Estimation of lot processing time in an IC fab

Cycle time plays an important role in an IC fab. Cycle time consists of pro cessing time and waiting time. This paper proposes a numerical algorithm to estimate the processing time. This approach is based on the tool model, wh ich characterizes the dynamics of a single tool and is divided into two par ts: a processing time model and a waiting time model. In our tool models, t he wafer processing time is the time required for wafer processing on this tool. Similarly, the waiting time is the time between the end of the curren t process and the beginning of the next process. The estimated IC product c ycle time is the sum of the time estimated by all tool models. Since the pr ocessing time could be influenced by many parameters, only the most importa nt parameters are taken into consideration. These parameters include wafer quantity, technology and others. In this paper, three different algorithms analyze the processing time model: Gauss-Newton based Levenberg-Marquardt a lgorithm (GN algorithm), back propagation of neural-network (BP), and adapt ive neural-fuzzy inference system (ANFIS). Compared with the other two meth ods, the ANFIS algorithm provides the most accurate prediction result at th e expense of the highest computation cost. The adoption of a fuzzy inferenc e system provides greater physical insight for engineers to understand the relationship between the parameters. Significance: This paper proposes an approach to construct a tool model to predict lot processing time by applying three different techniques. A compa rison and discussion is provided.