Etch rate prediction in plasma etching using feed forward Error-Back Propagation neural network model

To study the residual settlement of goaf’s law and prediction model, we investigated the Mentougou mining area in Beijing as an example. Using MATLAB software, the wavelet threshold denoising method was used to optimize measured data, and the grey model (GM) and feed forward back propagation neural network model (FFBPNN) were combined. A grey feed forward back propagation neural network (GM-FFBPNN) model based on wavelet denoising was proposed, the prediction accuracy of different models was calculated, and the prediction results were compared with original data. The results showed that the prediction accuracy of the GM-FFBPNN was higher than that of the individual GM and FFBPNN models. The mean absolute percentage error (MAPE) of the combined model was 7.39%, the root mean square error (RMSE) was 49.01 mm, the scatter index (SI) was 0.06%, and the BIAS was 2.42%. The original monitoring data were applied to the combination model after wavelet denoising, and MAPE and RMSE were only 1.78% and 16.05 mm, respectively. Compared with the combined model before denoising, the prediction error was reduced by 5.61% and 32.96 mm. Thus, the combination model optimized by wavelet analysis had a high prediction accuracy, strong stability, and accorded with the law of change of measured data. The results of this study will contribute to the construction of future surface engineering in goafs and provide a new theoretical basis for similar settlement prediction engineering, which has strong popularization and application value.

Plasma etch is a complex semiconductor manufacturing process in which material is removed from the surface of a silicon wafer using a gas in plasma form. As the process etch rate cannot be measured easily during or after processing, virtual metrology is employed to predict the etch rate instantly using ancillary process variables. Virtual metrology is the prediction of metrology variables using other easily accessible variables and mathematical models. This paper investigates the use of Gaussian process regression as a virtual metrology modelling technique for plasma etch data.

In this paper, the effectiveness of the genetic algorithm-based back propagation (GABP) neural network model and its application to the breakout prediction in the continuous casting process are investigated. The formation of the sticking-type breakouts and the prediction principle of thermocouple thermometry method are analysed firstly. Then the genetic algorithm-based back propagation neural network model is proposed by fusing genetic algorithm (GA), and error back propagation neural network to offset the demerits of one paradigm by the merits of another. Finally, the GABP neural network model is applied to the breakout prediction in the continuous casting process; and the feasibility of the model is verified by the testing result with the accuracy rate of 97.56% and the prediction rate of 100%.

Virtual metrology (VM) is attracting much interest from semiconductor manufacturers because of its potential advantages for quality control. Plasma etching equipment with state-of-the-art plasma sensors are attractive for implementing VM. However, the plasma sensors requiring physical understanding make it difficult to select input parameters for VM. In addition, those sensors with high sensitivity frequently cause several issues in terms of VM performance. This paper will address plasma sensor issues in implementing a robust VM, where self-excited electron resonance spectroscopy, optical emission spectroscopy, and VI-probe are utilized for critical dimension prediction in a plasma etching process. An optimum sensor selection technique which can give insight into effectiveness of plasma sensors is introduced. In this technique, a numerical criterion, integrated squared response, is proposed for effective selection of important sensors for particular manipulated variables. Sensor data shift across equipment preventive maintenance (PM) and its impact on VM performance are also addressed, where a recursive data centering technique is introduced to handle PM-to-PM sensor data drift in a cost-effective way. The application of the technique introduced in this paper is shown to be effective in dynamic random access memory manufacturing. Hopefully, these results will encourage further implementation of robust virtual metrology in plasma etching for semiconductor manufacturing.

Presence of high degree of orthotropy enhances shear lag phenomenon in laminated composite box-beams and it persists till failure. In this paper three key parameters governing shear lag behavior of laminated composite box beams are identified and defined by simple expressions. Uniqueness of the identified key parameters is proved with the help of finite element method (FEM) based studies. In addition to this, for the sake of generalization of prediction of shear lag effect in symmetrical laminated composite box beams a feed forward back propagation neural network (BPNN) model is developed. The network is trained and tested using the data base generated by extensive FEM studies carried out for various b/D, b/tF, tF/tW and laminate configurations. An optimum network architecture has been established which can effectively learn the pattern. Computational efficiency of the developed ANN makes it suitable for use in optimum design of laminated composite box-beams.

