When does temperature coefficient matter?

[(In)Sanity]

So I've been doing a whole lot of temperature coefficient checking lately due to a questionable batch of resistors.   Well it got me thinking about why am I so concerned, and when does it really matter ?   

I started to compile a list of areas where low quality parts might not make a difference.   I encourage feedback on this list and would love for more items to add to the list. 

In the case of a resistor,  when does temperature coefficient not really mater ? 

1.   For a pull up resistor.
2.   For a pull down resistor.
3.   For a current limit on a simple charge circuit.
4.   When used with a basic audio amp circuit.
5.   When used as a bias or current limit on a switching circuit.  Such as BJT drive or FET limit.
6.   When used as a half divider,  assuming both resistors drift equally and in the same direction (Crossed this one off for being impractical)
7.   For an LED voltage drop resistor.

When does temperature coefficient matter ?

1.   In a divider that's not doing exactly half.
2.   In an op-amp circuit where the resistor is controlling gain and it needs to be accurate.
3.   In a current sense device where accuracy matters. 
4.   In a DAC type circuit R2R ladder,  etc.  This really goes right back to #1.

Perhaps I'm just tired while writing this,  but I'm finding it hard to think of cases where temperature coefficient of a resistor counts to a large degree outside of unequal dividers,current sense circuits and precision amps.   Situations where an exact current needs to be drawn from a divider might also be one of them.   Feedback loops really fall back in to that divider area (#1).   

I think it's beneficial for everyone to understand where they should and shouldn't skimp on parts.   I'm very confident others will have a ton of items that don't fall in to #1,2 or 3.   

Thanks,

Jeff

Edit:

Not sure what I was thinking.   If the TC of the resistors in a divider are all equal and the sign of the TC is equal for all of them then in theory the divider won't produce error caused by the TC.   This is not at all likely to occur.   

[Jay_Diddy_B]

Hi Jeff,

There is nothing really magic about a divider whose division is ration is 0.5. Tempco still matters in this case.

The reason is that tempco is stated without reference to a the sign. Two 100 ppm resistor can drift with temperature in opposite directions.

The manufacturers aim to get a tempco of zero mixing various components. If the mixture was very uniform then all the resistors would have a low temperature coefficient. If the mixture is not uniform there will be a spread in tempco between the parts.

The other consideration is operating temperature. If the unit does not experience wide temperature variation, it may not be so important.


Jay_Diddy_B

[(In)Sanity]

Hi Jeff,

There is nothing really magic about a divider whose division is ration is 0.5. Tempco still matters in this case.

The reason is that tempco is stated without reference to a the sign. Two 100 ppm resistor can drift with temperature in opposite directions.

The manufacturers aim to get a tempco of zero mixing various components. If the mixture was very uniform then all the resistors would have a low temperature coefficient. If the mixture is not uniform there will be a spread in tempco between the parts.

The other consideration is operating temperature. If the unit does not experience wide temperature variation, it may not be so important.


Jay_Diddy_B

True,  I didn't really think about real world manufacturing processes.   So we can cross that one off the list for anything outside of hobby projects.   

Jeff

[alm]

Any differential circuit, since CMRR directly depends matching the gain for the positive and negative part. TC tracking is more important in this (and many other) cases, but how do you ensure tracking if they only specify a max TC? You also need to make sure they're at the same temperature.

A bad TC can even introduce distortion in some amplifier designs: more current flows at the peaks of a sine, this might heat a resistor, change its value, affecting the gain. This produces harmonic distortion.

[(In)Sanity]

Any differential circuit, since CMRR directly depends matching the gain for the positive and negative part. TC tracking is more important in this (and many other) cases, but how do you ensure tracking if they only specify a max TC? You also need to make sure they're at the same temperature.

A bad TC can even introduce distortion in some amplifier designs: more current flows at the peaks of a sine, this might heat a resistor, change its value, affecting the gain. This produces harmonic distortion.

Alm,  yah that was my #4 on the first list.  "Basic Audio amp".   Without a doubt for anything of quality it needs to have good TC. 

Thanks,

Jeff

[Mechatrommer]

imho... no buddy (and i swear )... there is only one simple answer to this, it will answer all of your questions, we can make an infinite length of talk but it will comes down to one simple thing.... tempco matters when you are doing precision instrument. how precise? that is you to define it. precise i mean accurate, low drift, repeatable etc anything you can find in the book. cheers.

