Resistance to Solder Heat, practical values, SMD resistor

The other side of this sort of calculation is that the manufacturers are covering themselves against improbably bad devices. It's tedious and expensive to test enough resistors to have a clear idea of the distribution of resistance shifts after exposure to solder heat, or to

70C for 1000 hours, so they cut the numbers and allow for the reduced precision by allowing for a bigger spread than they actually saw in reality.

The statistical expectation of a lot of different, independent shifts is the root mean square sum of all the individual standard deviations, and while there is a tail on this disribution, it gets down to one or two in a million a about six times the aggregate standard deviation.

At this level of improbability, the parts that don't work the way you promised are more likely to have been screwed up by being struck by lightning.

-- Bill Sloman, Nijmegen

Our experience is that a 1% resistor really is a 1% resistor. We design as such.

John

John Larkin a écrit :

My experience is that a few (4 out of a batch of 250) 1% 2.7Meg 0201 have drifted by -20 to -30% after soldering, cleaning, PCB soldering then cleaning again. We first thought it was the cleaning process that left some flux residual, but not.

Still under investigation...

--
Thanks,
Fred.

Normally a 1% resistor is supposed to stay within 1%. But probably it is like with medicine, manufacturers have to protect against litigation. If one guy went to a pub after taking a pill, filled up to the gills and then dropped dead it must say under side effects "can cause death" ;-)

This datasheet contains some realistic values:

One of the things that should really be avoided with precision resistors is hefty surges. Much depends on how it was trimmed, meander cut versus just one deeper cut and so on. This is also why I don't like precision current resistors being loaded so they go many tens of degrees centigrade above ambient. It can make them drift slowly.

--
Regards, Joerg

http://www.analogconsultants.com/

"gmail" domain blocked because of excessive spam.
Use another domain or send PM.

Did you try cleaning the bottom of those resistors? My experience is that the coating on the resistance element keeps contamination from the element but flux and residue can still reside on the substrate material and affect the resistance. Try mounting them 'face down' so the substrate can be easily cleaned. regards, al

Try this: Remove one of them, scrub it real good, measure resistance. That is likely to result in a big surprise :-)

--
Regards, Joerg

http://www.analogconsultants.com/

"gmail" domain blocked because of excessive spam.
Use another domain or send PM.

Are you kidding? I mean, do you really think it wasn't my first thought? The resistors have been unsoldered, thoroughly cleaned. Same outcome. The big surprise is for you :-) : it definitely *is* the resistor that shifted.

BTW the circuit has somewhere Gohm order impedances with 0201 size SMDs and some other nasty packages which required some pretty carefully cleaning processes and that I sure won't disclose :-)

--
Thanks,
Fred.

Besides the fact that cleaning isn't at fault at all, I suggest you to suggest *your* production people to have some 0201 resistors mounted face down, and see what they say :-)

--
Thanks,
Fred.

Ok, Gohms in 0201 is a highly unusual situation. The next item on my list would then be to send these resistors to a forensic lab. They can (destructively) find out whether any flux or other residue has contaminated the resistor body. Those guys can find miniscule traces. Probably even find out whether the assembly guy smoked Gitanes or Gauloises after lunch :-)

Four out of 250 sound like you got some bad apples in the shipment though. If not then this might require switching the brand or hand soldering. No fun with 0201.

--
Regards, Joerg

http://www.analogconsultants.com/

"gmail" domain blocked because of excessive spam.
Use another domain or send PM.

Before, or after they send him through the reflow oven?

--
You can't fix stupid. You can't even put a Band-Aid? on it, because it's
Teflon coated.

That's not a tolerance drift- it's a failure. Solution: dump that manufacturer.

Nope- the IEC recommendation is to tighten up the so-called 'production instability', or mean deviation of the product of the manufacturing process so that the deviations under testing work as advertized. The 1% resistors should have a production deviation of

0.1% or less.

Since when is an expectation a standard deviation?

IEC publication 115 and its various versions outline the methodology of recommended testing- I don't think there's a whole lot of guardbanding in the resistor industry...and there is no way of knowing from the specifications the worst case deviation under application of simultaneous environemnetal stress conditions. Some of these are cancelative, for example, the high humidity test causes resistance to decrease whereas high temperature causes an increase- but the solder stress with really high temperature is more of a recovery from a deformation.

My formulation skated over a number of implicit assumptions. The biggy is that deviations are expected to show a Gaussian distribution, and resistors have been known to show very non-Gaussian distributions, when the parts you buy have been selected out of the more or less Gaussian distribution that came out of the production process.

None the less, the first cut at any question about the total effect of the accumulation of a number of independent imperfections is going to be to assume that the various imperfections show a Gaussian distribution, and the way of going from there to the distribution of the accumulated errors is to estimate the standard deviation of the final distribution as the root-mean square sum of the standard deviations of the various contributing errors.

Once you've got a number for the standard deviation of the accumulated errors, and are prepared to assume the the distribution is still Gaussian, you've got an expectation about what proportion of the parts will deviate from the mean by more than some multple of the standard deviation.

-- Bill Sloman, Nijmegen

t

al

ld

o ,

The power of the Gaussian distribution as far as most statistical sampling is concerned derives from the central limit theorem, an important result which as applied to sample means shows that ,regardless of the distribution of the deviation of an individual resistor, the mean of a large enough random sample set of those resistors, itself regarded as a random variable, will approach a Guassian distribution. So knowing the sample mean and its standard deviation together with the fact that it is negligibly close to being Gaussian, you can establish a confidence interval that every resistor deviation lies within an interval about the sample mean with such and such probability. No assumptions are necessary. There is a little empirical testing required to arrive at an estimate of the variablity of the data to try to establish the size of the sample sets to begin with, but the whole process is sort of self-validating. Since each environmental or handling stress produces a mean deviation as well as std deviation, I don't see where you can use the manufacturers' data to predict a worst case confidence interval under conditions of multiple stresses.

a

ort

tial

en

ould

s s

to

ed

ts

ns,

or

.

It's true that - because the stresses can interact - the processes that occur when a part is subjected to multiple stresses aren't the same as those you are looking at when investigating a part's response to any one of these stresses, so the estimates that you make by effectively assuming independence can wrong - and sometimes wildly wrong.

But the assumption of independence does give you a place to start.

-- Bill Sloman, Nijmegen