Temperature Rise - ICs

Thread Starter

toffee_pie

Joined Oct 31, 2009
233
Hi Guys,

Is there an easy way to extrapolate temp rise from an IC if you know its power dissipation? I am looking into a new IC and want to predict what the overall temperature rise will be Vs the previous component - (hopefully much lower temp!)

say component A - Power dissipation 1.7W and component B - Power dissipation .96W - is there a quick way to gauge the peak temperature rise?

thanks.
 

ebp

Joined Feb 8, 2018
2,332
There will be a "thermal resistance" specification for the package, usually expressed as Kelvins per watt. Depending on the device, there may be multiple specs - die to package, package to free air, package to [some particular arrangement of copper & pc board], die to ..., etc. Some parts will have curves for temperature rise but that isn't common for low-power ICs that don't actually control power in some fashion.
 

ebeowulf17

Joined Aug 12, 2014
3,291
There will be a "thermal resistance" specification for the package, usually expressed as Kelvins per watt. Depending on the device, there may be multiple specs - die to package, package to free air, package to [some particular arrangement of copper & pc board], die to ..., etc. Some parts will have curves for temperature rise but that isn't common for low-power ICs that don't actually control power in some fashion.
Yep, and even after combining the thermal resistance figures with the dissipation figures, you still have to take into account the ambient temp surrounding the IC (unless it's in free, open air at room temperature.) If the air is trapped in such a way as to warm up as the circuit operates, or if the ambient conditions are already starting out hot, that will affect the temperature rise.

EDIT: On second thought, I might be going overboard here. If you're only trying to gauge the relative difference between two devices, maybe ambient conditions affect both ICs more or less the same way and cancel out. I'm not sure. I'd have to think more about that and probably do some math (sounds hard and I'm on a bus, so I'll pass!)

But if you want to come up with actual numbers for how hot a component is getting, you definitely can't ignore ambient temperature.
 

ebp

Joined Feb 8, 2018
2,332
No, I don't think you're going overboard. Thermal resistance predicts temperature rise above ambient and it is the actual working temperature that counts in predicting reliability. It can be difficult to accurately predict "final" temperature. All the values that appear on the data sheet are based on certain assumptions, and if those assumptions are dubiously valid then the values may not be good predictors.

What? You live in some oppressive place where you aren't allowed to do math on a bus? Shocking! ;)
 

ebeowulf17

Joined Aug 12, 2014
3,291
No, I don't think you're going overboard. Thermal resistance predicts temperature rise above ambient and it is the actual working temperature that counts in predicting reliability. It can be difficult to accurately predict "final" temperature. All the values that appear on the data sheet are based on certain assumptions, and if those assumptions are dubiously valid then the values may not be good predictors.

What? You live in some oppressive place where you aren't allowed to do math on a bus? Shocking! ;)
Ha! Wish I could blame someone else for my laziness!

I haven't *really* done math since high school, and my forays into math for the last 10+ years fall into two categories.

One category is finding a way to make a spreadsheet do the math for me, often taking a brute force approach to finding an answer where a more clever person could've solved an equation for whichever variable they needed and then had a relatively simple formula to give the final answer.

The other category is when I actually try to do the formula solving. I'm awful at it compared to what I once knew. I remember nearly nothing of calculus and I sometimes struggle with even relatively basic polynomial rearrangements. I can't remember all of the tricks and techniques. It's quite frustrating... So when I want to do one of those, I need a lot of scratch paper!

Anyway, in this particular case, if I had all of the datasheet values and the ambient temp value in front of me, I could do a passable comparison of two components pretty easily, maybe even getting a decent ballpark approximation done in my head without paper or electronics... but if I wanted to work out the theory on whether or not certain terms cancel out by solving equations... I'd need lots of paper (and I might still want spreadsheet data to back up my "proof" when I was done!)

Sorry for the rant. I have a love hate relationship with math. The things you can do with math are so wonderful, but for some reason I can't make myself spend the time studying and getting good at it again. Much more fun to work on electronics or coding projects with instant gratification - or at least where you make tangible progress towards gratification. Math homework is too tedious for me, even though I know the reward would be worth it. I'm thinking of taking classes at a local school, just to force myself to do the homework. It feels dumb to pay for classes when I could just study books I already own and do the practice exercises in them, but I just can't seem to make myself do it. Blurgh...
 

WBahn

Joined Mar 31, 2012
26,409
Yep, and even after combining the thermal resistance figures with the dissipation figures, you still have to take into account the ambient temp surrounding the IC (unless it's in free, open air at room temperature.) If the air is trapped in such a way as to warm up as the circuit operates, or if the ambient conditions are already starting out hot, that will affect the temperature rise.

