Semiconductor Etch Process Research Articles

An adaptive imbalance robust graph embedding broad learning system fault diagnosis for imbalanced batch processes data

Modern batch processes develop toward higher levels, and devices usually work at normal conditions with generally very few faults. Therefore, fewer fault data are collected than normal data, which makes the normal and fault modes imbalanced, and the fault diagnosis model is incapable focusing on the minority of fault samples as much as on most normal samples, consequently leading to insufficient generalization ability for the model. For that, we propose an adaptive imbalance-robust graph embedding broad learning system (AI-RGEBLS) in this paper. Firstly, it achieves adaptive correction of unbalanced samples by niche technique and synthetic minority over-sampling technique (SMOTE) with improved Mahalanobis distance. Then, broad learning system (BLS) is regularized by graph embedding, and further introduced the L2,1 norm constraint, which makes it possible to enhance the robustness of the model while considering the local manifold information of the data in the feature extraction process. Finally, the incremental learning approach is applied to the model to avoid the disastrous forgetting problem caused by training the whole model from zero. The effectiveness of the proposed method is verified by penicillin fermentation process and semiconductor etching process. It can effectively improve the speed of model training and provide better fault diagnosis for imbalanced batch processes compared with the existing methods.

Effect of radiofrequency bias power on transmission spectrum of flat-cutoff sensor in inductively coupled plasma

Real-time monitoring of plasma parameters at the wafer plane is important because it significantly affects the processing results, yield enhancement, and device integrity of plasma processing. Various plasma diagnostic sensors, including those embedded in a chamber wall and on-wafer sensors, such as flat-cutoff sensors, have been developed for plasma measurements. However, to measure the plasma density on the wafer surface in real-time when processing plasma with bias power, such as in the semiconductor etching process, one must analyze the transmission spectrum of the flat-cutoff sensor in an environment with bias power applied. In this study, the transmission-spectrum and measured plasma-density characteristics of an electrode-embedded flat-cutoff sensor are analyzed via electromagnetic simulations and experiments under applied bias power. Our findings indicate that the flat-cutoff sensor accurately measures the plasma density, which is equivalent to the input plasma density under low bias power. Conversely, under high bias power, the plasma density measured by the sensor is lower than the input plasma density. Also, a thick-sheath layer is formed owing to the high bias power, which may complicate the measurement of plasma parameters using the flat-cutoff sensor. Plasma diagnostics using a flat-cutoff sensor in thick-sheath environments can be achieved by optimizing the flat-cutoff sensor structure. Our findings can enhance the analysis of plasma parameters on-wafer surfaces in processing environments with bias power applied.

Effect of the Ion Flux Incident on the Wafer Depending on the Structure and Material of the Focus Ring in Capacitively Coupled Plasma Sources for the Semiconductor Etching Process

Effect of the Ion Flux Incident on the Wafer Depending on the Structure and Material of the Focus Ring in Capacitively Coupled Plasma Sources for the Semiconductor Etching Process

Effective thermo-electric-mechanical modeling of capacitively coupled plasma in low-pressure conditions: Modeling and application in dry etching

In semiconductor dry etching process, capacitively coupled plasma (CCP) is widely used in low-pressure environments. However, the kinetics of species particles in low-pressure conditions are different from the general environments. Traditional continuum models are popular in industry due to their simplicity, but they struggle to account for the kinetics of particles in low-pressure conditions. Although particle-in-cell (PIC) modeling is suitable for simulating the behavior of a single particle, it is difficult to perform such industrial simulations due to the high computational cost. To account for both kinetics and engineering applicability, this study presents an effective thermo-electric-mechanical modeling structure of CCP at low-pressure conditions. This modeling structure includes balance laws of continua derived from the global energy equilibrium containing thermal, electric and mechanical energies. The ion momentum, which is ignored in traditional plasma continuum models, is considered in the balance laws, and electron energy distributions are obtained by solving the Boltzmann equation to obtain the properties of electron transport. In addition, the model variables are calculated based on mixed kinetic theory accounting for the fraction of each species. The results of the proposed modeling are compared to those of a PIC model to validate its ability to describe the distribution of electrons with respect to pressure conditions. Based on the proposed numerical formulations, this study built a MATLAB-based program to guide the design of a screen baffle in dry etching equipment. This program captures the critical aspects of the practical application of screen baffle design in the context of the dry etching process. The results of this paper showed that the proposed model can be applied to engineering applications under low-pressure conditions.