[(In)Sanity]

imho... no buddy (and i swear )... there is only one simple answer to this, it will answer all of your questions, we can make an infinite length of talk but it will comes down to one simple thing.... tempco matters when you are doing precision instrument. how precise? that is you to define it. precise i mean accurate, low drift, repeatable etc anything you can find in the book. cheers.

Yes,  but where exactly in that high precision circuit ?   Are you going to use a 0.5ppm TC resistor to drive a status LED?   I think it goes without saying that dividers (in any form) should use high quality low TC components,  I also think they should be of the same sign and perhaps depending on the situation be the opposite of other components in the design.   For example countering the negative drift of silicon with positive drift from another component..etc.   Temperature compensation is another area I would love to touch upon at some point.   

Perhaps it would be easier to list areas where TC doesn't really matter.   

Thanks,

Jeff

[c4757p]

In my opinion, these sort of questions are symptomatic of the wrong approach to engineering. Does it matter? Yes. Everything always matters. The question is "how much?". Before you design your circuit you should come up with a plan for its capabilities and tolerances, and if you're wondering if the tempco of something is relevant, you need to calculate how much it affects the circuit, and whether that puts you inside or outside your specifications. A quick order-of-magnitude calculation in your head will suffice when you have a huge margin, like pullup resistors.

Binary "does it matter" questions lead to fantastic overengineering where it's not needed, limited by blind underengineering in the rest of the circuit.

[(In)Sanity]

In my opinion, these sort of questions are symptomatic of the wrong approach to engineering. Does it matter? Yes. Everything always matters. The question is "how much?". Before you design your circuit you should come up with a plan for its capabilities and tolerances, and if you're wondering if the tempco of something is relevant, you need to calculate how much it affects the circuit, and whether that puts you inside or outside your specifications. A quick order-of-magnitude calculation in your head will suffice when you have a huge margin, like pullup resistors.

Binary "does it matter" questions lead to fantastic overengineering where it's not needed, limited by blind underengineering in the rest of the circuit.

For the most part I already know the areas where it matters and where it doesn't.   I thought it might be an interesting exercise to narrow down and list out where it "really matters".   If your new to the game you might ask questions such as does the TC of a pullup resistor really matter ?   Well,  maybe it does.  Perhaps it will decrease in resistance which causes heating in another component which sits next to something else and causes drift.    So your correct,  it always matters

[c4757p]

I think a better way to phrase it would be "When should temperature coefficient matter?", i.e., if your design requires precision low-tempco pullup resistors, don't buy precision low-tempco resistors - redesign your circuit. I'm sure somebody would find a way to make a circuit where it matters and be damn confused because he was told it doesn't 

[(In)Sanity]

I think a better way to phrase it would be "When should temperature coefficient matter?", i.e., if your design requires precision low-tempco pullup resistors, don't buy precision low-tempco resistors - redesign your circuit. I'm sure somebody would find a way to make a circuit where it matters and be damn confused because he was told it doesn't 

Good point,  so perhaps this is a dead thread.   I'm always looking for those weird exceptions where two 1000ppm resistors can form a highly precise divider that doesn't drift at all.   Too much theory and not enough practice.   I just starting to touch on temperature compensation on some of the projects I'm working on.  My problem is I always want to just fix it in software

Thanks,

Jeff

[ftransform]

I am developing a negative association to resistor after spending so much time with those resistors. It's like eating Thanksgiving turkey for 5 days in a row.

We must find another mostly overlooked passive component property to nitpick now, at least for a while. Any suggestions?
Perhaps inductor tempco? noise? Microphonics studies? Giving up and playing with a arduino?

I feel like this forum recently went through a really STRONG ESD phase, I don't want to resurrect that one quite yet.

[alm]

You can spend much more time on capacitors. All kinds of fun parasitic effects to talk about: ESR, ESL, voltage coefficient, dielectric absorption, microphonics, aging and even tempco .

Transformers and inductors could be another fun topic, but is mostly limited to power supplies these days.

[(In)Sanity]

You can spend much more time on capacitors. All kinds of fun parasitic effects to talk about: ESR, ESL, voltage coefficient, dielectric absorption, microphonics, aging and even tempco .

Transformers and inductors could be another fun topic, but is mostly limited to power supplies these days.

Let's not forget leakage current

Great now I'll be messing with the LCR meter and my little environmental test chamber.   

Actually I just tested a  Fairchild LM78L05.   74ppm/C or 0.37mv/C drift.    I think temperature compensation is a better subject,  it's kind of cops and robbers.   So in theory I should be able to get this 78L05 down to very little drift with the correct reverse TC components adding in to the mix.   That of course opens up the can of worms called linearity.   But it's all fun.

