ROOT CAUSE ANALYSIS IN PROCESS-BASED INDUSTRIES

Horev brings 20 years of experience to bear in this methodical treatment of problem solving in the semi-conductor industry, written with wider use of these analytical techniques in mind.

The processes by which silicon ingots are cut into wafers, etched and assembled into microchips are highly complex, meticulous down to microscopic detail and demanding complete cleanliness and quality-control on both macro- and microscopic levels. Horev’s manual emerges from the management of these systems, encompassing the problem solving necessary to suss out production issues on an active line where controlling time and material loss, as well as maintaining quality control and product reliability, are paramount. Such manufacturing processes are so complex that exhaustive monitoring of every parameter would require, according to Horev’s calculations, millions of monitoring sights for a single process parameter, or the allowable range of results for a certain component. But with a more realistic quantity of monitoring sites, data collection and analytical acumen, the sources of problems can be deduced. He begins by cataloging the types of process noise, or deviation from anticipated results, and how these scenarios will appear in measured fluctuations over time, as visualized in trend charts, ranging from single events to patterns of repetition that might be rooted in either human or machine defects—the wearing out of a polishing disk or incorrect maintenance by workers. He outlines the sequence of cause analysis, a continuous cycle that encompasses problem definition, problem characterization, model building and model validation. Here, the complete and useful model has three components: conditions or initial qualities; properties, or the impact that these conditions have; and behavior—how these conditions reflect on the system. While simpler models work in the description of sequential events, coincidental events might be involved in the root cause, requiring the pursuit of a number of simultaneous and interrelated conditions, effects, symptoms, etc. This is where property trees and other models are used to enumerate then eliminate paths from observed effects to possible root causes as well as the interplay between a model’s many elements. In emphasizing human factors in both the manufacturing processes and the problem-solving team, and connecting problem-solving technique in a specific environment to applications in the world at large, Horev stresses the importance of ingenuity on the part of investigators, as well as a pragmatic observance of boundaries to an investigation—the extent, for example, to which an investigation ought to usefully be pursued.

Readers unfamiliar with the semiconductor industry might find the text daunting but the essential information is useful in its application across disciplines.