the chip (from internal or external heat flow) or stresses.

The tempco of the 'wiper' resistance is surely much, much higher but you say you will buffer that so it may not be an issue. It will also be somewhat sensitive to supply voltage.

When you get down to the ppm level things like the stress on the die (from board flexing, from soldering, from environmental changes) can affect the value of "semi-precision" resistors like the on-chip parts in question. If it's designed to be a high precision part they can put extra cost in there to compensate (perhaps by using symmetrical resistors that compensate for each other). We don't really know how carefully they've designed that chip.

I don't see what good a SPICE model would do you for a part like this- you're looking at deviations from the nominal so a calculator and sharp pencil (and a caffeine-sharpened mind) are usually your best tools.

Bottom line- if you need guaranteed specifications you should specify parts with guaranteed performance. Otherwise you may find yourself trying to make a silk purse of a sow's ear.

Spehro Pefhany
  • 376,485
  • 21
  • 320
  • 842
  • There is a 30x difference in cost between parts like the AD5121 and the AD5781 so I am trying to get a feel for what is necessary to do the job. I would like the stated noise performance of the AD5781 but I don't care at all about dynamic performance or resolution. I'll probably just need to get one of these on the bench and try it out. – Mike Oct 23 '18 at 02:44
  • 1
    Contact the ADI Applications Engineers. I suspect they answer these sort of tradeoff questions all the time. – analogsystemsrf Oct 23 '18 at 05:03