Jeff

[ftransform]

Oh that's an interesting idea I wonder if you can use a resistor to fix that drift. That would actually be pretty neat.

I wonder what photonicinduction thinks of this topic.

[PA4TIM]

On Circuits online we gad a topic frim someone who tried to get the best out of a 7805 and he came pretty far ( not like a Vref but still much further as I had expected)

Capacitors are fun, inductors even more. They have tempco, harmonic distortion, Q, saturation, current cooficient, eddy currents, voltage coofficient, skineffect, Ohms serie resistans, formfactor, distributed "capacitance" ( real capacitance and Phase delay) , parallel resistance. Are very influenced by the surrounding materials, selfresonant ( unwanted or just planned) and for most, very hard to measure.

[chickenHeadKnob]

I am developing a negative association to resistor after spending so much time with those resistors. It's like eating Thanksgiving turkey for 5 days in a row.


I feel like this forum recently went through a really STRONG ESD phase, I don't want to resurrect that one quite yet.

Amen brother, actually there is a deeper issue with this forum insidiously inducing volt-nuttery in hapless bystanders such as myself. I thought I was immune but I find myself cruising ebay for precision  references and bench DMM's more and more. What the hell happened to me! it is like a crystal meth addiction. Friends don't let friends volt-nut.

[PA4TIM]

It can be worse, i'm a time-nut too.
Besides volt and components nut.

[hans]

A little sidejump to the original question. The answer is pretty simple.

Compute the gain formula for your 'blackbox' analog circuit. Have an input, have an output, and it's relationship.

All values from components in that formula have a temperature coefficient. Some have bigger influence than others. Therefore, if you requirement is 0.1%, a influence of 1/10000 drift doesn't matter, but 1/500 drift does matter.
With temperature coefficients, operating range, and initial accuracy + trimming/compensation, you can work out how precise your system is.

Good luck writing a shitload of math down 

[Harvs]

There's a pretty good explanation of an error budget in the Chap 7 precision circuit, of The Art of Electronics which most people seem to have.

You can pretty much wrap most of these discussions up into that type of analysis.

[Mechatrommer]

When does temperature coefficient matter ?

1. In a divider that's not doing exactly half.

i want to trigger (output hi) the comparator at slightly +ve voltage. 0.1, 0.2,0.3v doesnt matter as long as it doesnt trigger at ambiguosly 0V, so i make a voltage divider something like 100K+1K. tempco? doesnt matter!

2.   In an op-amp circuit where the resistor is controlling gain and it needs to be accurate.

configuration usually found or same as in...

4.   When used with a basic audio amp circuit.

doesnt matter! please note i've striked the accuracy part of the quote.

where in one situation it doesnt matter, in another situation, it does matter. its quite difficult to classify. i agree though some app, are almost certainly doesnt matter such as digital pull-up, current limit protection, lighting up LED, capacitor drain etc. the bottom line is, as long as you are not going to specify its accuracy, it doesnt matter.

[ftransform]

It can be worse, i'm a time-nut too.
Besides volt and components nut.

at least you don't collect human ears 

[(In)Sanity]

I am developing a negative association to resistor after spending so much time with those resistors. It's like eating Thanksgiving turkey for 5 days in a row.


I feel like this forum recently went through a really STRONG ESD phase, I don't want to resurrect that one quite yet.

Amen brother, actually there is a deeper issue with this forum insidiously inducing volt-nuttery in hapless bystanders such as myself. I thought I was immune but I find myself cruising ebay for precision  references and bench DMM's more and more. What the hell happened to me! it is like a crystal meth addiction. Friends don't let friends volt-nut.

I second the Amen.   While it's good to be precise,  perhaps being productive is slightly more important.   

Maybe I should have titled this topic as "When can I get away with using low grade crap without the boss noticing?".    I also completely botched the whole divider statements.   I was too focused on resistors in this case that basic logic escaped me.    Two resistors equal temp co,  same direction...yah those will make a great divider.   Too bad you can't use that concept in production.   

[krenzo]

This is a bit off topic, but as an ex-nuclear guy and seeing "when does temperature coefficient matter", I've just been dying to point out something like this.

[(In)Sanity]

This is a bit off topic, but as an ex-nuclear guy and seeing "when does temperature coefficient matter", I've just been dying to point out something like this.

Umm,  point taken.

