Issue link: https://iconnect007.uberflip.com/i/1307491
14 DESIGN007 MAGAZINE I NOVEMBER 2020 of the via structure is very important. No matter which via technology you utilize in your design—whether it is standard PTH, HDI, micro- vias, or a combination of the three—you want it designed so that the via is in the board fabricator's sweet spot while meeting your design require- ments and DFM for down- stream optimization. If you're not doing that, you'll be hav- ing a lot of problems. I am still surprised to find there are many people who aren't talking to their suppliers, and they're just throwing the design over the fence. Then, you have to understand the material you're choosing. Are you choosing the right material for your application and the widget that you're building? Are you designing in the core sweet spots for success, or are you push- ing the rail on complexity? You had better be speaking and understanding the industry ter- minology and language when talking to your supplier to make sure that you're on the same page so that there are no surprises at the end. Happy Holden: If an OEM is naïve enough to send a board out to multiple vendors, they shouldn't be surprised that they'll come back different because every fabricator optimizes drilling, desmear, metallization, and plating to be as reliable as they can possibly make it. But because of the multiple vendors, machines, and processes, there are 10,000 different per- mutations and combinations that make up every multilayer. An OEM deserves whatever they lose if that's the way they're going to man- age their supply chain. They should develop a qualification process that's statistically signifi- cant, but they won't, and the fabricators won't tell them that they're foolish. And many OEMs don't understand reliability, so they ask for 15 boards with a cross-section. That does not guarantee any kind of reliability at all because 15 boards are not statistically significant unless your entire year's purchase is 15 boards. Also, a cross- section has been proven to be so unreliable to tell you about what's really happening on the interconnect because you're only looking at a plane cut through the hole. You're not looking at all 360 degrees. When companies hired me, I gave them a quality program that included reliability and used the IPC PCQR 2 meth- odology, which meant that the fabricator had to produce 250,000 holes in the qualifica- tion, or half a million holes if they chose the blind laser drill to panel. But those boards were tested so, there was a 99.5% reliability confidence that there was less than a 0.5% error over the entire life- time. And they redid that qualification every year. If there were coupons and things like that, then they tested on the same panel as the PCQR 2 so that there was traceability. Most people don't want to do that even though that's money well spent. I worked a lot with Asian fabricators that supply Intel, and Intel's qualification is 20,000 boards and mod- ules, which Intel pays for, and they are split over an entire year. They want to know that the boards are made day and night, Monday through Friday or Saturday, summer and win- ter, because the measurements go up and down depending on seasons and a whole bunch of other factors. But the people who really sink their money into reliability, like the packages for the Intel microprocessors, do it properly, and pay the fabricator to do that, so they have extremely high confidence and know exactly what they're doing. But don't order three boards or 15 boards and cross-section them, and think, "That's a good fabricator." They might be, but maybe not. Do you think people really understand reli- ability and statistical significance? Chavez: I'd like to think so, but there are still many that simply don't understand or don't Stephen Chavez