EDIT: On second thought, I might be going overboard here. If you're only trying to gauge the relative difference between two devices, maybe ambient conditions affect both ICs more or less the same way and cancel out. I'm not sure. I'd have to think more about that and probably do some math (sounds hard and I'm on a bus, so I'll pass!)

But if you want to come up with actual numbers for how hot a component is getting, you definitely can't ignore ambient temperature.
No, you can't ignore that even when doing a comparison. If the air has poor exchange properties, then it will heat to a higher temperature with the higher power dissipation. That means that the ambient temperature for the higher power IC ends up at a higher temperature than for the lower power IC.

EDIT: Fix typos.
 
Last edited:

ebp

Joined Feb 8, 2018
2,332
I'm a big believer in spreadsheets. Far too late I adopted a policy of using them even for relatively simple calc's, Invariably, if I do calc's on paper I want to go back and fiddle something then have to do it all the hard way again. I generally do "manual" calc's so I can be confident that my spreadsheet is correct. I use the brute force approach quite often (well, used to).

I am terrible at manual algebra. I have an amazing talent for dropping minus signs or other dumb things that scramble everything, especially for complex numbers. Stupid j. Stupid, stupid j.

returning to the topic at hand ...
One thing that you sometimes run into with going to newer components is that you're forced to use a different package, which makes you do the arithmetic. For example, I used to use microcontrollers in PLCC packages which are now gauche et passé, replaced by quad flatpacks, QFNs and the like. Lots of the new packages have an exposed metal slug on the back and the thermal info in the datasheet assumes a four layer board with a bunch of vias from the slug pad to an inner copper plane.
 

ebeowulf17

Joined Aug 12, 2014
3,291
I'm a big believer in spreadsheets. Far too late I adopted a policy of using them even for relatively simple calc's, Invariably, if I do calc's on paper I want to go back and fiddle something then have to do it all the hard way again. I generally do "manual" calc's so I can be confident that my spreadsheet is correct. I use the brute force approach quite often (well, used to).

I am terrible at manual algebra. I have an amazing talent for dropping minus signs or other dumb things that scramble everything, especially for complex numbers. Stupid j. Stupid, stupid j.

returning to the topic at hand ...
One thing that you sometimes run into with going to newer components is that you're forced to use a different package, which makes you do the arithmetic. For example, I used to use microcontrollers in PLCC packages which are now gauche et passé, replaced by quad flatpacks, QFNs and the like. Lots of the new packages have an exposed metal slug on the back and the thermal info in the datasheet assumes a four layer board with a bunch of vias from the slug pad to an inner copper plane.
Just a few weeks ago I had my first experience thinking about the impact of pads and layers on dissipation. Trying to determine acceptable dissipation levels through a protection diode, I found a variety of thermal specs depending on pad size and layout. Fortunately the other design requirements were loose enough that it was easy to get what I needed with comfortable margins, so I didn't have to fret too much splitting hairs or triple checking tricky math.

I also learned the value of those thermal vias on a design we contracted out about a year ago. Early versions of an LED driver chip board had lots of inexplicable problems, despite the fact that the math said we had a huge safety margin in terms of thermal limits. Even though it *shouldn't* have made any difference, adding a few thermal vias under that big center pad eliminated >90% of our problems!
 

WBahn

Joined Mar 31, 2012
26,409
Early versions of an LED driver chip board had lots of inexplicable problems, despite the fact that the math said we had a huge safety margin in terms of thermal limits.
The problem wasn't the math, it was the model. The thermal model you were using was not the one for the problem you were trying to solve (or, at the very least, it lacked the necessary fidelity). So the correct math gave you a correct solution to a different problem.

Recognizing that we need a different model than the one we've been using or are about to use is a big part of engineering -- and it is not always easy to discern that this is the case until after a lot of failures and failure analysis.
 

Picbuster

Joined Dec 2, 2013
1,038
A number of things are playing a roll:
1:total power consumption
2:thermal resistance between source of heat and body.
3:thermal resistance from body to environment.
4:enviromental temperature and moisture.

Point one and two are calculated and measured by the chip producer
The other points are subject to worry they are hardly controllable in a open environment.
Worst case: outside 65 Celsius and low 10% humidity. save max chip temperature 110 Celsius.
Delta T = 45 Celsius.
This leads to very complicated calculations.
Please read Heat transfer from John R. Howell Robert Siegel Pinar Menguc.
This should give you an impression of the complexity.

Picbuster
 
Top