The impact of data fragmentation on processing time modeling in semiconductor manufacturing : *Topic: IE: Industrial Engineering

Given their complexity, semiconductor manufacturing processes are highly fragmented. Hundreds of different products are processed by complex machines (so-called cluster tools) under different settings. In their turn, cluster tools are divided into different modules (generally into cassette, process, and transfer modules). Modeling semiconductor manufacturing processes for each of the resulting processing paths (i.e., combinations of modules, machine settings, machines, etc.) becomes intractable. This paper tackles the issue of data fragmentation in modeling the processing times associated with a Chemical-Mechanical Planarization (CMP) machine. Based on a cloud of available observations, the variability of processing times is modeled with different granularities. The distribution characteristics of empirical processing time data are determined via a moment-based approximation. Through an illustrative study conducted on a benchmark dataset, this paper quantifies statistically the value (resp. loss) induced by a fine (resp. macroscopic) representation of the processing times of a CMP machine.