Plasma etch is a key process in modern semiconductor manufacturing facilities as it offers process simplification and yet greater dimensional tolerances compared to wet chemical etch technology. The main challenge of operating plasma etchers is to maintain a consistent etch rate spatially and temporally for a given wafer and for successive wafers processed in the same etch tool. Etch rate measurements require expensive metrology steps and therefore in general only limited sampling is performed. Furthermore, the results of measurements are not accessible in real-time, limiting the options for run-to-run control. This paper investigates a Virtual Metrology (VM) enabled Dynamic Sampling (DS) methodology as an alternative paradigm for balancing the need to reduce costly metrology with the need to measure more frequently and in a timely fashion to enable wafer-to-wafer control. Using a Gaussian Process Regression (GPR) VM model for etch rate estimation of a plasma etch process, the proposed dynamic sampling methodology is demonstrated and evaluated for a number of different predictive dynamic sampling rules.

In order to realize the modernization of power grid management, mid-term power load forecasting has become an important research in power dispatching. Accurate mid-term load forecasting is of great significance to future power system dispatching and safe operation, and is a prerequisite for safe and economical operation of power systems. Based on error back propagation (BP) neural network has the advantages of generalization ability, self-learning ability and self-adapting ability, and simple local calculation, this paper proposes to use the BP neural network model to forecast the power load in the alpine region. Firstly, the historical load data and meteorological data of a certain area in Harbin are studied, analyzed and screened, and they are used as the input data of the network, a mid-term load forecasting model based on BP neural network is established, and then the model is simulated and improved through simulation to predict changes in the future development of electric load demand. The prediction data results and images show that the medium-term load prediction accuracy and training speed based on the BP neural network can reach the target, and the performance is good.

Development of a novel cloud-based multi-tenant model creation service for automatic virtual metrology

Background: E-beam metrologies, both critical dimension scanning electron microscope metrology and defect scan metrology, have been playing a very critical role in gating patterning quality. SEM images can provide rich visual information for engineers to do qualitative and quantitative analyses. However, the low e-beam metrology tool throughput makes it impossible to obtain SEM images for larger area. Monte Carlo-based SEM image simulations or other SEM image simulations require postlithography or postetch pattern three-dimensional structures as prerequisite, and the simulation speed is not sufficiently fast for full chip implementation. Aim: We aim to develop machine learning SEM models with sufficient accuracy and speed for full chip application in semiconductor manufacturing environment. Approach: We have proposed a virtual SEM metrology solution based on U-Net neural network with physics-based feature maps as model input. With information in aerial image space encoded properly, SEM images of both postlithography and postetch can be predicted accurately enough for practical applications using our proposed virtual SEM metrology models. Equipped with GPUs, the machine learning-based SEM image models are fast enough to make it possible to realize full chip SEM generation from post-OPC data. Results: Our machine learning SEM image models can predict SEM images with normalized cross correlation around 0.95 in reference to ground truth SEM images, each SEM image (512 × 512 in size) takes about 800 ms using single CPU, and the speed can be accelerated to about 10 ms with single GPU. Conclusions: Using U-Net structure and physics-based feature maps as model inputs, machine learning-based SEM image models can be developed. The models are sufficiently accurate and fast to find their applications in semiconductor manufacturing, and they can be used as independent model for OPC data verification or generate SEM images as reference for SEM defect scan metrology.

