Galileo Galilei once quoted: “Measure what is measurable, and make measurable what is not so”. In silicon manufacturing R&D phase, it often happens that engineers would like to access some parameter values that are not easy, even impossible to measure. When looking at a CDSEM image, the parameters of interest seem easy to extract but in practice getting access to them in a robust and reliable way is not always simple. Developing a contour-based metrology tool coupling robust contour extraction with a comprehensive contour metrology environment could help to bridge this gap. In previous works, it has been shown that SEM images contain significant amounts of information that can be extracted and analyzed using efficient contour extraction and analysis toolboxes [1, 2]. Also, the concept of implementing remote contour-based metrology has been introduced. The present work continues to unveil what can be achieved with such solutions. For that, the example of implant layers’ process assumption will be explored. During this process step, counter doping problems can occur for example when the distance between layers deviates from nominal. Therefore, it is crucial for design rule control to measure some critical dimensions such as minimum distance between layers, corner rounding, slope, etc. However, given the characteristics of the different structures in the images, which may come from different layers and/or processes steps, the measurements are not straightforward to extract with standard CDSEM metrology algorithms. Moreover, recipes are complex to setup, measurements by themselves are not very stable, and usually an indirect determination of the key figure is performed. In this paper, we will show that multilayer contour-based metrology, mixing image contour and GDS layout, allows to overcome the previously mentioned difficulties, as well as to generate measurements that are not possible to be performed by using standard algorithms.

/pdf/contour-based-metrology-make-measurable-what-is-not-so-1pbdc2ghxd.pdf

Contour based metrology: “make measurable what is not so"

(1993). Evaluation and Control of Measurements. Technometrics: Vol. 35, No. 2, pp. 223-223.

Evaluation and Control of Measurements

Contour detection of an object is a fundamental computer vision problem in image processing domain. The goal is to find a concrete boundary for pixel ownership between an OOI (object-of-interest) and its corresponding background. However, contour extraction from low SN SEM images is a very challenging problem as different sources of noise shadow the estimation of underlying structural geometries. As device scaling continues to 3nm node and below, the extraction of accurate CD contour geometries from SEM images especially ADI (after developed inspection) is of utmost importance for a qualitative lithographic process as well as to verify device characterization in aggressive pitches. In this paper, we have applied a U-Net architecture based unsupervised machine learning approach for de-noising CD-SEM images. Unlike other discriminative deep-learning based de-noising approaches, the proposed method does not require any ground-truth as clean/noiseless images or synthetic noiseless images for training. Simultaneously, we have also attempted to demonstrate how de-noising is helping to improve the contour detection accuracy. We have analyzed and validated our result by using a programmable tool (SEMSuiteTM) for contour extraction. We have de-noised SEM images with categorically different geometrical patterns such as L/S (line-space), T2T (tip-to-tip), pillars with different scan types etc. and extracted the contours in both noisy and de-noised images. The comparative analysis demonstrates that de-noised images have higher confidence contour metric than their noisy twins while keeping the same parameter settings for both data input. When the ML algorithm is applied, the contour extraction results would have higher confidence numbers comparing with the ones only applied the conventional Gaussian or Median blur de-noise method. The final goal of this work is to establish a robust de-noising method to reduce the dependency of SEM image acquisition settings and provide more accurate metrology data for OPC calibration.

Unsupervised machine learning based SEM image denoising for robust contour detection