[ftransform]

russian resistor tempco must be pretty bad

[JoeyP]

You can spend much more time on capacitors. All kinds of fun parasitic effects to talk about: ESR, ESL, voltage coefficient, dielectric absorption, microphonics, aging and even tempco .

OK, I'll bite The voltage coefficient of large ceramic caps is the one I find most amazing. If you use them anywhere near their rated voltage, they practically disappear. Take a look at this chart for a 4.7uF 50v X5R cap. Notice that at 50v its value has been reduce by more than 90%. It makes a better varactor than a capacitor!

[PA4TIM]

Wow, i knew X5R had a rather large VC not this extreme.  Upto now I never measured this. I was plannimg because I used a X5R 50V 470 nF in a boostconverter casacde for a pulser. I made the same one, using the same coil before. The old one died while exerimenting so I reused the coil.  The new one i made in smd. Instead of MKT 470 nF cps I now used 470nF X5R caps. The first trew out 130V, the new one stuck at 81V. ( changing the feedback solved this but it was strange)

Now seeing your graph it makes sense. I use them very near to their max voltge rating.
My bridges can work with a DC bias. So that gives inspiration for some nice tests.

[Jay_Diddy_B]

Hi,
That graph for the capacitor variation with applied voltage does not look the graph for X5R. It looks like a graph for Y5V or Z5U.

Jay_Diddy_B

[PA4TIM]

http://www.niccomp.com/help/VoltageCoefficientofCapacitors-032012-R1.pdf
More info about VC including that same graph

[Neilm]

The VCR of a cap depends on how the cap has been constructed, as well as the dielectric. I have seen and used caps with X5R dielectric that still had 80% of rated capacitance at full voltage (3kV)

Neil

[JoeyP]

Hi,
That graph for the capacitor variation with applied voltage does not look the graph for X5R. It looks like a graph for Y5V or Z5U.

Jay_Diddy_B

The data sheet is attached below. Take a look for yourself.

Here are some actual measured values for a couple of parts. Lower value parts seem not to be as bad, but still very significant. I looked at different brands as well and found similar results:

Taiyo Yuden 4.7uF 50V X5R 1206 (UMK316 BJ475KL-T)

0v: 4.11uF
10v: 2.56uF
20v: 1.63uF
30v: 1.10uF
40v: 0.78uF
50v: 0.61uF !!!

Kemet 1uF 10V X5R 0603 (C0603C105K8PAC7867)

0v:0.891uF
1v: 0.887uF
2v: 0.854uF
3v: 0.804uF
4v: 0.740uF
5v: 0.675uF
6v: 0.611uF
7v: 0.545uF
8v: 0.495uF
9v: 0.443uF
10v: 0.400uF

[ftransform]

I wonder if this is easy to test using 555 timer (at different voltages) and performing frequency measurements.
i have not played with one in a while but iirc their frequency output varies quite a bit with voltage.

What kind of impact does this voltage dependant capacitance have on decoupling supplies? I figure the 100nF decoupling caps commonly used are not so bad, but how about more extreme cases?
Take the http://www.ti.com/lit/ds/symlink/tps7a3301.pdf for example. They say one should use 47uF ceramic decoupling caps for the output. Maybe this works OK when the unit is outputting 2 volts, but what if I want a -30 volt supply? Should I rethink their use of ceramic capacitors?

I'm afraid the designers did not take this into consideration? They recommend using high value X5R type capacitors. If you look at the spectral noise density graphs over a voltage range (page 9) you can see a increase in 2x over a 5 volt increase in output voltage. Perhaps this is because the ceramic capacitor is crapping out at this point?

[JoeyP]

The January 2013 EDN has an article which discusses ceramics used in capacitors. The author points out another interesting point about this voltage dependence, which is that it is mostly independent of the voltage rating of the cap. It's much more dependent upon the physical size of the cap. He uses an example of two different 4.7uF caps one rated at 16v, one rated at 25v. When biased to 12v, the value of either cap ends up being about the same.

[marshallh]

Here's some graphs on how package size alone will affect effective capacitance...

http://www.maximintegrated.com/app-notes/index.mvp/id/5527

[JoeyP]

Here's some graphs on how package size alone will affect effective capacitance...

http://www.maximintegrated.com/app-notes/index.mvp/id/5527

That looks like the text of the same EDN article. I see that it indicates it also ran in EDN in November 2012. They must be desperate for material to run it again so soon.

[Bored@Work]

That looks like the text of the same EDN article. I see that it indicates it also ran in EDN in November 2012. They must be desperate for material to run it again so soon.

Or EDN got paid a second time for running it.