11

When designing precision analog circuits, I often come across parts which seem to be more than accurate enough for my purposes, but where the datasheet does not specify how key parameters will change over time.

Right now, I am looking at datasheets for difference amplifiers (e.g. the INA157), and the CMRR looks better than what I would achieve by using affordable matched resistor dividers (e.g. the MAX5490). However, the resistor ratios will drift over time, which will reduce the CMRR.

Resistor dividers often give a typical value for this drift in ratio, so I can estimate how long my circuit can go without recalibration. However, while some of the difference amplifiers I saw specify input offset drift over time, I didn't see one yet which specifies the change in CMRR or the resistor ratio matching over time.

I'd assume that the parameters won't drift much beyond the initial limits over time, and this seems to be true e.g. for the offset voltage of many op amps, but on the other hand, I remember seeing 0.1% resitors which were only specified to drift less than 2% (or something along that magnitude) within a few thousand hours.

Now I'm wondering: Is there some rule of thumb for estimating how the CMRR (or similar parameters without aging specification) will develop? Can I assume that it will remain above the "minimum" specification even after some years of use? If not, for how many hours of use does the datasheet specification actually remain valid?

gsills
  • 7,133
  • 16
  • 22
Medo42
  • 607
  • 5
  • 17
  • 1
    While resistor values in difference amplifiers will drift over time, they're constructed so that they'll all change in a similar fashion, so CMRR shouldn't vary because of resistor changes. – Scott Seidman Dec 05 '14 at 13:23
  • 2
    The resistors in matched networks are also constructed this way, and their ratios do drift (albeit slowly), since the resistors don't age *exactly* the same way. For example, the MAX5490 matched resistor divider specifies a ratio change of 0.03% typical over 2000 hours at 70°C. That seems very small, but high CMRRs demand very precise matching. For example, a CMRR of 80dB requires the ratios of the two resitor dividers to be matched within 0.02%. – Medo42 Dec 05 '14 at 13:37
  • Have you asked the manufacturer yet? I certainly don't know the answer. But I imagine there are people in the business who do know, and they probably work for national, linear, TI, Maxim or whatever. No offense to anyone I left out. – user57037 Dec 16 '14 at 04:25
  • @mkeith Not yet. I'll try sending an eMail to Analog, because they make the IC that I settled on for now, and I'll report if / when I get an answer. – Medo42 Dec 16 '14 at 15:39
  • 2
    I re-posted my question in Analog Devices' Engineer Zone now: https://ez.analog.com/message/168215. – Medo42 Dec 16 '14 at 17:25
  • I guess someone should also note that the question has received a fairly satisfactory answer there (on AD site). In summary: they test a few high-end parts explicitly for this issue, but for the lower end stuff they just assume the fabrication process ensures stability via equal drift over time to a certain extent. They also say that most drift occurs during a "break in" period (my term) and so suggest to [re]test yourself after about 1000 hours (one month) of continuous use. – Fizz Dec 21 '14 at 21:15
  • I already started writing an answer based on that reply, but wanted to wait until I had a bit more clarification. The way I understood the answer, a device will typically still be within the initial min/max specs after 1000 hours, and drift slows down over time, so it probably won't go much beyond the min/max specs within several thousand hours. I'm not 100% sure about that interpretation though, so I'm waiting for an answer to my follow-up question. – Medo42 Dec 21 '14 at 21:37
  • @Medo42 and that link is dead now, which is why we encourage people to post content here instead of just links – endolith Aug 08 '19 at 18:14

1 Answers1

2

If the datasheet doesn't show the details of a particular contribution to the error of some value, then you have to assume that error contribution is already taken into account in the min/max range of the value.

In your example, the datasheet doesn't show CMRR drift over time. You can therefore assume that this drift is already accounted for in the min CMRR spec. That doesn't mean the CMRR won't drift over time, temperature, phase of the moon, or what you had for breakfast, but that it won't go below the specified minimum under all combinations of valid conditions.

Design your circuit to tolerate the minimum CMRR, and all should be OK. Most of the time you will get better than that, but you can't rely on that.

Note that this also means performing calibration to initial conditions found at first powerup provides no guarantee. In fact, it makes the worst case worse since this will add a offset to the error value. For example, let's say a parameter is guaranteed to be from 95 to 105, with 100 being the nominal. That means the device could at any time, for any reason, drift around between 95 and 105. If it happens to be at 105 when you perform calibration, by having the external circuit subtract 5, you will get 90 when the device drifts to 95.

To perform calibration with any guarantees, the parameter you calibrate out has to be separated into initial accuracy and a long term drift components. If the device you are trying to use isn't specified that way, either talk to the manufacturer to see if more detail might be available, or use a different part.

Olin Lathrop
  • 310,974
  • 36
  • 428
  • 915
  • I find that a surprisingly strong interpretation of the specs. I always assumed that manufacturers would mention it in their datasheets if they had long-term tests of their components - and without those, how could they make long-term promises? I'm sure you have experience behind that claim, but can you point me at some textbook / manufacturer statement as source? – Medo42 Dec 23 '14 at 13:25
  • @Med: You are trying to read too much into the datasheet. You are only guaranteed what it says you are guaranteed. Trying to guess what else you can get away with will get you into trouble. Manufacturers usually know a lot more about their devices than they put in datasheets. If you think a device may have additional characterstics that are not specified, talk to a applications engineer from the company. – Olin Lathrop Dec 23 '14 at 13:58
  • I'm not sure how to understand that. You seem to be taking two opposing points in your answer and comment - first you suggest that I can rely on the min/max values to hold long-term if long-term drift is not specified in the datasheet, and then you say I shouldn't rely on things not specified in the datasheet. Which is it? – Medo42 Dec 23 '14 at 14:12
  • @Med: You can only rely on things specified in the datasheet, unless you get additional information from the manufacturer. However, if drift is not specified separately, then the long term range the value *is* specified in the datasheet. If no mention is made that a value is only valid for some time, then you can assume it applies over the life of the part, assuming the absolute maximum limits were adhered to at all times. – Olin Lathrop Dec 23 '14 at 14:21
  • Ok, thanks for clarifying and also for your answer. I only wanted to ask above whether you can provide a source to confirm your answer (i.e. that long-term drift is included in min/max specs unless seperately mentioned), because I always assumed min/max specs only gave you *initial* accuracy. So, I'm reading less into the datasheet than you are, and thus found your comment surprising. – Medo42 Dec 23 '14 at 14:36