Semiconductor Manufacturing Process Research Articles

Application of 1D ResNet for Multivariate Fault Detection on Semiconductor Manufacturing Equipment.

Amid the ongoing emphasis on reducing manufacturing costs and enhancing productivity, one of the crucial objectives when manufacturing is to maintain process tools in optimal operating conditions. With advancements in sensing technologies, large amounts of data are collected during manufacturing processes, and the challenge today is to utilize these massive data efficiently. Some of these data are used for fault detection and classification (FDC) to evaluate the general condition of production machinery. The distinctive characteristics of semiconductor manufacturing, such as interdependent parameters, fluctuating behaviors over time, and frequently changing operating conditions, pose a major challenge in identifying defective wafers during the manufacturing process. To address this challenge, a multivariate fault detection method based on a 1D ResNet algorithm is introduced in this study. The aim is to identify anomalous wafers by analyzing the raw time-series data collected from multiple sensors throughout the semiconductor manufacturing process. To achieve this objective, a set of features is chosen from specified tools in the process chain to characterize the status of the wafers. Tests on the available data confirm that the gradient vanishing problem faced by very deep networks starts to occur with the plain 1D Convolutional Neural Network (CNN)-based method when the size of the network is deeper than 11 layers. To address this, a 1D Residual Network (ResNet)-based method is used. The experimental results show that the proposed method works more effectively and accurately compared to techniques using a plain 1D CNN and can thus be used for detecting abnormal wafers in the semiconductor manufacturing industry.

Optimal Ventilation Design for Flammable Gas Leaking from Gas Box Used in Semiconductor Manufacturing: Case Study on Korean Semiconductor Industry

Highly flammable substances such as hydrogen and silane are used in the semiconductor manufacturing process. When gas leaks, it is mixed with outside air and connected to a treatment facility through the duct inside the gas box. This study investigated optimal exhaust design to prevent fire explosions and health problems by optimizing the exhaust volume when hydrogen leaks from the gas box of semiconductor manufacturing equipment. After selecting the leakage rate amount based on the KS C IEC 60079-10-1, SEMI S6-0707E, and SEMI F-15 standards, a gas box was manufactured. Subsequently, the fan speed required to ventilate the gas box more than five times per minute according to the SEMI standard and the opening area and location that can reduce the lower explosive limit (LEL) to less than 25% in the event of hydrogen leakage were determined. When the air intakes were placed on the left and right, the flow rate was measured at 32 L per minute (LPM), and the maximum concentration was measured at 9111 ppm. This is less than 25% of the LEL of hydrogen and is believed to be capable of preventing fire and explosion, even if a similarly flammable gas leaks inside the gas box.

Discrete Active Disturbance Rejection Control for Semiconductor Manufacturing Processes With Dynamic Models

Discrete Active Disturbance Rejection Control for Semiconductor Manufacturing Processes With Dynamic Models

Scalable photonic integrated circuits for high-fidelity light control

Advances in laser technology have driven discoveries in atomic, molecular, and optical (AMO) physics and emerging applications, from quantum computers with cold atoms or ions, to quantum networks with solid-state color centers. This progress is motivating the development of a new generation of optical control systems that can manipulate the light field with high fidelity at wavelengths relevant for AMO applications. These systems are characterized by criteria: (C1) operation at a design wavelength of choice in the visible (VIS) or near-infrared (IR) spectrum, (C2) a scalable platform that can support large channel counts, (C3) high-intensity modulation extinction and (C4) repeatability compatible with low gate errors, and (C5) fast switching times. Here, we provide a pathway to address these challenges by introducing an atom control architecture based on VIS-IR photonic integrated circuit (PIC) technology. Based on a complementary metal–oxide–semiconductor fabrication process, this atom-control PIC (APIC) technology can meet system requirements (C1)–(C5). As a proof of concept, we demonstrate a 16-channel silicon-nitride-based APIC with (5.8±0.4)ns response times and >30dB extinction ratio at a wavelength of 780 nm.

