Issue link: https://iconnect007.uberflip.com/i/1285883
SEPTEMBER 2020 I DESIGN007 MAGAZINE 15 and I was always able to get some advancement on those things for the tools that I was using. They're out there to make a profit, and they will change wherever enough people are asking them to go. The push needs to be back on the design teams to ask for their specific needs. Johannes Adam and I have been friends since about 2002, and I've used a lot of tools for solv- ing various thermal problems. His tool is one of the best I've used for trace heating. However, it is just focused on the circuit boards, where- as a lot of the bigger tools are focused more on the system level, and then they may or may not have features that will address the board level. It takes those larger companies time, and pres- sure from the people buying their tools, to in- corporate the different features that are needed to solve the various thermal issues that come up in designs. Overall, the EDA industry has some awesome tools. I envision fully automated artifi- cial intelligence features in the future EDA tools. Shaughnessy: Other than using the completely manual way, putting down heat sinks and fans, what advice would you give somebody who is new to doing thermal design? Jouppi: Understand your design from the funda- mental side. I use a control volume and spend time defining all of the sources of power into the board and the heat transfer paths into and out of the board. Model those powers as accu- rately as possible to represent the physical de- sign. I talked to some people who account for the power losses from their traces and asked how they model those losses. They would smear it over the entire board in their thermal model, and that's at least one step closer, but you could possibly be missing a hotspot. It's important to understand the compo- nent thermal resistances and associate those numbers to your hardware. There's so much to gain from looking at the hardware and how it's mounted, as well as knowing what the lead materials are, the thermal resistance through those leads, how it is mounted to the board, and what the gap between the component and the board might be. Do all your homework by reading papers that people have written, fully understanding every aspect of the design. Then perform thermal testing to validate your mod- els. There is a lot to learn from testing. That is a topic in itself. Shaughnessy: As always, communication is king. If you have any thermal problems, then your designers should talk to the CAM person. Jouppi: We're always coordinating with the CAM person, but the problem is that the whole team is under a time crunch. Early temperature predictions help to minimize changes to the layout and keep the design team on schedule. Shaughnessy: IPC-2152 is pretty central to all of that. Jouppi: The idea is central to it, but IPC-2152 isn't used the way I hoped it would be used. I wanted to see people use IPC-2152 as a base- line configuration and used to create trace heat- ing charts for their own specific design config- urations. The nice thing about IPC-2152 is that it's a well-defined set of numbers that were put together through testing and documentation. There's a large appendix that has a lot of in- formation about the testing we did, as well as the history of the previous charts that were last published in IPC-2221. When IPC-2152 was put together, we had a long list of items to investigate. We prioritized that list and worked through what we could. The appendix was developed so that we could add to the document over time. The main area was trace heating for internal and external trac- es, based on temperature measurements fol- lowing IPC-TM-650 2.5.4.1a. We also wanted to look at the impact of cop- per planes on trace temperatures and illustrate the temperature change that occurs when you introduce a copper plane and how the distance from the trace to the copper plane would im- pact the trace temperature. In addition, we wanted to address working with vias and how much current you can put through vias and via patterns. Many people still use rules of thumb to say one amp per via or two amps per via, but I wanted to see us do