Integrated Manufacturing Decisions for Next Generation Factories – IMAGINE

The project initially focused on applied research based on advanced statistical methods to build and use an equipment behavior model: Multivariate Capacity Index, Moving Variance/Covariance. In addition, this provided highly effective methods to quickly support real-time decision-making by: Classification of failure modes correlated to the health index and real-time monitoring of this index using advanced statistical control charts. In addition, research was conducted on task scheduling integrating equipment behavior (its health). This involves scheduling tasks by integrating the fact that the machine is only capable (qualified) of executing families of tasks over a certain period of time depending on what has been previously executed. In summary, the following research questions were addressed: a) Use of correlations between an equipment failure mode, the induced effects of detections in order to define the origin of the causes of equipment failures. , b) Construction of a health index based on the failure mode. c) development of optimization algorithms for a scheduling problem on parallel machines based on the health status of the equipment by constraint programming and linear programming.

Faced with the complexity of scheduling in the semiconductor industry, marked by advanced process constraints, this project developed two exact solution methods: a linear integer program and a constraint programming model. In order to optimize the search for solutions, particularly when these methods are time- or resource-intensive, two heuristics were proposed: a recursive procedure and a simulated annealing algorithm. Tests on random instances validated the effectiveness of these approaches. The study highlights the importance of equipment condition in scheduling, revealing a direct link between equipment health and production performance. It introduces condition-based monitoring (CBM) as an innovative means of managing equipment deterioration, with a focus on failure diagnosis and prediction. This approach, validated by a case study, demonstrates its effectiveness. In sum, this work enriches scheduling management by integrating equipment state, and proposes innovative models for improving reliability and efficiency in semiconductor production.

To make the proposed equipment behavior prognosis approach practical for the industry, the deterioration models should be integrated into the monitoring scheme of the equipment condition. The monitoring mechanism can be utilized to optimize the production control, such as the job scheduling and maintenance planning. Batching is often a critical workshop in semiconductor manufacturing and so scheduling while incorporating time constraints to maintain machine performance (achieved through machine health) with batching is relevant.

Kao, Y.-T.; Dauzère-Pérès, S.; Blue, J.; Chang, S.-C. Impact of integrating equipment health in production scheduling for semiconductor fabrication. Computers & Industrial Engineering. 2018, 120, 450-459.

Rostami, H.; Blue, J.; Yugma, C. Automatic equipment fault fingerprint extraction for fault diagnosis on batch process data. Applied Soft Computing. 2017, 68, 972-989.

Nattaf, M.; Obeid, A.; Dauzère-Pérès, S.; Yugma, C.; Wu, C.-H. Parallel machine scheduling with time constraints on machine qualifications. Computers and Operations Research. 2019, 107, 61-76.

Rostami, H. Equipment Behavior Modelling for Fault Diagnosis and Deterioration Prognosis in Semiconductor Manufacturing. Thèse de doctorat, Université Jean-Monnet, Saint-Etienne, 2018.

A prototype has been developed in Matlab, providing prognostic-diagnostic solutions for industrial semiconductor data.

By extensive use of product metrology data, traditional Statistical Process Control (SPC) is performed to ensure an in-control process and thus quality products. As the feature size of the high-tech products is getting smaller and lighter, the manufacturing processes, such as IC fabrication processes, are becoming very sensitive to disturbances and variations. Various advanced control methods, e.g., Advanced Process Control Advanced Equipment Control (AEC) and Fault Detection and Classification (FDC), have been therefore developed to further monitor tool parameters for a better control of the delicate processes, in particular, the technology node under 90nm. To achieve effective control, it is critical to base these control algorithms and models on the knowledge of the process physics. Compared to other industries, it must be noted that the current availability and reliability of semiconductor production tools are yet to be improved to a satisfactory level. Equipment utilization has to be optimized in terms of production throughput and product quality, or to a greater extent, in terms of the overall equipment efficiency, measured by the total number of hours a tool is up and running within a certain normalized timeframe. OEE even worsens in production lines having a high mix of products and technologies that have to be processed on the same equipment set. Scheduling consists in optimally assigning resources to a set of tasks, in order to perform all these tasks under imposed constraints and to optimize indicators. Scheduling and control are interdependent. For example, control requires information from scheduling (where jobs are processed) to sample the better lots that bring information on tools.To improve the current availability and reliability of the production tools, it is necessary to move beyond the present APC and FDC capabilities towards a predictive approach. The first task of the project is to develop an Equipment Health Index model for Predictive Equipment Behavior and the second task is how to use EHI in sampling strategies and scheduling to guide the lots which are more reliable in the production and metrology line.The EHIS3 project is an industrial research project that aims at proposing novel models and algorithms based on the tool parameters and the underneath failure mechanisms to better describe the behavior of the process equipment. The objective is to develop an Equipment Health Index (EHI) for failure identification and to integrate the EHI into the production decision system for taking better decisions for sampling, scheduling on metrology and process tool and make preventive maintenance. It will enable engineers to align the manufacturing plans in accordance with the real "health" state of the equipment. The Equipment Health Index for Predictive Equipment Behavior will directly contribute to the OEE by improving equipment availability and predictability. With the failure predictability at the equipment level, process excursions can be greatly minimized and the unscheduled downtime can be cleverly reduced. The capability to integrate these new features together with existing production control systems will further reduce the variability of both equipment availability and performance and in turn improve the production yield and cycle time.In particular, an EHI is based on actual measurements of lots on metrology tools. Thus, important questions are how to sample lots and schedule them on metrology tools in order to ensure that the EHI remains relevant. It is also crucial to increase the number of measurements when a process tool is at risk, i.e. to sample more lots processed on the tool when its EHI indicates that the probability of failures is increasing. Hence, to meet the two previous objectives, the EHIS3 project will also aim at developing new strategies and algorithms for sampling lots and scheduling them on metrology tools.