Some time back we were testing a flip-chip package in a top-cold-plate harness to measure the package junction-to-case thermal resistance (ΘJC, RJC). We performed a number of tests where we were consistently measuring a higher value of thermal resistance than was reported by the vendor. How could we and the vendor both be correct?!?
To understand this further I built an analysis model to study the influence of the TIM 2 (thermal interface material) thermal resistance on the overall RJC prediction. For those of us that can’t remember where TIM 2 resides in the overall stack-up refer to my artistic representation below.
The analysis model varied the TIM 2 thermal conductivity from 0.01 to 1000 W/mK, essentially from “very poor” to “unrealistic”. The charts below show the results of the analysis, where the second chart is a zoom-in of the overall results. From the results we can see that once the TIM 2 thermal resistance reaches about 1 °C/W the predicted RJC is independent of TIM 2. The predicted RJC value levels out at about 0.8 °C/W. Below the value of 1 °C/W we are seeing the effect of heat spreading through the lid of the BGA.
After studying the results I still wonder what would be the best value to report for RJC ? Since I have a bit of Electronics Cooling Design experience allow me to opine. Thermally speaking the flip-chip BGA does not pop up unless there is a significant amount of heat generated, let’s say greater that 10 W. It is also pretty safe to say that the thermal designer is going to draw the heat out the top of the package. For me, a flip-chip BGA without a heat sink is , shall we say, underdressed. For a high power application it simply won’t be possible to adequately cool the device with an interface material in the 1 °C/W range. Because of this, if I were the BGA vendor, I would not report this high end value though it is the most prudent value.
Let’s look at the opposite end of the spectrum where we have an idealistic TIM resistance. If the BGA vendor is characterizing the thermal performance of the package in an analysis tool why not perform the RJC calculation with no TIM 2 at all? It isn’t controlled by the vendor so why not allow there to be perfect thermal contact between the BGA and the cold-plate? The reason is illustrated in the chart above. With a perfect contact assumption the RJC will be predicted at a value that the vendor’s customer would never be able to achieve.
I guess if it were up to me I would use a TIM 2 resistance that was consistent with the high performance TIM materials currently available on the market. That is the best they can do. The truth is every situation will be different because it is dependent on the TIM as well as the performance of the heat sink.
There are a number of articles written about predicting the effect of heat spreading as related to electronics cooling but here are a couple that I found particularly useful:
Electronics Cooling Magazine-Seri Lee
Electronics Cooling Magazine-Clemens J.M. Lasance
Also, this is a great online calculator for heat spreading
MHTL Simulation Tools