Virtual metrology (VM) is the prediction of end-of-batch properties (i.e., metrology data) using process variables and other information available for the process and/or the product (i.e., machine data) without physically conducting property measurement. VM (sometimes augmented with existing metrology) has been utilized in semiconductor process monitoring and control. Besides the economic benefit of replacing or reducing metrology tools, due to the instant availability of high frequency machine data, a good VM can actually provide better process monitoring and control performance compared to the same monitoring and control schemes based on the physical metrology data which often obtained at lower frequencies and usually with delays. In this paper, we propose a statistics pattern analysis (SPA) based VM approach for predicting sheet resistance using optical emission spectroscopy (OES) data. The advantageous properties of the SPA based VM are discussed. And the performance of the SPA based VM is compared with several commonly used VM algorithms in terms of prediction accuracy.

Virtual metrology (VM) predicts on-wafer properties, such as thickness and uniformity, using equipment data (e.g., sensors, constants) and potentially on-wafer properties or predictions from previous steps. While much literature is devoted to developing VM models, maintaining them (e.g., so they can be used for months in production) presents a different set of challenges. Behavior of equipment can change over time, causing degradation of VM models. Maintenance events can abruptly reset sensors and process conditions, invalidating VM models. In addition, sensor noise, sensor imperfection and aging parts can disturb performance of VM models. While some of these issues can be addressed during modeling by incorporating certain variables that capture the dynamics associated with these issues, others are harder to foresee. Resolving these issues may require tuning the models as they age or manually rebuilding the models after they become ineffective. To mitigate these problems, techniques including monitoring residual error and tool degradation, compensating for sensor reset and residual error, and auto-retraining models can significantly help VM models maintain their performance during deployment.

This paper proposes to develop a data-driven via's depth estimator of the deep reactive ion etching process based on statistical identification of key variables. Several feature extraction algorithms are presented to reduce the high-dimensional data and effectively undertake the subsequent virtual metrology (VM) model building process. With the available on-line VM model, the model-based controller is hence readily applicable to improve the quality of a via's depth. Real operational data taken from a industrial manufacturing process are used to verify the effectiveness of the proposed method. The results demonstrate that the proposed method can decrease the MSE from 2.2 × 10−2 to 9 × 10−4 and has great potential in improving the existing DRIE process.

Statistical Approach to Neural Network Model Building for Gentamicin Peak Predictions

To resolve the problem of short-term power load forecasting, we propose a self-adapting particle swarm optimization (PSO) algorithm to optimize the error back propagation (BP) neural network model. The proposed model is called PSO-BP model which employs PSO to adjust control parameters of BP neural network. In order to verify the performance of PSO-BP, the practical datum of a city in China are selected in the experiments. Simulation results show that our approach outperforms simple BP neural network.

The River Ganga (2,525 km long) is the largest River basin in India, covering 26.2 percent of India's total geographical area and recently granted living entity status by the court. It is the holiest River and also among the dirtiest in the world. That’s why it is mandatory to maintain its water quality (WQ). Though, monitoring and assessment of WQ of a River is a very challenging task. In this research work, Soft Computing (SC) based popular and commononly used Artificial Neural Network (ANN) technique has been used for modelling the WQ of the Ganga River by developing a prediction model based on six different training functions. Five sampling stations along this River stretch were selected from DEVPRAYAG to ROORKEE in the Uttarakhand state of India. The monthly data sets of five water quality parameters temperature, pH, dissolved oxygen (DO), biochemical oxygen demand (BOD) and total coliform (TC) for the time period from 2001 to 2015 have been taken. The feed forward error back propagation neural network method has been used to develop the WQ-prediction model by conducting various experiments following a neural network structure of 5-10-1, 0.1 as a training goal and various training functions. Using the Mean square error (MSE) statistical method the prediction performance of the developed model was evaluated. The model developed with traincgp (Conjugate Gradient with Polak-Ribiere Restarts) comes out to be the worst one (MSE=0.786) while the other model with trainlm (Levenberg-Marquardt backpropagation) rule proved to be the best one (MSE=0.163) among others. Consequently, it is found that ANNs are capable of predicting WQ of the River Ganga with acceptable results.