Methods to achieve zero human error in semiconductors manufacturing

Abstract

The word of automation has been realized in semiconductors industry since assembly machine become automatic since mid 80's. The major enabler was implementation of pattern recognition system (PRS) in order to replace eyes oriented operation such as bonding location targeting. Over the last two decades, many machine manufacturers have managed to achieve fast, accurate, and repeatable bonding without much manual assistance. However, all the assembly machine manufacturers have not achieved one point process window, meaning major parameters still need to be in a specific optimum range. In many machines, password control has been known as the most often used method to control risks of human errors. Even that, several major risks of human errors could not be contained or controlled at some operation steps. One of those steps was requirement to perform recipe download, which was necessary to run various devices and packages in production. Meanwhile, recipe editing could not be avoided especially due to almost impossible opportunity in obtaining one hundred percent of PRS lighting robustness over time. Despite that, some of the non-closed loop machine parameters such as actual bond force, temperature and ultrasonic power would still require manual recording in checklist. On the other hand, most of the material handling such as bonding wire, and leadframe also require operator assistance and judgment. In order to resolve these potential human errors, a new advanced automated method was created base on the concept of computer integrated manufacturing (CIM). With commercial communication protocols, machines work interactively with host computer at various operation steps. The new CIM system has enabled various human dependant steps either automated or effectively controlled. The first few steps governed possible errors in production lot, machine ID, operator ID, and material handling. This information would be verified against multiple system databases, including planning, bill of material, machine maintenance record, operator skill certification, etc. After that, all major machine parameters would be obtained and sent to host computer with predetermined frequency, performed without human assistance. Upon receiving these data, they would be compared against process window, and any violation will trigger machine shut down. In the final stage, the correct recipe will be downloaded into machine and start allowing material flow in production. The whole system was designed and incorporated into Web-base access, convenient to extract any information at anytime and anywhere. One of the significant differences in comparing to an in-line automated system, this CIM system development and deployment do not require high cost, and hence it applies to a wide range of machines. This CIM system has proven to be a cost effective method to achieve zero human error in semiconductors manufacturing