Search

As this Special Section goes to publication, semiconductor manufacturing in the United Status, and globally, continues to expand at a seemingly torrid pace. Assisted by government funding and driven in part by artificial intelligence workloads that gobble up increasing amounts of data center computing capacity, Intel and TSMC fabs are going up in Arizona, TI and Samsung fabs are coming to Texas, and Micron has big plans in New York. Unfortunately, just like those flying cars we were once promised, AI has not yet eliminated the need for the skilled trades and engineers required to build and successfully operate a fab. As a result, workforce development has become an important part of the increasingly complex semiconductor manufacturing process: Where are the thousands of engineers the semiconductor industry needs to staff these new fabs going to come from? How can we make more students excited about science and engineering? While the Guest Editors don’t have all the answers, we are happy that ASMC contributes to the solution by actively supporting student presentations and posters and annually recognizing the best student paper of the conference. And, maybe some day, the artificial intelligence systems that semiconductor manufacturing has enabled will give us those Star Wars or Star Trek robots that can build fabs and make chips too. [ABSTRACT FROM AUTHOR]

Semiconductor manufacturing processing can be jeopardized due to process fluctuations, and the degradation of equipment parts can significantly influence process variation. Timely diagnosing equipment faults causing process variations is desired in current high-end product manufacturing. This paper proposes a diagnostic method for the SiH4 gas flow rate drift using N2 vibrational transition in oxide deposition. In this research, optical emission spectroscopy (OES) and quadrupole mass spectrometer (QMS) are employed as condition monitoring sensors serving as a reference model to compare the diagnostic performance for gas flow rate drift. The study observes that the OES model exhibits much higher performance for minor diagnoses of less than 5% drift. The diagnostic model performance can be enhanced by incorporating plasma condition and gas indicators compared to when these indicators are used individually. This suggests that when conducting diagnostics for equipment and processes, it is crucial to consider indirect indicators like plasma indicators along with direct indicators such as gas radical density. The comprehensive use of both types of indicators enhances the diagnostic performance, providing a more accurate assessment of the conditions and potential problem in semiconductor manufacturing. [ABSTRACT FROM AUTHOR]

This paper studies simultaneous scheduling of production and material transfer that arises in the semiconductor photolithography area. In particular, the right reticle and right job both need to be present to process the job. Jobs are transferred by a material handling system that employees a fleet of vehicles. Reticles serving as an auxiliary resource are also transferred from one place to another by a different set of vehicles. This intricate scheduling challenge, encompassing jobs, reticles, machines, and two distinct sets of vehicles, is explored here for the first time. The paper introduces a multi-stage methodology that involves relaxation, a constructive heuristic, constraint programming, and a warm-start approach to address this complex problem. [ABSTRACT FROM AUTHOR]

This is a panel paper that discusses the use of Artificial Intelligence (AI) technologies to address production and supply chain level problems in semiconductor manufacturing. We have gathered a group of expert semiconductor researchers and practitioners from around the world who have developed AI solutions for various semiconductor problems. This paper aims to provide their answers to an initial set of questions and provide an overview of the AI developments and empirical studies to make suggestions for future directions in this arena. [ABSTRACT FROM AUTHOR]

This paper proposes a linear regression model for scalar-valued responses and two-way functional (bivariate) predictors. Our motivation stems from the quality evaluation of products based on optical emission spectroscopy data from virtual metrology of semiconductor manufacturing. We focus on multivariate cases where the smoothness and shapes of the data vary significantly across variables. We propose a two-step solution to this problem, consisting of decomposition and prediction. First, we decompose the two-way functional data into pairs of component functions using functional singular value decomposition. Next, we build functional linear models for the decomposed functional variables and obtain the final predictor by averaging the models. Results from numerical studies, including simulation studies and real data analysis, demonstrate the promising empirical properties of the proposed approach, especially when the number of predictors is large. [ABSTRACT FROM AUTHOR]

