Issue link: https://iconnect007.uberflip.com/i/1406534
SEPTEMBER 2021 I SMT007 MAGAZINE 65 design looks like, and what are the con- straints and parameters? at's why we said, "If we ever want to go into that space, we have to do the 'unsexy' work first: digitizing and connecting all the parts of the supply chain, where important information is hidden that is needed in the design process." But it's basical- ly "in the drawer," and we are focusing on this other part first. Matties: Do you think this might evolve into an AI tool that will autoroute complex PCBs? Schaal: e question is, really, what's the best way of delivering insight to a designer? We're focused on moving from EMS to OEM cus- tomers and tapping into the designer space. But I think the first step for us is not going to be building autorouting and autoplacing fea- tures. ere are other functions that are low- hanging fruits and that can help folks avoid res- pins without automating all their work, by de- livering information to the point where they need it. In general, I'm a believer. Placement and routing are constraint optimization problems, a fairly complex one, and mathematically this is something that a tool can at least support you with. Maybe we will reach that point where we will have a "compiler for hardware" that brings your problem schematics all the way to the physical layout without you having to do a lot of double-checking. But until we reach that point where we have this "compiler status," we have a long way toward a design tool. I see two trends. One, from the front end, is moving toward an easier schematic set-up, eas- ier library management, and compiling this all into a physical design which can still be opened in one's design tools to do double-checking. e other one, from the backend, where one may still be working in a similar workflow with the same design tools but would be given a lot more insight. is would mean knowing that the work you do is meaningful and that you'll avoid respins. Happy Holden: at's how we got into our au- tomatic board design system at HP, the Board Construction Advisor, which was an automa- tion of our DFM manual. is allowed us to set the goals for the design and the boundary con- ditions, and with our three-dimensional field solver, it would provide feedback on material stackup and geometries. But when they tested the soware on new electrical engineers, there was a problem—it was great at giving answers, but they didn't know the questions. HP Labs went back and said, "What we need is soware that asks the questions that has 20 years of experience, like Happy's group, be- cause we can hire electrical engineers out of college, but it takes a long time to grow expe- rienced people." at's where self-learning ar- tificial intelligence came in. e eight PhDs who developed this were from Carnegie Mel- lon and had a background in AI. As it designed each one of these boards, it learned the rules and techniques. Schaal: Happy, this is something revolution- ary you did, and you jumped right in and start- ed with the "automation process" itself. For us, in our two years as an AI solutions company Sebastian Schaal