Below are the questions asked during the presentation, along with their respective answers.
Q: I was wondering why 200C was chosen. Isn’t that way too high?
A: It is high for electronics generally but this was a test harness where we needed to have a significant amount of heat conduction for any Copper distribution. Neither the Copper nor FR4 have conductivities that are temperature dependent so the choice of 200C didn’t influence the results.
Q: Why is there such a difference between the analytical and substitute layer method?
A: The analytical method doesn’t capture the reality that while there may be a significant amount of Copper much of it doesn’t provide a continuous conduction path. If the Copper distribution is highly connected then the temperature predictions between the two methods would be more similar.
Q: Were there any experimental comparisons made?
A: Yes, we validated the approach against two sets of measurement results.
Q: If we assume the PCB as a single cuboid, what would you recommend for in-plane and through-plane thermal conductivities? Assuming we have 10-20 layers.
A: For typical PCB thicknesses and % conductor, an In-Plane thermal conductivity of 16 W/mK and an Axial Thermal Conductivity of .83 W/mK would be reasonable assumptions.
Q: How do you decide which PCB modeling approach to take between compact, detailed, and explicit?
A: Many times the decision comes down to what information is available. Another consideration is the detail associated with the IC package model. For instance, I wouldn’t recommend modeling explicit PCB Copper if the components are modeled as blocks with heat. Another consideration is the amount of heat conducted through the PCB. If the critical IC has a large heatsink mounted on top then the PCB doesn’t make much difference in the accuracy of the prediction. As a general rule PCBs should be modeled as detailed, where each layer is modeled individually. One step beyond that is to subdivide the signal layers to capture areas of high or low Copper density. For power and ground layers it is usually acceptable to go with a % Copper area weighed conductivity.
Q: How to obtain substitute layer thermal conductivity values using Flotherm XT?
A: In FloTHERM XT the choices for effective PCB thermal conductivity are Analytical and Empirical. Empirical refers to the Substitute Layer Method. Modeling explicit traces with Thermal Territories is also an option.
Q: Can you share a link to the original work?
A: The Substitute Layer Method is patent pending and the original work hasn’t been released yet.
Q: Do you have this in a paper format?
A: The Substitute Layer Method has only been discussed in online presentations.
Q: Are there industry standards for determining Keff of a PCB?
A: I’m not aware of any standards.
Q: What method do you use to calculate the % of Copper per layer?
A: In FloTHERM XT it is based in on volume fraction of the physical CAD parts. If only images of the PCB layout are available the % Copper could be determined by comparing the number of pixels representing Copper to the total number of pixels.
Q: Did you compare the different methods against physical testing? Or is it only between simulations?
A: Yes, the methodology was validated against two sets of customer measurement results.
Q: Were actual measurements made on the board and devices to compare results with explicit, analytical and substitute models?
A: Measurement results were compared against the Analytical and Substitute Model approaches. The Substitute Model approach compared much more favorably to the measurement results as compared to the Analytical approach, which significantly underestimated temperatures.
Q: Can you give a resource for subdividing PCB in cases where the Copper is not uniform?
A: There are some options within FloTHERM XT. Generally speaking the strategy would be to either subdivide the PCB in a uniform set of NxM subdivisions or to nominate particular areas within the PCB to capture individually.
Q: Is this method more suitable for signal layers?
A: The improvement in accuracy is more notable in signal layers. If there is a lot of Copper in a specific layer and is uniformly distributed then the difference between the Analytical and Substitute Layer approaches are smaller.