Junction temperature estimation 1

[thermdes]

If you dont have any thermal information on a chip and need to estimate the temperature of the junction, how would you do this? Also, I dont have any dimensions of the die or any internal chip info. What I have been doing is estimating a combined thermal conductivity representing the chip and calculation a maximum temperature using a CFD program. Then I use the thermal resistance if I have it available. If not, I use thermal resistances of similar packages to estimate junction. Anyone know of a better way?

thermdes

 

[ko99]

Use a reasonably conservative estimate of the thermal resistances of a similar package (j-c and j-b). If you have good margin, you're done.

If not, try harder to get the data from the manufacturer -- you may have to get your VP to call their VP. Or my preference, if possible, is to get a sample and cut it in half.

ko (

http://www.ecooling.biz)

 

[sailoday28]

What information, other than computer programs, do you have?

 

[IRstuff]

Don't you have a datasheet? There is usually sufficient info on a datasheet to get sufficiently accurate estimates

TTFN

 

[mecheng66]

Not so, IRstuff. The datasheets might give a junction-to-case and but rarely specify junction-to-board, or die size, or package details, or anything else you need to know to figure out what's really going on.

 

[ko99]

Yes, some so-called "datasheets" are incredibly useless when it comes to thermal data. But don't get me started...

To be fair, die and package construction details are proprietary and the trend is slowly going in the right direction: more manufacturers are now providing j-b and some even provide compact models.

ko (

http://www.ecooling.biz)

 

[IRstuff]

You should be able to get a decent estimate with junction to case or junction to ambient for the estimate that the OP is trying to get

TTFN

 

[ko99]

Yes, j-a is useful in lieu of a model IF your application is similar to their test setup. And, yes, you can ignore j-b and just use j-c in a model IF the dominant thermal path is through the top.

For all other cases, you need more data. The thermal path to the circuit board is very application-specific, anywhere from 5 to 95% of the heat can take that path.

That reminds my of another way to estimate j-b:
If the supplier's test setup is defined, including pcb copper content, you can model their test and vary j-b until your model matches the j-a test data.

btw, what's an OP?

ko (

http://www.ecooling.biz)

 

[IRstuff]

original poster

TTFN

 

[thermdes]

Thanks for all that answered my questions...but I am still at a loss to accurately estimate junction temperature... are there any papers you may be aware of to do this in tandem with a CFD package? I read some info on the DELPHI models but its not easy to boil it down and use it if you are not using Flopak. Any suggestions?

thermdes

 

[IRstuff]

Just what do you mean by "accurately?" and whydo you need to do this? The variability of assembly is such that any analysis that is "accurate" is probably inaccurate.

TTFN

 

[thermdes]

Many times we make product decisions on the junction temperature. As you know, the reliability hangs on this number. Our products must be reliable. That is why I think it should be accurate.

 

[IRstuff]

??? Only if you have total control over the usage of your product.

Typically, we ahve a -30ºC to +50ºC temperature range of usage with nary a iota of data on actual temperatures and actual usage and actual duty cycles. That results in something like a 10⁴range on lifetime. A 10ºC error in junction temperature in your model would harly make a dent.

TTFN

 

[ko99]

Yes, it's virtually impossible to predict the relation between a die temperature and the end-product's actual MTBF.

A better way to make design decisions is to keep all dice below their maximum operating temperature when your product is operating in it's specified worst-case environment.

Thermdes, there are a wealth of papers on using CFD to model IC packages. Browse through these sites

http://www.hitech-projects.com/euprojects/profit/publications.htm

http://www.flomerics.com/flotherm/technical_papers/

http://www.electronics-cooling.com/html/articles.html

Still, the bottom line is if you don't know the internal construction of a package and do not have a detailed or DELPHI model, then you cannot "accurately estimate junction temperature". Even if you have Flopack, you must either enter the precise details or use estimates based on similar packages...

When you need more accuracy, the only choices I'm aware of are what I already mentioned, all of which have their own set of pitfalls.
1. Get the details from the supplier
2. Take the package apart
3. If the supplier has j-a test data, simulate the supplier's test and adjust your package model to match their test at several points

ko (

http://www.ecooling.biz)