Similarity searching for fault diagnosis of defect patterns in wafer bin maps

Due to the growing complexity of processes in semiconductor manufacturing, high volumes of data are automatically generated, resulting in a greater challenge for fault detection and yield improvement. Wafer bin maps (WBMs), which represent the spatial distribution of defective dies on the wafer, may contain specific defect patterns that offer useful insight into the underlying causes of anomalies in the processes. Hence, the identification of these defect patterns helps the early detection and diagnosis of the faults. Nowadays, rare and mixed-type defects are more frequently observed, which increases the difficulty for the recognition of defect patterns. In this study, we propose a similarity searching approach for the identification of defect patterns and their potential causes. The comparison of similar patterns provides valuable information to trace the problems in the process history which may narrow the scope of troubleshooting. In particular, the tensor voting algorithm is applied to highlight the structural information in the patterns, and then the weighted best-buddies similarity (WBBS) is proposed to measure the degree of similarity. Experimental results verify the effectiveness of the proposed method in the context of both single and mixed-type defect patterns.

Soft-Sensing Regression Model: From Sensor to Wafer Metrology Forecasting.

The semiconductor industry is one of the most technology-evolving and capital-intensive market sectors. Effective inspection and metrology are necessary to improve product yield, increase product quality and reduce costs. In recent years, many types of semiconductor manufacturing equipments have been equipped with sensors to facilitate real-time monitoring of the production processes. These production-state and equipment-state sensor data provide an opportunity to practice machine-learning technologies in various domains, such as anomaly/fault detection, maintenance scheduling, quality prediction, etc. In this work, we focus on the soft-sensing regression problem in metrology systems, which uses sensor data collected during wafer processing steps to predict impending inspection measurements that used to be measured in wafer inspection and metrology systems. We proposed a regressor based on Long Short-term Memory network and devised two distinct loss functions for the purpose of the training model. Although the assessment of our prediction errors by engineers is subjective, a novel piece-wise evaluation metric was introduced to evaluate model accuracy in a mathematical way. Our experimental results showcased that the proposed model is capable of achieving both accurate and early prediction across various types of inspections in complicated manufacturing processes.

Artificial-Neural-Network-Driven Innovations in Time-Varying Process Diagnosis of Low-K Oxide Deposition.

To address the challenges in real-time process diagnosis within the semiconductor manufacturing industry, this paper presents a novel machine learning approach for analyzing the time-varying 10th harmonics during the deposition of low-k oxide (SiOF) on a 600 Å undoped silicate glass thin liner using a high-density plasma chemical vapor deposition system. The 10th harmonics, which are high-frequency components 10 times the fundamental frequency, are generated in the plasma sheath because of their nonlinear nature. An artificial neural network with a three-hidden-layer architecture was applied and optimized using k-fold cross-validation to analyze the harmonics generated in the plasma sheath during the deposition process. The model exhibited a binary cross-entropy loss of 0.1277 and achieved an accuracy of 0.9461. This approach enables the accurate prediction of process performance, resulting in significant cost reduction and enhancement of semiconductor manufacturing processes. This model has the potential to improve defect control and yield, thereby benefiting the semiconductor industry. Despite the limitations imposed by the limited dataset, the model demonstrated promising results, and further performance improvements are anticipated with the inclusion of additional data in future studies.