High-Speed Design with Separated Tapering for Reducing Cogging Torque and Torque Ripple of a 3 kW Dry Vacuum Pump Motor for the ETCH Process

This paper proposes a design method to reduce cogging torque and torque ripple in the concentrated winding of IPMSMs (Interior Permanent Magnet Synchronous Motors) used in motors for the semiconductor ETCH process. IPMSMs can utilize reluctance torque through the difference in inductance between the d axis and q axis, but they are at a disadvantage in terms of reducing cogging torque while tapering the rotor and stator to reduce torque ripple. In addition, the existing single tapering can push the permanent magnets into the rotor. If the rotor’s permanent magnets are embedded, the magnetic reluctance will increase, and the overall performance of the motor will decrease. However, an optimum design method was derived in which the magnets do not move during rotor tapering. This geometric design is an optimum design method that reduces cogging torque and torque ripple. This paper compares and analyzes four models, the concentrated winding model, distributed winding model, conventional tapering model, and separated tapering model, using 2D and 3D finite element analysis (FEA).

SiameseDeNPEnetwork framework for fault detection of batch process

AbstractIn batch processes, it is crucial to ensure safe production by fault detection. However, the long batch duration, limited runs, and strong nonlinearity of the data pose challenges. Incipient faults with small amplitudes further complicate the detection process. To achieve safe production, motivated by deep learning strategies, we propose a new fault detection method of batch process called Siamese deep neighbourhood preserving embedding network (SDeNPE). First, the DeNPE network is constructed by means of NPE and kernel functions, which utilizes the different types of kernel functions in the kernel mapping layer to extract diverse deep nonlinear features and overcome strong nonlinearity in the process data. Then, the Siamese network is used to obtain the different features between the data and improve the recognition of incipient faults. In addition, the deep extraction and Siamese network allow for batches of training data reduction without diminishing the performance of fault detection. Finally, we utilize monitoring statistics to complete the fault detection process. Two batch process cases involving the penicillin fermentation process and the semiconductor etching process demonstrate the superior fault detection performance of the proposed SDeNPE over the other comparison methods.

Multimode process monitoring strategy based on improved just‐in‐time‐learning associated with locality preserving projections

AbstractIn this paper, a multimode process monitoring strategy based on improved just‐in‐time‐learning associated with locality preserving projections (IJITL‐LPP) is proposed. First, raw data are projected into the feature space using locality preserving projections (LPP). Second, IJITL searches for similar samples of the query sample in the feature space by introducing a variational inference Gaussian mixture model (VIGMM). Finally, the new statistic named average distance is created to complete process monitoring. In the IJITL, the introduced VI can automatically determine the number of modes, thereby accelerating the efficiency of selecting similar samples. In the process monitoring phase, the average distance can reduce the impact of different mode dispersion on fault detection. In addition, LPP can render the model less sensitive to outliers. Compared with principal component analysis (PCA), LPP, K nearest neighbour rules, Gaussian mixture model (GMM), K‐means based‐PCA, and just‐in‐time‐learning (JITL)‐based LPP, the proposed method has better performance in a numerical case, the Tennessee Eastman process, and the semiconductor etching process.

Multimode Process Monitoring Based on Modified Density Peak Clustering and Parallel Variational Autoencoder

Clustering algorithms and deep learning methods have been widely applied in the multimode process monitoring. However, for the process data with unknown mode, traditional clustering methods can hardly identify the number of modes automatically. Further, deep learning methods can learn effective features from nonlinear process data, while the extracted features cannot follow the Gaussian distribution, which may lead to incorrect control limit for fault detection. In this paper, a comprehensive monitoring method based on modified density peak clustering and parallel variational autoencoder (MDPC-PVAE) is proposed for multimode processes. Firstly, a novel clustering algorithm, named MDPC, is presented for the mode identification and division. MDPC can identify the number of modes without prior knowledge of mode information and divide the whole process data into multiple modes. Then, the PVAE is established based on distinguished multimode data to generate the deep nonlinear features, in which the generated features in each VAE follow the Gaussian distribution. Finally, the Gaussian feature representations obtained by PVAE are provided to construct the statistics H2, and the control limits are determined by the kernel density estimation (KDE) method. The effectiveness of the proposed method is evaluated by the Tennessee Eastman process and semiconductor etching process.

