Abstract: Semiconductor manufacturing plays a critical role in modern industry. With the rapid growth of this field, semiconductor companies are challenged in both R&D and manufacturing domains. However, the increasing complexities of devices, structures, materials, processes, and even changing physical models pose great challenges for both modeling and computation. In addition, the growing amount of data and the large number of people involved in the R&D, design, and fabrication of semiconductor components complicated the production systems and slowed down the overall throughput. For the R&D and production of semiconductor devices with advanced technology nodes, the integration of Physics-based modeling, Computer Aided Engineering tools, parameter optimization, and Machine Learning methods constitutes a new vector for promoting innovation and productivity in the semiconductor industry.
Machine Learning methods empower automated, efficient, and intelligent solutions for semiconductor modeling by modeling data with Non-linear regression, Principal Component Analysis, clustering, classification, and generative methods. Physics-based regression methods such as Fourier Series expansion and Polynomial Chaos are established to establish neural networks for topology optimization, sensitivity analysis of processes with uncertainty quantification, inverse characterization of materials, and accelerated simulation. Reinforcement Learning tools have been successfully developed for the early-stage optimization of processes and design. Meanwhile, Deep Learning-based tools such as Generative adversarial networks and convolutional neural networks have been developed for the design of structures/gate layouts and the qualification of patterns.
On the other hand, modern semiconductor manufacturing consists of multiple departments with complex production systems. Significant efforts have been made on modeling layout storages and new equipment selection to optimize the bi-objective cost and yield in extreme scale layouts. Mathematical programming, agent-based models, and Reinforcement Learning methods have been proposed to optimize the scheduling of diverse wafer processing flows and streamline interactions at the Fab level between manufacturing equipment, input/output, and cost. Moreover, after a decade struggle, advanced Process control systems in conjunction with on-line monitoring monitors critical sensors and control actuators to solve quality issues on time. These large-scale systems save resources and improve quality at the cost of higher complexity. Process data analysis and fault detection methods such as Fourier analysis and kernel-based methods have been established to model non-linear propagation of disturbances and forecast machine states for predictive maintenance.
Keywords: Machine Learning, Semiconductor, Research, Manufacturing, Automation, Predictive Modeling, Process Optimization, Defect Detection, Yield Improvement, Data Analytics, AI in Semiconductors, Fabrication, Smart Manufacturing, Quality Control, Industrial AI
|
DOI:
10.17148/IJARCCE.2021.101274
