Using Heuristics to Identify Maverick Lots at In-Circuit Test
Yield and defect levels in the manufacturing of high-complexity and high-reliability Printed Circuit Board
Assemblies (PCBAs) are extremely sensitive to variations in the process,component quality and workmanship. Any
deviation from expected targets needs to be investigated,not only after the identification of the manufacturing issue,
but also proactively before the issue becomes more pronounced. Traditional control tools such as yield tracking
and/or Statistical Process Control (SPC) capture deviations in the overall defect levels but they have no knowledge
what specific types of defects to monitor. In addition,the discontinuous nature of production build schedules and
fluctuating product volumes in the high-mix and low-volume electronics manufacturing domains can render SPC
ineffective. This research proposes a new methodology that interprets and analyzes defect and yield data to detect
“maverick lots”. These defect clusters are delineated from large volumes of production data in a high-complexity
PCBA facility that contains accurate information mixed indistinguishably with “noisy” data. For this approach,
historical information from prior analysis is used to distinguish deviations and trends from the current data. This
procedure is generic in nature and could be simultaneously executed automatically at several appraisal points on the
manufacturing line. This methodology has been used effectively in the Endicott Interconnect In-Circuit Test process
and has identified shifts in our Complex Assembly processes that may have gone undetected using other techniques.