This paper addresses the quality control of the photolithography process in the semiconductor industry. Overlay errors in the process seriously affect the wafer yield, and cause the wafer to be forced to rework and affect the production efficiency of the equipment. We examine the current state of its process control, develop a novel overlay predict model, and verify the prediction results. This study proposes a Global Attention Generative Adversarial Networks (GAGAN) model to precisely predict the overlay error for the feed-forward data of the front layer, which is used as the important information and process parameters for the advanced process control of the current layer. Experiment results on a semiconductor shop-floor confirms that our proposed method achieves high predictive performance while maintaining extensibility and visual quality. [ABSTRACT FROM AUTHOR]

This paper studies the problem of scheduling machines in the photolithography area of a semiconductor manufacturing facility. The scheduling problem is characterized as an unrelated parallel machine scheduling problem with machine eligibilities, sequence- and machine-dependent setup times, auxiliary resources and transfer times for the auxiliary resources. Each job requires two auxiliary resources: a reticle and a pod. Reticles are handled in pods and a pod contains multiple reticles. Both reticles and pods are used on multiple machines and a transfer time is required if transferred from one machine to another. A novel constraint programming (CP) approach is proposed and is benchmarked against a mixed-integer programming (MIP) method. The results of the study, consisting of a real-world case study at a global semiconductor manufacturer, demonstrate that the CP approach significantly outperforms the MIP method and produces high-quality solutions for multiple real-world instances, although optimality cannot be guaranteed. [ABSTRACT FROM AUTHOR]

Semiconductor manufacturing has been extensively exploiting machine-learning (ML) to process equipment sensory data (ESD) for near-real time anomaly detection (AD). ESD characteristics are highly diversified and data lengths vary among processing steps and cycles. Cleansing ESD with minimum distortion (CMD) to fit the fixed-length input requirement by ML-based AD is critical to AD effectiveness and is challenging. This paper presents a novel CMD method of four innovations: i) statistical mode-based equalization of step data lengths for the least number of step data length changes, ii) importance indicator value (IIV) of a data sample based on its relative difference with the subsequent sample, and iii) step data segmentation into groups based on samples of significant IIVs and the least-entropy-group-to-cleanse-first rule, and iv) cleansing the least IIV sample(s) in the selected group for step data length equalization. CMD application to ESD demonstrates its characteristics preservation property. Simulation experiments are on an integration of data cleansing with an unsupervised ML-based AD system, STALAD. Comparisons with two benchmark methods over AD scenarios of small-scale drifts and shifts show that CMD not only is superior in facilitating accurate detection by STALAD but also helps detect anomaly much earlier than using the two benchmarks. [ABSTRACT FROM AUTHOR]

Identifying the patterns of defective chips in wafer bin maps (WBMs) in semiconductor manufacturing processes is crucial because different defect patterns correspond to different root causes of process failures. Recently, mixed-type defect patterns (i.e., multiple defect patterns in a single wafer) have become increasingly common owing to the increased complexity of semiconductor manufacturing processes. Previous methods for classifying mixed-type defect patterns in WBMs focused on outputting only the class labels of the defect patterns and not their locations, although location information of the defect patterns is useful for tracking the root causes of failure and improving processes. Moreover, most previous methods used only labeled WBM data, although a larger quantity of unlabeled WBM data are more accessible because of the costly process of label annotation. Therefore, in this paper, we propose a semi-supervised learning method for classifying mixed-type defect patterns and detecting their locations simultaneously using both labeled and unlabeled WBM data. The proposed method extends a recent unsupervised object detection method called Attend-Infer-Repeat in a semi-supervised manner to perform object detection and classification simultaneously. The performance of the proposed method is verified using WBM datasets of different sizes. The results demonstrate the effectiveness of the proposed method for classification and location detection. [ABSTRACT FROM AUTHOR]

Semiconductor manufacturing consists of multiple sequential processes. In addition, even in a single process, a wafer must pass several steps. Accordingly, a dataset generated in semiconductor manufacturing has sequential information. Thus, the sequential information between steps and processes must be considered when predicting a target variable such as a yield or defect status. This paper proposes a method that utilizes the concept of residual learning to capture sequential information. Specifically, we propose to learn several modules that use data gathered starting from different steps and obtain a final decision by combining all modules’ decisions. In each module including multiple models, the first model is trained to predict the target variable using the data from the earliest step, and the remaining models are trained to predict the residuals that cannot be explained by the models for the previous steps based on the concept of residual learning. We conducted extensive experiments using two real-world semiconductor manufacturing datasets and found that even though each module’s performance was not good, the final decision obtained by combining all the modules’ decisions achieved a significant performance improvement. As a result, the proposed method significantly outperformed the baseline models. [ABSTRACT FROM AUTHOR]

