Generating Cycle Time-Throughput Curves using Effective Process Time based Aggregate Modeling

Abstract

Abstract-In semiconductor manufacturing, cycle timethroughput (CT-TH) curves are often used for planning purposes. To generate CT-TH curves, detailed simulation models or analytical queueing approximations may be used. Detailed models require much development time and computational effort. On the contrary, analytical models, such as the popular closed-form G/G/m queueing expression, may not be sufficiently accurate, in particular, for integrated processing equipment that have wafers of more than one lot in process. Recently, an aggregate simulation model representation of workstations with integrated processing equipment has been proposed. This aggregate model is a G/G/m type of system with a workload-dependent process time distribution, which is obtained from lot arrival and departure events. This paper presents a first proof of concept of the method in semiconductor practice. We develop the required extensions to generate CT-TH curves for workstations in a semiconductor manufacturing environment where usually only a limited amount of arrival and departure data is available. We present a simulation and an industry case to illustrate the proposed method.

Capacity restrictions and supply chain performance: Modelling and analysing load-dependent lead times

“…Literature in this area mainly focus on developing and improving methods for generating CT-TP curves (see e.g. Nemoto et al, 2000;Park et al, 2002;Fowler et al, 2001;Leach et al, 2005;Yang et al, 2007;Pahl et al, 2007;Veeger et al, 2010;Kacar et al, 2016, among others), given that the generation of empirical CT-TP curves requires collecting large amounts of representative data (Mönch et al, 2013). In the context of our research, modelling lead times using a CT-TP curve may represent a reasonable procedure to come closer to the true essence of capacity problems in real-life supply chains.…”

Predicting Cycle Time Distributions with Aggregate Modelling of Work Areas in a Real-World Wafer Fab

“…The WIP-dependent EPT distribution can be calculated directly from actual arrival and departure data. (Veeger et al 2010a) demonstrated that the method of (Kock et al 2008) suffices to predict the mean cycle time, but does does not accurately predict cycle time distributions. This limitation is largely a result of the First Come First Served (FCFS) rule in the aggregate model.…”

“…One of the conclusions was that the EPT is an explicit function of utilization and henceforth a dynamic quantity. This problem seems to be solved by introducing the WIP-dependent EPT distribution used by both (Kock et al 2008) and (Veeger et al 2010a). Another systematic error that can occur is due to time-based interruptions in the system, e.g.…”

Fluid Models of Parallel Service Systems under FCFS

“…The literature on parallel service system is quite voluminous. An incomplete list would include an early study [18]; applications to manufacturing and supply chain management [26,31], applications to call centers and internet service systems [16,21,27,33], attempts to find optimal policies, mainly for small graph systems [7,8,9,17,29,34], heavy traffic and fluid approximations [22,23], and many-server scaling [1,19,20]. In view of [20,22,23], establishing fluid approximations is often the first step to solve the optimal dynamic control problem for parallel service systems.…”