Issue link: https://iconnect007.uberflip.com/i/1315894
DECEMBER 2020 I DESIGN007 MAGAZINE 35 work together with manufac- turing, where there's a con- tinuous flow of information between the two. As manufac- turing is usually remote from design, things happen—such as material shortages or pricing issues—that invoke alterna- tive materials from local sup- pliers to be chosen, for exam- ple. Manufacturing often uses materials that the designer did not authorize or even foresee. All of your DFM rules in the world would not have included logic about exceptions that happen beyond the control of the design. Therefore, there needs to be a collabora- tive platform on which the local manufac- turer, looking to source alternative materials that will satisfy the need of production in this instance, would seek and get permission to make the substitution from the designer, who will respond to confirm that the functionality, specification, and physical attributes—includ- ing the leads, pads, shapes, and size, etc.—all look good. This process doesn't happen today, or if it does, it's extremely ineffective, as such interactions occur too late, and the designer has moved on to something else. Holden: That's important because we have engineers who pick a component and every- thing seems okay, but if they ask the assembly test people, they would say, "No, we have a lot of failure and test issues with that component. This component over here does the same thing but has a much higher quality testability level than what you selected." Ford: That is often the case and may be a local- ized issue. It may depend on certain produc- tion equipment, as well as material vendors. It's really important to have interaction. I've seen that kind of feedback work very well, where instead of the legacy DFM problems I mentioned earlier, the manufacturer can show information, such as when you use this component, there is a first-pass yield issue of around 100 ppm; however, if you were to relocate or site that component with a differ- ent orientation, the first-pass yield defect reduces to 10 ppm. That form of information is very easy for designers to understand because the whole issue of domain-based techni- cal language has been elimi- nated through the use of data. Johnson: That sounds like a huge amount of data being turned into statistical AI rec- ommendations that can be fed back. It must mean a huge change in what the manufactur- ers are collecting for data, how they're present- ing it, and a great deal of AI and knowledge- based expertise going to the fabricators. Ford: The focus of a lot of people in manufactur- ing right now is to collect as much data as pos- sible; in most cases, they collect messages from machines. The IPC CFX standard is the most efficient way to collect the data from machines, but context is needed to create values. Qualified facts and events should be com- municated, but if you only exchange the raw data—which would incidentally be massive in terms of data size and complexity—how is the designer supposed to understand it? Any design-based computer software has a similar problem; in the same way, a human designer can't understand the manufacturing engineer, the designer's computer solution can't under- stand the manufacturing data. Contextual- ization of many data points together at the manufacturing side creates meaning for oth- ers to understand. We also have to eliminate all internal causes of variation that occur in manufacturing so as not to confuse contribu- tions to the root cause. We then have the basis of factual feedback to designers and their sys- tems in a way that they can understand. Johnson: Historically, it has been the situa- tion where you would write that up as a rule deck. In IC, in the early days at Mentor Graph- Happy Holden