Wafer Image Preprocessing Based on Density Based Spatial Clustering of Application with Noise

With the development of microelectronic manufacturing technology, semiconductor manufacturing presents the trend of maximization of scale and miniaturization of process size. Even if the wafer production now has a high automation of the production process, high-precision production equipment and advanced production technology, wafer abnormal situation is still inevitable. Abnormal conditions in semiconductor manufacturing process will reduce the yield of wafer products and increase production costs. The later analysis becomes the necessary means to improve the wafer yield. With the rapid development of modern computing power, the application of machine learning-based automatic detection method in semiconductor production is irresistible. In addition to its spatial pattern, there are many noises that can affect the classification of defects, so it is necessary to preprocess the wafer diagram. The traditional density based spatial clustering of application with noise (DBSCAN) algorithm needs to determine two clustering parameters artificially, and the choice of parameters can affect the clustering effect easily, the parameter list was obtained by K-mean nearest neighbor algorithm and mathematical expectation method, and the integrated parameters of intra-cluster density and inter-cluster density after DBSCAN clustering were selected as evaluation indexes to select the optimal parameters. Finally, a self-adaptive (SA)-DBSCAN map is obtained by retaining the largest cluster, adding feature points and feature clusters, thus improving the yield of wafer.

The Impact of Processing Time Variations on Swap Sequence Performance in Dual-Armed Cluster Tools

The performance of a swap sequence is analyzed by assuming cyclic scheduling in dual-armed cluster tools with processing time variations. A dual-armed cluster tool consists of multiple processing modules (PMs), one material handling robot that can hold two wafers at the same time, and loadlocks where wafer cassettes are loaded or unloaded. The swap sequence in a dual-armed cluster tool is widely used in practice and known to be optimal with deterministic processing times when the bottleneck PM’s workload is larger than the robot workload. However, in practice, processing times on a PM can have a small variation, which leads to a different processing time of each wafer on the PM. Hence, when the processing time variation is introduced, the performance of the swap sequence needs to be analyzed. This paper first defines a fundamental cycle and analyzes its cycle time. It then proposes optimality conditions of the swap sequence and performs numerical experiments to show the effectiveness of the sequence. <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Note to Practitioners</i> —A dual-armed cluster tool used for semiconductor manufacturing processes is usually operated with a swap sequence because it is simple, easy to control, and proven to be optimal with deterministic processing times. However, studies on the performance of the swap sequence are still limited with processing times varying in PMs which often occur in practice. Hence, this study shows the effectiveness of the swap sequence with the processing time variations by analyzing cycle times, optimality conditions, and performing numerical experiments.

Integration of Bayesian Adaptive Exponentially Weighted Moving Average Control Chart and Paired Ranked-Based Sampling for Enhanced Semiconductor Manufacturing Process Monitoring

Exponentially weighted moving average (EWMA) and Shewhart control charts are commonly utilized to detect the small to moderate and large shifts in the process mean, respectively. This article introduces a novel Bayesian AEWMA control chart that employs various loss functions (LFs), including square error loss function (SELF) and LINEX loss function (LLF). The control chart incorporates an informative prior for posterior and posterior predictive distributions. Additionally, the control chart utilizes various paired ranked set sampling (PRSS) schemes to improve its accuracy and effectiveness. The average run length (ARL) and standard deviation of run length (SDRL) are used to evaluate the performance of the suggested control chart. Monte Carlo simulations are conducted to compare the performance of the proposed approach to other control charts. The results show that the proposed method outperforms in identifying out-of-control signals, particularly under PRSS schemes compared to simple random sampling (SRS). The proposed CCs effectiveness was validated using a real-life semiconductor manufacturing application, utilizing different PRSS schemes. The performance of the Bayesian AEWMA CC was evaluated, demonstrating its superiority in detecting out-of-control signs compared to existing CCs. This study introduces an innovative method incorporating various LFs and PRSS schemes, providing an enhanced and efficient approach for identifying shifts in the process mean.

Dynamic mode decomposition with memory.

