Issue link: https://iconnect007.uberflip.com/i/1285883
46 DESIGN007 MAGAZINE I SEPTEMBER 2020 es. He was thrilled that someone finally took enough interest in it to come out and talk to him about it. Mike got to look at some of the original data. There were scrap boards that were lying around the office, and that was all the original investigators had. They tested those boards. When they published it, they said, "This is preliminary and uncontrolled. What we real- ly need is more money to do this more cor- rectly." But the money wasn't coming. The first curves were published with the heading that said "preliminary," and the internal curve data was not even experimentally derived. It was just, "The internal traces have to be hot- ter than the external ones, so we'll just derate them by a factor of two." Those are the curves we used for 50 years in the industry. Holden: They're like Moses' stone tablets. It's fighting words if you say that they're not correct. Brooks: We used them for 50 years, and we got away with it because they were so conserva- tive. Now, IPC-2152 is better data, but it is still conservative. I show in one chapter in our book that almost anything you do from a PCB design standpoint lowers the temperature of the trace. If you lengthen it and bring another trace near- by, you could put a plane underneath it—al- most anything you do on a board lowers the temperature of the trace because it improves the thermal conductivity. If you are still bas- ing designs on IPC-2152 data, you're going to be fine. It's just inefficient. You're making the traces too big. Holden: At HP, we designed processors with much higher performance than Intel. All Intel processors are licensed from HP. About four years after HP introduced the internal prod- ucts, they would come out with an Intel ver- sion. But the problem was keeping these HP processors cool enough because they're using NMOS 3 with a lot of current. They had the heat sinks, but they couldn't get the massively parallel CPUs close enough. They were forced to go to optical busing between CPUs because of the noise on the high-speed buses separat- ed by the distance it took just to handle the heat. They didn't want to go with liquid cool- ing on their systems. They wanted to stay with air cooling. Brooks: In 1968, I worked with Texas Instru- ments, and we built the very early stage of ESL integrated circuits. They designed the ILLIAC IV supercomputer for the government. We had as much difficulty designing leak-proof plumb- ing through the substrate as we did designing the chips to get them all together. Those were water-cooled through channels in the sub- strates. Shaughnessy: Well, thermal continues to be a hot topic; I had to say that. This has been great. Thank you for your time, gentlemen. Adam: Thanks, Andy. Brooks: Thank you. Always a pleasure. DESIGN007