Development of Virtual Metrology Using Plasma Information Variables to Predict Si Etch Profile Processed by SF6/O2/Ar Capacitively Coupled Plasma.

In the semiconductor etch process, as the critical dimension (CD) decreases and the difficulty of the process control increases, in-situ and real-time etch profile monitoring becomes important. It leads to the development of virtual metrology (VM) technology, one of the measurement and inspection (MI) technology that predicts the etch profile during the process. Recently, VM to predict the etch depth using plasma information (PI) variables and the etch process data based on the statistical regression method had been developed and demonstrated high performance. In this study, VM using PI variables, named PI-VM, was extended to monitor the etch profile and investigated the role of PI variables and features of PI-VM. PI variables are obtained through analysis on optical emission spectrum data. The features in PI-VM are investigated in terms of plasma physics and etch kinetics. The PI-VM is developed to monitor the etch depth, bowing CD, etch depth times bowing CD (rectangular model), and etch area model (non-rectangular model). PI-VM for etch depth and bowing CD showed high prediction accuracy of R-square value (R2) 0.8 or higher. The rectangular and non-rectangular etch area model PI-VM showed prediction accuracy R2 of 0.78 and 0.49, respectively. The first trial of virtual metrology to monitor the etch profile will contribute to the development of the etch profile control technology.

Utilization of TRIZ to Solve the Quality Problems in Semiconductor Etching Process

Utilization of TRIZ to Solve the Quality Problems in Semiconductor Etching Process

Multistage Condition Monitoring of Batch Process Based on Multi-boundary Hypersphere SVDD with Modified Bat Algorithm

Multistage characteristic has become one of the essential issues of batch process and several stage division approaches have been introduced to monitor the process. As the non-Gaussian and nonlinear problems of batch process, a hybrid intelligent method is developed to monitor the multistage conditions in this paper. The proposed algorithm includes converged stage division (CSD), multi-boundary hypersphere support vector data description (MH-SVDD), and modified bat algorithm (MBA). CSD algorithm is utilized to process the data and make the stage division, which consists of data length processing, three-dimension unfolding, and K-means clustering. MH-SVDD algorithm is to construct two hyperspheres, which can overcome the deficiency of traditional boundary SVDD. The Gaussian kernel function width parameter of MH-SVDD plays a very significant role in multistage fault monitoring, a modified bat algorithm is established to select the optimal parameter. The experimental of the semiconductor etching process is described, and the results demonstrate that the proposed model can gain higher fault monitoring accuracy in multistage condition monitoring of the batch process.

A Novel Monitoring Strategy Combining the Advantages of NPE and GMM

Semiconductor manufacturing process data usually have multimodel and multiphase characteristics which do not meet the application assumptions in neighborhood preserving embedding (NPE). Aiming at the above limitations of NPE, a novel monitoring strategy combining the advantages of the neighborhood preserving embedding and Gaussian mixture model(NPE-GMM) is proposed. Firstly, the window data are obtained by the default window width. Next, the score of the current window data set are calculated by NPE. After that, some Gaussian components of the score are determined by GMM. Finally, a quantification index is proposed to monitor process status. NPE-GMM can not only maintain more local structure information of the current window data set in the feature subspace, but also reduce the computational complexity of GMM in fault detection processes. By introducing the new statistic, NPE-GMM can effectively improve the fault detection rate of some multimodel batch processes. The effectiveness of the proposed method is verified in a numerical case and the semiconductor etching process. The simulation results indicated that the proposed method has a higher fault detection rate than traditional methods.

Surface coupling of plasma optical emission spectra with bent metal-clad waveguide

We present the design of compact bent metal-clad waveguides to collect the vertically emitted light and to transmit to the horizontal direction. The intention of the design is to collect the plasma optical emission spectra inside the semiconductor etch process chamber to monitor plasma gas phase chemistry resides in the process chamber. Coupler A has been designed for visible spectra range from 748 to 753 nm at which the maximum coupling efficiency is over 52%, while Coupler B operates at the wavelength range from 775 to 780 nm with coupling strength of 62% and Coupler C covers the visible spectrum from 808 to 813 nm with its maximum efficiency of 60%. The surface couplers reveal good spectral characteristics, considerable alignment tolerance and minimal size of 9 × 5 × 1.6 µm3 each. Relatively simple geometrical structure over conventional grating coupler is advantageous to lower the fabrication cost as well.