This study proposed a numerical method of dynamic mode decomposition with memory (DMDm) to analyze multidimensional time-series data with memory effects. The memory effect is a widely observed phenomenon in physics and engineering and is considered to be the result of interactions between the system and environment. Dynamic mode decomposition (DMD) is a linear operation-based, data-driven method for multidimensional time-series data proposed in 2008. Although DMD is a successful method for time-series data analysis, it is based on ordinary differential equationsand thus cannot incorporate memory effects. In this study, we formulated the abstract algorithmic structure of DMDm and demonstrate its utility in overcoming the memoryless restriction imposed by existing DMD methods on the time-evolution model. In the numerical demonstration, we utilized the Caputo fractional differential to implement an example of DMDm such that the time-series data could be analyzed with power-law memory effects. Thus, we developed a fractional DMD, which is a DMD-based method with arbitrary (real value) order differential operations. The proposed method was applied to synthetic data from a set of fractional oscillators and model parameters were estimated successfully. The proposed method is expected to be useful for scientific applications and aid in model estimation, control, and failure detection of mechanical, thermal, and fluid systems in factory machines, such as modern semiconductor manufacturing equipment.

Supervised contrastive learning for wafer map pattern classification

In the semiconductor manufacturing process, analyzing the defect patterns on a wafer map is crucial for identifying the causes of the defects. The advent of convolutional neural networks (CNNs) has significantly increased the accuracy of automated wafer map pattern classification. Generally, the use of a larger training dataset results in higher classification accuracy. However, collecting a large number of wafer maps and labeling them with their defect categories is expensive and time-consuming. In this paper, we present an improved training method under data insufficiency for wafer map pattern classification. We apply supervised contrastive learning to train a CNN by exploiting the rotational-invariant characteristic of wafer map labeling. The CNN is trained by simultaneously minimizing two loss functions: classification loss and contrastive loss. The first loss function is to classify the rotational variants of wafer maps accurately. The second loss function is to align the representation vectors for the rotational variants of wafer maps with similar labels to be close to each other. Using two benchmark datasets, WM-811K and MixedWM38, we demonstrate that the proposed method enhances classification accuracy compared with existing methods, particularly when the training dataset is small.

Physiochemical etching characteristics and surface analysis of Y2O3-MgO nanocomposite under different CF4/Ar/O2 plasma atmospheres

During plasma etching in the semiconductor manufacturing process, not only the wafers but also the ceramic components that make up the interior of the chamber are influenced by the highly reactive plasma. Because chemical and physical etching work concurrently, and selective etching is possible by controlling the plasma gas composition, it is essential to develop novel ceramics with outstanding resistance to various plasma environments that do not produce contamination particles; this will allow us to elevate production yield of chips and overall equipment effectiveness (OEE). Herein, we report the plasma etching behavior of 50:50 vol% Y2O3-MgO nanocomposite ceramics under different CF4/Ar gas ratios conditions. It was found that the erosion depths and surface roughness changes of this material were much lower than those values of reference materials under both chemical and physical etching. Effects of plasma environment on the penetration depth of fluorine on the nanocomposite surface and of physical sputtering attack were also investigated. The results indicate that a nanocomposite with fine grains and high density can decrease particle generation and improve its life cycle under strong bias voltage and various plasma gas conditions, thereby showing itself a promising material for next-generation ceramic components in the semiconductor manufacturing process.

A novel carbon reduction engineering method-based deep Q-learning algorithm for energy-efficient scheduling on a single batch-processing machine in semiconductor manufacturing

The semiconductor industry is a resource-intensive sector that heavily relies on energy, water, chemicals, and raw materials. Within the semiconductor manufacturing process, the diffusion furnace, ion implantation machine, and plasma etching machine exhibit high energy demands or operate at extremely high temperatures, resulting in significant electricity consumption, which is usually carbon-intensive. To address energy conservation concerns, the industry adopts batch production technology, which allows for the simultaneous processing of multiple products. The energy-efficient parallel batch scheduling problem arises from the need to optimise product grouping and sequencing. In contrast to existing heuristics, meta-heuristics, and exact algorithms, this paper introduces the Deep Q-Network (DQN) algorithm as a novel approach to address the proposed problem. The DQN algorithm is built upon the agent’s systematic learning of scheduling rules, thereby enabling it to offer guidance for online decision-making regarding the grouping and sequencing of products. The efficacy of the algorithm is substantiated through extensive computational experiments.

