Issue link: https://iconnect007.uberflip.com/i/1174596
14 DESIGN007 MAGAZINE I OCTOBER 2019 sometimes they would fail, but sometimes they would pass. Too many people don't know how to design to pre- vent things from failing, es- pecially in the automotive world. For example, high-end airbag electronics have high layer-count boards. But for a lot of the microcontroller cir- cuits and low-end things that go into cars, we expect to have low layer-count boards. Dan Feinberg: I totally agree with you. One of the things that has happened in the last few weeks is the introduction by AMD of their 570 series chipset using their new 3900 processors. It totally freaked out Intel and kicked them in the butt. Those processors are using seven- nanometer technology. I'm preparing to build a rig using one as soon as I receive the chip, which should be soon. And these processors have a huge amount of power. This particular one is going to be a 16-core processor—about 32 threads—and the total power requirement under load will be 105 watts. Hartley: That doesn't shock me. I worked as a consultant for AMD around 1997 when they were doing the K7 processor, and even that with its memory drew almost 100 amps of power during peak power draw times. I can't even imagine what's happening now. Feinberg: And that chipset is the only one that can handle it. There's not another computer in the world from any manufacturer that can come anywhere near that right now. But it's going to be interesting to see what Intel does over the next six months to a year. Shaughnessy: Mike, do you want to chime in? Mike Creeden: As Rick said, the current land- scape is broad. Many people are trying to de- sign for production where there is still some slower-speed stuff going on, but most North American fabricators tend to be more in the R&D sector. I'm seeing that the micro-designs, as far as packaging goes, and the pin pitch are becoming so small that the manufactur - ing is extremely challenging. You see SAP, mSAP, and even EFAP technology as well as different types of manufactur - ing that are trying not to get a trapezoidal line but a squared line in the 1-mil range and vias in that 2–3 mil range. Dielectrics are cra - zy thin, so there are a bunch of manufacturing challenges that, as they become smaller, are a perfect fit for high-speed applications because you see thinner dielectrics. I recently transferred to Insulectro to be part of their team working with high-speed mate- rials. In the communication world, I'm see- ing some incredible speeds right now. I've also heard that there are going to be three different evolutions to it. You have the frequencies for 5G RF phased array antennas attempting to en- ter into ascending frequency ranges from 30– 300 GHz. You can't make the mistakes that we used to be able to make even five years ago. Is- sues of signal loss will be dramatic and affect performance. In the high-speed digital circuits, I see the same thing with memory: DDR3, DDR4, and DDR5. The PCIe-4 and PCIe-5 protocol ver- sions are going to be right behind it. It's be- coming hard to use standard connectors anymore. The parasitics on those connectors can make them ineffective; there's too much signal degradation. Signal performance is mak- ing new challenges for today's circuit boards. Happy Holden: I heard someone say that if you can breadboard your circuit using wires and pegboards, or some other kind of technique, and it works, then you're not going to have sig- nal integrity problems. Hartley: Eric Bogatin says that. Holden: But does anybody except hobbyists use that anymore? With the sophisticated chips, and the speeds people are talking about, you can't breadboard it; you're going to have to de- Mike Creeden Rick Hartley