Issue link: https://iconnect007.uberflip.com/i/1470111
16 DESIGN007 MAGAZINE I JUNE 2022 All the reviews came back. We had raised 130 or so different issues collectively. Only one was asked by everybody: "Can I thieve cop- per?" Other than that, they all found differ- ent issues in the same design using the exact same tools, just with different people. ere's no such thing as black and white in design. We don't have robots doing DFM reviews. ere are a lot of human interpretation factors. I used to have some of the front-end people go out and visit a designer because they'd say, "How can that idiot do this every time?" en they'd see all the inputs that the layout person was having trying to lay that board out. ey came back and said, "Wow, there's a lot of fac- tors they're accounting for." We'd bring the designers back in to look at the front-end CAM people and say, "Why do those guys keep ask- ing those questions?" Once they saw each oth- er's constraints, they both had more apprecia- tion for each other and it worked out better. But I love when a customer says, "e last company didn't question that." I'll say, "It's a law of physics problem. ey couldn't have built it that way. It's impossible. ey edited your data and didn't tell you." Oops. Shaughnessy: Apparently, true DWM is going to be bidirectional. e assembly folks will know enough about design intent that they can swap out a 1206 for another part and know that it will work. Korf: at's why, conceptually, all this work in digital twin intrigues me and it's a nontrivial answer. But I wish there was a way that we could automate the transfer of just the good knowl- edge, not the outliers, and modify the design practices or tools or rules, or the design system. We wouldn't have to rely so much on a human to feed it back, and there's a better chance of everyone's quality improving. I think there must be true feedback for this to really make sense. Eventually, we'll have AI assisting this flow. Holden: Yes, but we're probably not going to see anything for a while. One of these years, a PhD grad student working on some area of artificial intelligence is going to poke into the printed circuit informational arena as the focus of his artificial intelligence. Korf: One of our grandkids will figure out AI for PCB design. What's stopping us from doing this? It's possible. Automate the front end tooling process. Well, not the whole front end, but a high percentage of it could be auto- mated. We're not that far away from being able to do it. at's part of the drive for IPC-2581. e drawings should not contain any design infor- mation. It should be a report or a form. If you look at 2581 now, you don't need to send a board outline drawing. You don't need to send a tolerance. You don't need to send an imped- ance table, a bill of materials. You don't need to send drill codes. All of that is already in the database. ey don't need to call out, "ese pads get immersion gold. ese pads get elec- trolytic gold." You can label the pads on the board. To your point, Happy, we have to get rid of all this duplicate information, this hand- drawn, extracted information that's so error- prone. For some reason, people are still stuck with drawings in their heads. So, Andy, when you talk to people, how do they define DWM vs. DFM? Do you get any kind of input on what they think the difference is? Shaughnessy: Basically, the designer, fabrica- tor, and assembly provider will all be on the We don't have robots doing DFM reviews. There are a lot of human interpretation factors.