All-dielectric terahertz metalens using 3D-printing

Terahertz (THz) electromagnetic waves offer a new way of understanding various phenomena, owing to their non-ionizing properties resulting from their low energy, and non-destructive inspection features in dielectrics. Although THz emitters and detectors have been developed with much interest, the performance of optics in the THz band is still limited by the properties of the available materials. Metamaterials, which are the artificial subwavelength structures for implementing optical properties that do not exist in intrinsic materials, proposed as a new approach. However, their fabrication requires a complex semiconductor manufacturing process for its elaborate structure. In this study, we demonstrated the fabrication of all-dielectric THz metamaterial optics using an accessible 3D printing technology. A meta-element consisting of a square pillar exhibited full-wave phase controllability. The metalens was designed for 0.2 THz. The metalens fabricated using a low-cost LCD-type 3D printer and photopolymer resin showed good focusing performance and polarization insensitivity.

GC-ICP-MS for Process and Quality Control in Semiconductor Manufacturing

Gas chromatography (GC) coupled to inductively coupled plasma–mass spectrometry (ICP-MS) is a powerful and flexible analytical solution that is well-established for measuring organometallic compounds such as organotin, organomercury, and organolead in environmental samples, foodstuffs, and consumer products. GC–ICP-MS is also used for industrial applications, such as monitoring catalyst poisons and environmental contaminant elements in petrochemical processing and plastics manufacturing. The semiconductor industry uses a range of high-purity specialty gases and volatile liquids as precursors in wafer substrate production, and for processes such as plasma etching and deposition of thin films by chemical vapor deposition (CVD). Semiconductor manufacturers are constantly developing new integrated circuit (IC) chips with smaller sizes, higher speeds, lower power consumptions, and greater transistor densities. This trend means that contaminants must be controlled at even lower levels in process chemicals and precursors. In this article, we show how GC–ICP-MS, particularly using triple quadrupole or tandem ICP-MS (ICP-MS/MS), enables determination of the lowest levels of contaminants in the specialty gases and volatile liquids used to make the most advanced electronic devices.

The Effectiveness of Pressure Safety Valves in Chemical Supply Systems to Prevent Fire, Explosion, and Overpressure in the Korean Semiconductor Industry

This study was conducted to review the safety and appropriateness of PSV (Pressure Safety Valve) installation in the supply tank, which is a pressure vessel included in supply systems dedicated to supplying the acid/alkaline substances used in the Korean semiconductor manufacturing process. Three aspects of design, risk assessment, and regulations were reviewed to determine if there is a source of pressure higher than the maximum allowable working pressure (MAWP) of the supply tank that could cause fires, explosions, and overpressure. In the case of the design review, all 17 overpressure scenarios described in API Standard 521, i.e., pressure-relieving and depressuring systems, were reviewed, and there was no overpressure scenario above the maximum allowable working pressure (MAWP). Then, the risk assessment, i.e., the Hazard and Operability Study (HAZOP) technique, was used, and as a result of reviewing all possible risk situations, we can state that there were no overpressure scenarios that can exceed the design pressure of the supply tank; thus, we decided that the installation of a PSV on top of the supply tank is unnecessary. Finally, accident prevention measures against overpressure, such as the KS B 6750-3 system design and the Korean Industrial Standard, were reviewed from a legal point of view. It was confirmed that the hazardous chemical supply system for the semiconductor industry designed in this study has several protective functions to prevent fires, explosions, and overpressure. As a result of reviewing the above three aspects, it can be said that there is no need to install a pressure safety valve in a pressure vessel storing hazardous chemicals.

