Over the past few years, noise2noise, noise2void, noise2self, and unsupervised deep-learning (DL) denoising techniques have achieved great success, particularly in scenarios where ground truth data is not available or is difficult to obtain. For semiconductor SEM images, ground truth or clean target images with lower noise levels can be obtained by averaging hundreds of frames at the same wafer location, but it is expensive and can result in physical damage to the wafer. This paper’s scope is to denoise SEM images without clean target images and with limited image counts. Inspired by noise2noise, we proposed an additive noise algorithm and DL U-net. We achieved good denoising performance using a limited number of noisy SEM images, without the clean ground truth images. We proposed the “denoise2next” and “denoise2best”. We compared generative adversarial network(GAN) generated images and Additive noise images for data augmentation. This paper further quantified the impact of image noise level, pattern diversity, and continuous (aka transfer) learning. The data sets used in the work include both line/space and logic pattern.
SEM image processing is an important part of semiconductor manufacturing. However, one difficulty of SEM image processing is collecting enough defect-containing samples of defect-of-interests (DOI) because many DOIs are very rare. This problem becomes more prominent for Machine Learning (ML) or Deep Learning (DL) based image processing techniques since they require large amount of samples for training. In this paper, we present a Generative Adversarial Networks (GAN) based defect simulation framework to tackle this problem. The fundamental insight of our approach is that we treat the defect simulation problem as an image style transfer problem. Following this thought, we train a neural network model to turn a defect-free image into a defect- containing image. We evaluate the proposed defect simulation framework by using it as a data augmentation method for ML/DL based Automatic Defect Classification (ADC) and Image Quality Enhancement (IQE) on a Line Pattern Dataset, which is collected with ASML ePTMand eScan R series inspection tools from an ASML standard wafer. The experimental results show a significant performance gain for both ADC and IQE. The result proves our defect simulation framework is effective. We expect GAN based defect simulation can have a broader impact in many other SEM image development and engineering applications in the future.
Continuous reduction in pattern size, the primary path of advancement for the semiconductor industry, has greatly increased resolution and throughput demands for defect inspection and metrology, where Electronic-beam (E-beam) wafer inspection equipment has been commonly used for both purposes. High resolution is specifically needed in order to inspect or measure these smaller patterns and is accomplished by either decreasing pixel size or increasing frame averages. Both of these adjustments come with a big penalty of throughput, which is an extremely important metric as large areas of the wafer must be inspected in a reasonable time to meet semiconductor development, yield ramp and high volume manufacturing process control requirements. A slow inspection means more inspection tools are required and lots are delayed by the longer process time. In order to regain throughput, it is common to try to back off on the frame averages, but this often results in low quality images with noise, blurring effects, and distortions. The end result is less defect sensitivity for inspections, lower CD measurement accuracy and precision for metrology. Image quality enhancement (IQE) algorithms can compensate for this and thereby play a significant role in achieving higher throughput while keeping sufficient sensitivity. In recent years, deep learning methods have demonstrated superior performance to traditional algorithms for IQE. However, these methods often require clean ground truth data for supervised training purposes, which is extremely difficult and expensive to achieve. For example, ground truth images with lower noise levels can be obtained by averaging hundreds of frames at the same location, but, in addition to taking a very long time, can cause permanent physical damage to the wafer due to the E-beam wafer imaging process, and unexpected artifacts or shadowing effects. In order to alleviate these issues, we propose an unsupervised machine learning- based image quality enhancement framework (uMLIQE) using deep learning methods, which does not require clean target images for the training process. In fact, only one or a few images are required since the required information can be extracted by segmenting the available image. The performance of this system was compared both via simulation and experimentally to a comprehensive list of alternate IQE approaches. The wafer we used for data collection was generated with standard semiconductor processing representative of CMOS processing across the industry. The unsupervised approach is clearly superior to all alternatives both qualitatively and quantitatively. Our proposed unsupervised deep learning IQE framework for SEM images has proven superior for throughput enhancement for high resolution E-beam wafer imaging.
In semiconductor industry, as physical sizes of integrated circuit (IC) components continue to shrink, critical dimension (CD) metrology plays an important role in manufacturing process monitor and control. However, when prior knowledge of E-beam tool conditions and statistics of underlying imaging samples are limited or missing, metrology parameters (such as imaging conditions and CD measurement parameters) are often selected empirically and not optimized in terms of measurement accuracy or precision. Common practice involved in fine-tuning some of the parameters may result in a time-consuming trial-and-error cycle.
In this paper, we propose a guidance system to provide an optimized set of metrology parameters given a line/space pattern image or images of scanning electron microscope (SEM). The proposed system models input condition with a comprehensive set of model parameters and then statistical analysis is done based on modeling outputs. A set of metrology guidelines, such as measurement parameters and achievable precisions, can be recommended by the proposed system. The validity of our method has been demonstrated by comparing the recommended parameters with the optimal parameters found by brute-force search on a set of 100 SEM images of line/space patterns.
As device size continues to shrink, stochastic-induced roughness of resist features exposed by photolithography is of increasing concern to the semiconductor industry. In this paper, we propose an end-to-end approach for line roughness analysis by using the Line Roughness Module from our CDU solution family, which is a part of HMI’s metrology SEM tool the eP5. Simulated Scanning Electron Microscope (SEM) images of line/space patterns are used to verify the ability of the Module to reliably extract roughness related metrics. A set of imec EUV ADI images collected on our metrology SEM tool are analyzed by the Line Roughness Module, and wafer signature maps of various roughness metrics are obtained. These wafer maps not only help to analyze different roughness contribution sources, but also provide insights about feature roughness in a systematic way. Such information can be further used in a feedback loop to the scanner for model correction and process control.
A novel rate-optimal rate allocation algorithm is proposed for parallel transmission of scalable images in multi-channel
systems. Scalable images are transmitted via fixed-length packets. The proposed algorithm selects a
subchannel as well as a channel code rate for each packet, based on the signal-to-noise ratios (SNR) of the
subchannels. The resulting scheme provides unequal error protection of source bits. Applications to JPEG2000
transmission show that significant UEP gains are achieved over equal error protection (EEP) schemes.
In this paper, a robust image transmission scheme based on JPEG2000 is proposed for packet erasure channels. Error resilience functionalities provided by JPEG2000 are utilized to control the source coding efficiency and the robustness according to channel conditions. Furthermore, together with the proposed interleaving scheme, some erasures can be recovered. Experimental results show the effectiveness of the scheme.