Design of a self-tuning adaptive model predictive controller using recursive model parameter estimation for real-time plasma variable control

The semiconductor etching process, which is the most important part of the semiconductor manufacturing process, requires higher sophistication as 10 nm semiconductors are mass produced. Etching methods utilizing plasma are getting increasingly popular with the miniaturization of the etching process. As the process performance depends on the state of the plasma variables, such as electron density, it is essential to measure and control these variables in real time. Moreover, to control the plasma-based system, the sensitive and time-varying characteristics of plasma should be considered. This paper proposes a self-tuning adaptive model predictive controller that can effectively perform electron density control. As a first step, an integral squared error-based Bayesian optimization is used to tune the model predictive controller, and its performance is verified on a drift-free Ar plasma system. The self-tuning adaptive model predictive controller is constructed by combining a recursive model parameter estimator with a model predictive controller. The recursive model parameter estimator is designed using a recursive least squares algorithm with Kalman filter interpretation. The effectiveness of the proposed controller is verified through control simulations and a set-point tracking experiment on the electron density with artificial drift in real time. The experimental results show that the performance of the proposed controller is 21% better than that of the conventional model predictive controller. We expect that this result will make a significant contribution to control processes utilizing plasma.

Incipient Fault Detection for Multiphase Batch Processes With Limited Batches

Sufficient batches are in general, required for fault detection of batch processes. However, sometimes, it is difficult and may be impractical to conduct multiple cycles and wait until enough batches are available. Without batch-wise normalization step, the nonstationary cannot be efficiently removed by time-wise normalization. The mean values of the nonstationary variables are still time varying and the interval of fault-free data is very wide in each phase. Therefore, the incipient fault, which has a small magnitude including early changing and the slow developing, may be buried by nonstationary trends resulting in low fault detection rate. To address the above issue, a two-layer fault detection method is proposed to detect the incipient fault for multiphase batch processes with limited batches. First, a concurrent variable separation strategy is proposed to distinguish nonstationary variables from stationary variables for multiple batches in each phase. Second, a two-layer fault detection model is constructed to detect the incipient fault. Cointegration analysis-based fault detection model is built to investigate the relationship between nonstationary variables, which can effectively distinguish the incipient fault from the normal trend of the nonstationary variables. Principal component analysis is adopted to describe the correlation of stationary variables. Afterward, a total fault detection model is constructed to monitor the relation between nonstationary variables and stationary variables. To illustrate the feasibility and effectiveness, the proposed algorithm is applied to two multiphase batch processes including fed-batch penicillin fermentation process and semiconductor etch process.

Dynamic hypersphere based support vector data description for batch process monitoring

Support Vector Data Description (SVDD) is an efficient monitoring method that captures the spherically shaped boundary around the normal batch data and sets the control limit related to support vectors (SVs) for online monitoring. Using nonlinear transformation functions, SVDD constructs an irregular hypersphere in high dimensional space. When the batch process is complicated, the accuracy of monitoring will decrease with traditional control limit of SVDD. In this paper, dynamic hypersphere based support vector data description (DH-SVDD) is proposed for batch process monitoring. In training process, static hypersphere is built by the important SVs of training dataset. In testing process, dynamic hypersphere is built by the important SVs of combined dataset with current test sample and training dataset. If there is a significant change between these two hyperspheres, it means that the current test sample is an outlier. Thus, DH-SVDD has a relatively high monitoring accuracy because it fully considers relationship between the current test sample and the historical training dataset in high dimensional space. Comparison is conducted between the proposed DH-SVDD and traditional methods such as K-chart-SVDD, max limit SVDD and validation limit SVDD. The effectiveness of the DH-SVDD is also verified by a semiconductor etch process and a fed-batch penicillin fermentation process.

A geometric method for batch data visualization, process monitoring and fault detection

Batch processing is used extensively in the production of high value products, and there are strong economic incentives for developing methodologies for ensuring the successful completion of batches via process monitoring and fault detection. Building on our previous work using time-explicit Kiviat diagrams for continuous processes, this paper introduces data visualization, data-driven process monitoring and fault detection for batch systems. Handling batch data, including unfolding and alignment are addressed as well. The proposed methodology is demonstrated on data obtained from a benchmark bioreactor simulator and a semiconductor etching process.

