Semiconductor Production System Research Articles

Reinforcement learning for adaptive order dispatching in the semiconductor industry

The digitalization of production systems tends to provide a huge amount of data from heterogeneous sources. This is particularly true for the semiconductor industry wherein real time process monitoring is inherently required to achieve a high yield of good parts. An application of data-driven algorithms in production planning to enhance operational excellence for complex semiconductor production systems is currently missing. This paper shows the successful implementation of a reinforcement learning-based adaptive control system for order dispatching in the semiconductor industry. Furthermore, a performance comparison of the learning-based control system with the traditionally used rule-based system shows remarkable results. Since a strict rulebook does not bind the learning-based control system, a flexible adaption to changes in the environment can be achieved through a combination of online and offline learning.

Material Flow Control Procedure Considering the Status Information of Transporters

A semiconductor production system has sophisticated manufacturing operations and needs high capital investment for its expensive equipment, which warrants efficient real-time flow control for wafers. To dispatch carriers of wafers in the semiconductor lines, this paper presents an unload request logic, which determines the destination of a carrier of wafers when a carrier finishes its operation, and a load request logic, which determines the next carrier to fill the newly available buffer space when a carrier leaves a buffer. The dispatching procedure first determines the transportation time of each carrier to its destination by each unit of transportation equipment and determines the destination machine by the unload request logic or target carrier by the load request logic. When there is no available buffer space at the machine tool, the procedure allows carriers to stay at the current buffer and determines the delay time, which is used to determine the destination in the unload request logic. Performance of the material flow control procedure is verified by an example.

Modified base-stock policies for semiconductor production system with dependent yield rates

We consider a two-stage production system faced by semiconductor manufacturing which produces a hierarchy of multiple grades of outputs. In the first stage, a single type of input (wafer) is used to produce multiple types of semi-finished parts with dependent yield rates, and in the second stage, each type of semi-finished parts can be transformed into a corresponding type of final products, or downgraded to a type of lower grade final products. Random customer demands are faced on the final products, and demands of different types of final products are not allowed to be substituted. The advantage of this production system is that it can prevent unhealthy ordering from customers who intentionally send out false demand signals for high grade products and revise the orders to lower grade products when the delivery time is close, which was observed in semiconductor manufacturing. The objective of the study is to plan the quantity of the input at the first stage and the respective downgrade quantities at the second stage so as to meet the required service level at the minimum cost. With some common assumptions, we propose a modified base-stock policy for this two-stage production system and show that the occurrence of nil excess inventory above the base-stock level follows a renewal process. We further extend the modified base-stock policy to a better policy that invokes risk pooling over multiple grade products. The performance of these two polices are evaluated via simulation to provide managerial insights.

Dynamic Reentrant Scheduling Simulation for Assembly and Test Production Line in Semiconductor Industry

Production scheduling in semiconductor production line is a complex combinatorial optimization problem. It is featured by reentrant production, product variety, frequent machine breakdown and especially reentrant phenomenon and dynamic randomness, which conduces enormous complexity to production line scheduling and management. Based on the resource conflict resolution strategies, dynamic scheduling models for a semiconductor production system are proposed here, aiming at finding maximum machine utilization, optimal man-machine ratio (MMR) and maximum output. Then the validity of the model is illustrated by a simulation case.

Proactive maintenance scheduling in a re-entrant flow shop using Lagrangian decomposition coordination method

Proactive maintenance scheduling is necessary to prevent facility trouble and to ensure high productivity in semiconductor production systems. For such systems, it is usually difficult to determine a proper plan because of complex process flows, which necessitate that jobs often re-enter a facility. This paper presents a method for re-entrant production floor optimization using Lagrangian decomposition coordination. By regarding maintenance as jobs that are limited by a starting and finishing time, the proposed approach produces a schedule that can facilitate proper maintenance. Feasibility is discussed using computer simulation results: the proposed method produces feasible solutions for minimizing overall tardiness problems.

日本の半導体生産システムの競争力弱化要因を探る：Papert's Principleの視点から

日本の半導体生産システムの競争力弱化要因を探る：Papert's Principleの視点から

Performance analysis of two-loop closed production systems

A production line with a limited number of carts or pallets is regarded as a closed loop manufacturing system. Such production lines are very useful in various manufacturing systems, such as semiconductor production systems. In this paper, we study a two-loop closed production system by means of block-structured Markov chains, and propose an effective RG-factorization solution to evaluate performance measures of the system, including throughput and WIP level. Based on this, the system can be numerically analyzed with respect to several crucial parameters. This leads to improvement on the design of the carts quantities, which are always crucial factors in closed-loop production lines.

A Study on High Speed Nanometer Positioning Table System. Realization of Ultraprecision Positioning Using a Repetitive Control Method.

In order to realize higher machining accuracy and higher productivity, it is required to materialize the table system which has capabilities of nanometer order positioning accuracy and high speed feed drive. In particular, nanometer order tracking accuracy is urgently required in semiconductor production systems. Accordingly, it is increasing to apply a linear motor and aerostatic guideway to the table systems. The linear motor-driven aerostatic table system has the advantage of non-contact direct drive, however, disturbances such as the force ripple of the linear motor deteriorate directly tracking accuracy of the table. In this paper, therefore, the performances of the linear motor-driven aerostatic table system using a repetitive control method were evaluated. Furthermore, through a series of experiments, it was proved that the repetitive control method could restrain tracking error caused by various factors.

The pumping character of the magnetic suspension-type turbomolecular pump of high flow rate

The pumping characteristics of the suspension-type turbomolecular pump with a high flow rate, which is operable within a pressure range of 102 Pa or below required for the semiconductor production system (physical vapor deposition, chemical vapor deposition, etc.), are described below. Conventional turbomolecular pumps are available with the operation pressure only up to a molecular flow range of 10−8 to 10−1 Pa. After trial manufacture of various rotor profiles, the rotor which would make the pump operable up to a transition flow range of 102 Pa was reviewed in this experiment. As a result of the experiment, a rotor which could achieve a flow rate of 2900 standard cm3/min (sccm) (6×102 Pa) at a pumping speed of 600 1/s (N2: 10−1 Pa or less) could be obtained. It was also found that the flow rate characteristics differed according to the rotor profile and working pressure range. Also reviewed in this paper is the surface treatment method for protection of the rotor (Al alloy) from reactive gases (Cl, F, and Ga) used in the semiconductor production systems.