Issue link: https://iconnect007.uberflip.com/i/1511625
NOVEMBER 2023 I PCB007 MAGAZINE 31 Combining measurement data with design information, such as annular ring and drill to copper, we can decide whether a product meets requirements or whether process parameters need to be changed to improve yield and long- term product reliability. Integrating relevant data from multiple sources builds our understanding of the pro- cesses and materials. Combining inner layer scale factors with measured scale errors, we can determine the total material movement for each core in the construction and compensate for it. By comparing total material movement for materials across multiple batches and prod- ucts, we can identify unexpected behavior caused by a process or material change. Supplementing data with engineering exper- tise converts data into a dynamic knowledge base. If we can answer why unexpected behav- ior occurred, then we can truly enrich the data. Machine Learning for a Smart Factory e volume of data generated in the PCB fac- tory requires automated systems to be smart and flexible. Technology roadmaps only move toward the more complex while product life- cycles are shortened, resulting in higher prod- uct mixes. is in turn increases the number of variables involved and significantly increases the number of potential combinations of mate- rials, process, and product design. It is not feasible to acquire the volume of data required to provide empirical data for each of the possible combinations. e average fab- ricator will manufacture using multiple resin systems, each having numerous core thickness and constructions, prepreg glass cloths, and resin contents. is can quickly result in tens of millions of possible combinations for even the simplest products without even consider- ing the options for copper thickness, copper pattern, or processing routes. Increasing technical capability will have a significant effect upon the number of potential material and process combinations: increased layer counts, number of lamination cycles, and portfolio of materials results in yet more pos- sible combinations with little or no knowledge of how they behave. Historical data alone will never be able to provide all the answers required, and when Figure 3: An example data flow that provides proactive, adaptive control and monitoring.