Abstract
Wire bonding is one of the key processes which connects the wafer die and the lead frame in the IC packaging process. It is noted that process parameters influence both the speed and quality of bonding. Parameters, such as descending speed and cycles affect the bonding speed; while parameters such as bonding force and ultrasonic energy level affect the strength of joint. Both situations will affect the output (unit per hour, UPH), and lead to the variation of the production capacity as well.Nowadays, most companies in Taiwan still follow the rule of thumb when setting the process parameters. However, the large number of parameters in the bonding process makes it difficult for engineers to find appropriate parameters at once efficiently based on their experience only. When new products are introduced into mass production, engineers must go through multiple adjustments in a repeated trial and error way to find out the appropriate parameters, which takes lots of time, manpower and material resources.Using data mining techniques to establish predictive models for process parameters can not only increase the production capacity but also reduce the cost of the operation during adjustments. However, most studies in the field of wire bonding parameter optimization focused on increasing the bonding quality, while little consider the operation time and quality simultaneously. In this study, a data mining framework is proposed to extract the relationship between the bonding speed, bonding quality and the parameters.This study aims to increase production capacity by enhancing the bonding quality and the bonding speed simultaneously via recommending optimized process parameters. The proposed framework is based on a defect classification model constructed by random forest (RF) and extreme gradient boosting (XGBoost) method, and a multi-objective process parameters optimization model applying particle swarm optimization (PSO) method.An empirical study was conducted in a leading IC packaging and assembly company in Taiwan. The empirical result on 2 testing products reveals that the proposed approach increases the bonding quality by 70%, whereas the bonding speed is enhanced by 20%; overall, the proposed approach benefits output UPH by 26.6%, 5.0% for each product. Besides improving the production capacity, the proposed framework can also recommend parameters systematically and more quickly to enhance engineers' parameter tuning efficiency on-line and achieve the goal of reducing the introduction time of new products.