Equipment at the end of functional life poses several challenges for manufacturing operations and long-term asset sustainability. This is even critical for semiconductor manufacturing where equipment upgrades are capital intensive with a longer return horizon. This demands an objective and quantifiable approach to manage and monitor the end-of-life health of the manufacturing systems. The paper presents a practical three-level approach that brings together engineering, operational, and supply chain factors under a single indicator. A visualization heatmap correlates equipment health to manufacturing impact for engineering and commercial decision-making. The approach is simple yet proven to be effective in fab environments. The proposed analysis applies in formulating system priorities, upgrade strategies, and capital expenditures. Brief guidance on health improvement strategies, application to new fab, and cost-based evaluation are presented for practical use. [ABSTRACT FROM AUTHOR]

Beyond the widely-studied scheduling of wafers within cluster tools, a novel and important perspective is raised in this paper to tackle an upper-level optimization problem in real-world production, i.e., the assignment of hybrid types of wafer lots to a set of cluster tools with parallel modules to minimize the maximum completion time for the lots. The main difficulty in addressing such a problem is that the objective, i.e., the maximum completion time, cannot be calculated explicitly beforehand. To make this problem tractable, the associated maximal overlap among tools is utilized to heuristically evaluate the objective for the problem. Besides, since the cluster tools for processing are identical, we further tackle this problem as a clustering issue. Accordingly, a clustering algorithm based on greedy searching is proposed to allocate wafer lots into cluster tools while minimizing the maximal overlap. To elucidate our method and its significance in real-world production, the wet bench tool in wet cleaning process is taken as a case study. We compare the proposed algorithm with the empirical method in fabs and several intelligent optimization algorithms, and the experimental results verify the effectiveness of our proposed method in terms of improved efficiency. [ABSTRACT FROM AUTHOR]

In semiconductor manufacturing processes, yield analysis aims to increase the yield by determining and managing the causes of low yield. The variable data collected from semiconductor manufacturing processes, in which hundreds of unit processes are implemented according to specific conditions and sequences, are interdependent, and the variables related to previous processes influence the variables in subsequent processes. Therefore, the order of processes should be considered when building a model that searches for the causes of low yield. However, there have been few studies in this area. This paper proposes a low-yield root cause search method considering the order of processes using a long short-term memory with attention mechanism (LSTM-AM) model. Specifically, the LSTM-AM model is applied to data classified according to the process structure of semiconductor products, and the causes of low yield are determined considering the order of processes by extracting attention weights. Experiments are conducted to verify the suitability of the proposed method using real yield data from a semiconductor company. The experimental results confirm that the proposed method outperforms the existing low yield root cause search methods in terms of low yield prediction. [ABSTRACT FROM AUTHOR]

Data-driven Fault Detection and Classification approaches are becoming increasingly important in semiconductor manufacturing and in other industries aiming at implementing the Zero-defect paradigm. Two of the main challenges in developing such solutions are: (i) the complexity of sensor data, that typically presents themselves in the form of time-series, requiring the employment of time-consuming and possibly sub-optimal feature extraction approaches; (ii) the fact that faults/defects may be caused by more than a single process, but in many cases they are generated by a cascade of processes. In this paper, we tackle the first issue, by considering a two-stage case study consisting of a deposition process and a rapid thermal process. The proposed approach is based on convolutional deep autoencoders employed to perform feature extraction from time-series sensor data in frontend production equipment. We will show on the reported case study, how the proposed approach outperfoms key numbers-based approaches typically used in the industry. To allow reproducibility of the reported results and to foster research in the field, we publicly share the data used in this work. [ABSTRACT FROM AUTHOR]