Orientation-dependent etching of silicon by fluorine molecules: A quantum chemistry computational study

Anisotropic etching is a widely used process in semiconductor manufacturing, in particular, for micro- and nanoscale texturing of silicon surfaces for black silicon production. The typical process of plasma-assisted etching uses energetic ions to remove materials in the vertical direction, creating anisotropic etch profiles. Plasmaless anisotropic etching, considered here, is a less common process that does not use ions and plasma. The anisotropy is caused by the unequal etching rates of different crystal planes; the etching process, thus, proceeds in a preferred direction. In this paper, we have performed quantum chemistry modeling of gas-surface reactions involved in the etching of silicon surfaces by molecular fluorine. The results confirm that orientation-dependent etch rates are the reason for anisotropy. The modeling of F2 dissociative chemisorption on F-terminated silicon surfaces shows that Si–Si bond breaking is slow for the Si(111) surface, while it is fast for Si(100) and Si(110) surfaces. Both Si(100) and Si(110) surfaces incorporate a larger number of fluorine atoms resulting in Si–Si bonds having a larger amount of positive charge, which lowers the reaction barrier of F2 dissociative chemisorption, yielding a higher etch rate for Si(100) and Si(110) surfaces compared to Si(111) surfaces. Molecular dynamics modeling of the same reactions has shown that the chosen reactive bond order potential does not accurately reproduce the lower reaction barriers for F2 dissociative chemisorption on Si(100) and Si(100) surfaces. Thus, reparameterization is necessary to model the anisotropic etching process that occurs at lower temperatures.

Pulsed Rapid Thermal Annealing as a Low Thermal Budget Semiconductor Fabrication Process

Thermal annealing is required in the fabrication of many semiconductor devices. A proper annealing condition, e.g., temperature, atmosphere, and time, can improve the device performance, e.g., by reducing defect densities in bulk films and at interfaces. In addition, the annealing step can generate new structures in previously deposited thin films for various new functions. However, when the annealing condition is improper, e.g., the temperature is too high or the annealing time is too long, the device performance can deteriorate or dysfunction due to damage to the composing material, structure, or even the substrate. The thermal budget is critical in the thermal annealing step.Pulsed rapid thermal annealing (PRTA), which exposes the sample to repeated cycles of short-period heating and cooling, is a low thermal budget process. In this presentation, the principle of PRTA is presented. Examples of using PRTA in the crystallization of amorphous silicon on the glass substrate to polycrystalline thin films for thin film transistors (TFTs) and solar cells, improving high-k gate dielectric properties for MOS capacitors and solid state incandescent light emitting devices (SSI-LEDs), and preparation of nanocrystals embedded high-k structure for nonvolatile memories will be shown (1-4). Limitations of PRTA and possible improvements will be discussed. Kuo and P. M. Kozlowski, Appl. Phys. Lett., 69(8), 1092-1094 (1996).Kuo, C.-H. Lin, and M. Zhu, IEEE Photovoltaic Specialist Conf. PVSC 35, 3698-3701 (2010).Y. Kuo and C. H. Lin, ECS 237th Meeting # 130648, May 10-15, 2020.Y. Kuo, ECS 227th Meeting # 47306 (1333), May 24-28, 2015.

A nonparametric EWMA control chart for monitoring mixed continuous and count data

ABSTRACT Conventional statistical process control tools monitor either continuous or count data but rarely both simultaneously. While process data are becoming increasingly complex, there will be more data points containing both continuous and count information. In the case of mixed continuous and count data with unknown distributions, the traditional parameter control chart cannot be used to monitor them. It is proposed in this paper a novel nonparametric EWMA control chart to monitor mixed continuous and count data. The mixed continuous and count data are first transformed into categorical data, and then a log-linear model is utilized to analyze correlations between variables, followed by the construction of an EWMA statistic that is used to monitor mixed continuous and count data. Next, the proposed control chart is compared with several improved control charts for monitoring mixed continuous and count data. Based on the numerical simulation results, the control chart presented in this paper provides a superior method of detecting alarm signals in the process compared to some improved control charts. Finally, the proposed control chart is demonstrated to be effective and applicable using the semiconductor manufacturing process dataset from the UC Irvine Machine Learning Repository.