Periodic analysis of critical dimension measurements in the photolithography manufacturing process.

Abstract

ModIV is Nortel Networks' semiconductor fabrication facility. It specializes in producing complex designs at low cost in a short period of time. This feature allows ModIV to serve as both a production and research facility. The semiconductor manufacturing process is a series of precise layering steps carried out on silicon wafers. Each layering step occurs in one of three fabrication areas: device formation, photolithography and interconnect. One of the biggest difficulties associated with troubleshooting problems in the semiconductor process is due to the lack of measurements performed on production batches. Most of the production measuring steps occur in the photolithography area. In the photolithography area there are three sequential processing steps: coating, exposing and developing. For capacity and maintenance reasons, there are three pieces of equipment that can be used interchangeably at each of the three processing steps. This means that there are three coaters, three steppers and three developers. Measurements can only be taken after the developer step. The equipment interchangeability creates a problem when trying to detect the root cause of processing problems. Currently each processing level has its own control chart that incorporates all of the processing paths for that particular level. The primary objective of this work was to develop a method of isolating the root cause of processing problems. A data extraction and sorting scheme was developed to create data files that could be analyzed for periodic behavior. A model building algorithm was created to detect periodic behavior and a series of stochastic simulations were used to determine the quantity of data required to detect the underlying signals with different signal to noise ratios. An experiment was conducted in the photolithography area to intentionally cause a problem. This problem isolation method was used, and the correct root cause of the problem was found.