Algorithm for Ultrasonic Wire Bond Outlier Classification

Abstract

Wedge bonders use ultrasonic energy to bond metal wires onto metal substrates in a process that takes milliseconds. In high-volume production, failures cause downtime and costs. On-line monitoring systems are used to reduce the failures and determine the root cause. We developed and tested an algorithm to classify outliers in ultrasonic wire bond production. The algorithm is used in large wire wedge bonders to measure and analyze process signals and detect and classify bond outliers. It helps bonder operators, production supervisors and process engineers to detect process deviations and fix the underlying root causes. The algorithm measures bond signals, such as deformation, ultrasonic current, and ultrasonic frequency. Bonds are automatically divided into subgroups based on bond order and process parameters, and the signals within a subgroup are then normalized. For outlier classification, the features are extracted from the normalized signals and combined into failure class values. The failure classes such as contamination, no wire, high deformation, misaligned wire, and unstable substrate, are calculated independently. We measured the detection rates for large wire aluminum bond failure classes and demonstrate how the algorithm calculates failure class values from the signals. Additionally, we show how new signal features and failure classes can be defined to detect production-specific or rare failure classes.