Automatic Virtual Metrology and Target Value Adjustment for Mass Customization

One of the core values of Industry 4.0 targets to integrate people's demand into manufacturing for enhanced products, systems, and services for a wider variety of increasingly personalized customization of products. Thus, Industry 4.0 advances the traditional manufacturing techniques from mass production toward mass customization (MC). Take wheel machining automation (WMA) as an example. To meet MC expectations, manufacturers should offer customized wheels at a large scale with low cost, short lead-time, and high quality. Thus, WMA cells with MC capability are requested to be designed to have a high degree of quick responsiveness to accurately react on machining-condition changes for manufacturing different wheel types. To meet the requirements of MC production, this work proposes to apply the automatic virtual metrology (AVM) system together with the so-called target-value-adjustment (TVA) scheme. The AVM system was developed by the authors to convert sampling inspections with metrology delay into real-time and on-line total inspection, while the TVA scheme is designed to enhance AVM's adaptive customization capability for automatically and rapidly accomplishing the goals of MC production. To demonstrate the versatility of the AVM-plus-TVA approach, the other example in semiconductor etching process is also illustrated.

Fault detection of batch processes based on multivariate functional kernel principal component analysis

In some batch processes, the variables' trajectories often show obviously functional nature and can be considered as smooth functions rather than just vectors, then the collected three-way data array essentially can be transformed into a two-way function matrix. For this purpose, the approach of functional data analysis (FDA) is introduced to express the variables' functions in this article, which can highlight the subtle differences in the shape of variables' trajectories between normal batches and faulty ones, and can also easily handle the problems of irregular three-way data array, such as unequal batch length, different sampling rates, missing data, etc. Based on the function matrix, a novel monitoring method called multivariate functional kernel principal component analysis (MFKPCA) is proposed for fault detection of batch processes, which directly eigen-decomposes the two-way function matrix in the function space and uses the kernel trick to handle the nonlinear correlations. Different from the traditional PCA method, MFKPCA designs three complementary control charts for batch process monitoring based on three statistics including Hotelling's T2, squared prediction error (SPE) and mean squared error (MSE). Finally, some simulations and an industrial semiconductor etch process are used to illustrate the validity of the proposed monitoring method.

Fault Detection in the Semiconductor Etch Process Using the Seasonal Autoregressive Integrated Moving Average Modeling

Abstract —In this paper, we investigated the use of seasonal autoregressive integrated moving average (SARIMA) time series models for fault detection in semiconductor etch equipment data. The derivative dynamic time warping algorithm was employed for the synchronization of data. The models were generated using a set of data from healthy runs, and the established models were compared with the experimental runs to find the faulty runs. It has been shown that the SARIMA modeling for this data can detect faults in the etch tool data from the semiconductor industry with an accuracy of 80% and 90% using the parameter-wise error computation and the step-wise error computation, respectively. We found that SARIMA is useful to detect incipient faults in semiconductor fabrication. Keywords —Autoregressive Integrated Moving Average, Dynamic Time Warping, Fault Detection, Seasonal Autoregressive Integrated Moving Average, Semiconductor Process, Time Series Modeling 1. I NTRODUCTION Over the past decade of continuous improvements being made to semiconductor technologies, the device feature size has been scaled down tremendously. As a result, the processes have become more and more complex, and this calls for a much tighter process control than before. The faults in the manufacturing processes must be detected with precision to reduce the loss in manufacturing. Semiconductor equipment data has been shown to be helpful for the detection of any abnormal variation in the process. This is because any fault occurring during the process appears as some variation in the equipment tool data collected for that process. There has been active research conducted in the fault detection area by applying different statistical techniques to the tool data. Neural networks have been one of the most researched techniques for the purpose [1-3]. Principal component analysis (PCA) and its variants have been widely employed as well [4,5]. Control charts have also been used for simultaneous fault detection and classification [6]. The use of autoregressive moving average based time series modeling techniques have also been shown to be quite useful in this regard [7,8]. These works have shown that the analysis of tool data gives greater insight into the process variations and can be used to detect abnormalities in the process. In this research, the seasonal autoregressive integrated moving average (SARIMA) time series models are investigated for the detection of faults in etch tool data. The derivative dynamic time