Issue link: https://iconnect007.uberflip.com/i/1210212
FEBRUARY 2020 I DESIGN007 MAGAZINE 59 realm of mechanical engineers. You need to have a mechanical engineer who is doing more work for this. Very few electrical engineers can use MCAD tools. Electrical engineers like simplified models with "thermal resistances" and "thermal volt- ages." This way, we can use our known elec- trical simulation tools. The results are only rough approximations, but so far, this method has been quite successful. You get a rough es- timate on what will happen, and you use pro- totype measurements to verify your model. That's far away from detailed thermal simula- tions using ANSYS tools. Shaughnessy: It's cheaper to build it and model it. Prototype it and see what happens. Löwer: That's what I think. If you do not have very specific requirements, it's easier that way. For detailed simulations, you have to provide good models and detailed data on electrical power consumption. Most engineers will tell you, "We have a good estimate on how much power we will have from this ASIC, IC, etc.," but it's an educated guess—especially when using large FPGA or microcontrollers because you use average data for initial system design. You can run with the old models without too much trouble—as long as you stay in the mod- el's limitations. And with the first prototype and a more evolved software, given the appro- priate time, you have the chance to reduce the component costs for power supplies and cool- ing solutions, or you make a cost reduction in the next product evolution step. Shaughnessy: So, it's not to a critical point yet. They don't need to use the thermal manage- ment tools because what they're doing works, more or less. You build a model and do a pro- totype. Löwer: Yes. It's faster and cheaper—as long as you do not run into unforeseen issues and sur- prises. As mentioned, it depends on how well you know your product, and whether or not you can see thermal issues beforehand. Usu- ally, companies know their product range and try to avoid the—from their point of view— overhead of extensive simulations as long as possible. However, this changes as soon as there are good reasons for simulations. Signal integrity is a good example of this change. There, you need simulations for high-speed systems. That was not an issue until people started build- ing DDR memories or using ultra-high-speed SerDes with multiple lines. When the first seri- al I/O systems from Xilinx and Altera came on, then people understood. But until then, LVDS was the worst that could happen, and every- body made sure that they had to be careful and tiptoe around it, but it's okay and stable. They had no need to do the simulations. If high-speed requirements reach a point where we have to take thermal issues into con- sideration, this will change as well. But aside from vacuum applications or high-power den- sities, I have not seen thermal simulations on a regular basis. Matties: The other thing that Happy mentioned was a design for a robot or automated assem- bly. Is that a consideration that you conscious- ly make when you're designing a board? When someone's coming in, do you go to the assem- bler, and do you collaborate in advance? Löwer: It depends on the customer. Remem- ber, I come into a company and hope that they have something like a design guide that I can read. That's not always the case, or it is not as detailed as I want it to be. Often, I have to talk to the engineers and ask what they are doing to get a feel for their way of design- ing. Design philosophies in different compa- nies are quite different, and you can feel this as an engineer. If you are in a bigger company, especially in automotive, you have pretty strict design guides, and they take care of all kinds of pro- duction issues. If you're doing 10 million de- vices for VW, for example, then you have to make sure that you know how to do those. They have special review boards where mul- tiple people sit in a room, checking that each resistor is in the right alignment and that, for