The method to perform Optical Proximity Correction (OPC) model calibration with contour-based input data from both small field of view (SFoV) and large field of view (LFoV) e-beam inspection is presented. For advanced OPC models - such as Neural Network Assisted Models (NNAM) [1], pattern sampling is a critical topic, where pattern feature vectors utilized in model training, such as image parameter space (IPS) is critical to ensure accurate model prediction [2-5]. In order to improve the design space coverage, thousands of gauges with unique feature vector combinations might be brought into OPC model calibration to improve pattern coverage. The time and cost in conventional Critical Dimension Scanning Electron Microscope (CD-SEM) metrology to measure this large amount of CD gauges is costly. Hence, an OPC modeling solution with contourbased input has been introduced [6]. Built on this methodology, a single inspection image and SEM contour can include a large amount of information along polygon edges in complex logic circuit layouts. Namely, a better feature vector coverage could be expected [7]. Furthermore, much less metrology time is needed to collect the OPC modeling data comparing to conventional CD measurements. It is also shown that by utilizing large field 2D contours, which are difficult to characterize by CD measurements, in model calibration the model prediction of 2D features is improved. Finally, the model error rms of conventional SFoV modeling and LFoV contour modeling between SEM contours and simulation results are compared.

Realizing more accurate OPC models by utilizing SEM contours

Computational lithography applications for OPC/ RET utilize models that represent the lithographic process in simulations. The quality of OPC/ RET wafer results strongly depends on the quality of the model. Hence, achieving model quality and experimental match is the goal of the model calibration process where models are calibrated to experimental data. Ideally, the model would be calibrated and validated to a data set that completely covers the entire design space and all process conditions. A promising alternative to the traditionally applied SEM-CD-based model calibration is the calibration to pattern contours directly with benefits in design space coverage, reduced metrology effort and data preparation complexity. However, contour calibration also demands a new standard operating procedure for contour specific metrology, pattern design and calibration. Goal of this work is to develop and exercise a full contour-based calibration methodology. Firstly, we discuss preconditions for a successful calibration: good quality contour input data, predictive modeling of optics, mask topography and 3D resist and additional calibrator functionality to include aspects of alignment and pattern specific measurement confidence. Secondly, we assess pattern for their calibration-suitability using a metric for pattern information density. Experiments are performed to show the applicability of the metric and the potential to calibrate to a minimal set of patterns. A model calibrated to a well selected single 2.25 μm2 contour is predicting a large set of pattern contours, 3D resist characteristics and SEM-CD focus-exposure process windows.

Contour-based model calibration to a minimum number of patterns

The control and the characterization of semiconductor very fine devices on a wafer are commonly performed by mean of a scanning electron microscope (SEM) to derive a critical dimension (CD) from a pair of parallel edges extracted from the images. However, this approach is often not very reliable when dealing with complex 2D patterns. An alternative is to use SEM contour technique to extract all the edges of the image. This method is more versatile and robust but before being implemented in a manufacturing environment, it must demonstrate that it can be matched well with traditional CD-SEM. Aim: The objective of this work is to present a method to evaluate and optimize the CD matching between a reference standard SEM-CD and SEM-Contours. Approach: After describing the metric used to assess the matching performance, we propose to screen some important influent parameters to give an evaluation of the best matching that we achieved with our experimental data. Results: After optimizing the matching calibration parameters and optimizing the selection of the best anchor pattern for the matching we could achieve a 3s-Total Measurement Uncertainty of 0.8 nm and 3.2 nm for 1D and 2D patterns. Conclusions: We established a method to achieve good matching performance that should facilitate the introduction of SEM contour in a manufacturing environment.

/pdf/investigating-sem-contour-to-cd-sem-matching-38oixuebap.pdf

Investigating SEM-contour to CD-SEM matching

2D curvilinear patterns are more and more present in the lithography landscape. For the related devices, the line edge roughness (LER) is, as well as for lines and spaces, a critical figure of merit. In this article we propose to use a dedicated edge detection algorithm to measure LER of 2D curvilinear patterns on CD-SEM images. We present an original method to validate the algorithm, in the context of roughness measurement. It is based on the generation of realistic synthetic CD-SEM images with programmed roughness and a precise PSD analysis flow. We show excellent correlation (average R2 = 0.988) between the input roughness parameters and the measured parameters for both 1D and 2D synthetic images. Using synthetic images for different number of frames, the contour extraction sensitivity to noise is also explored. Finally, the methodology is successfully applied to experimental CD-SEM images for two classes of applications : photonic devices and DSA fingerprint patterns.

/pdf/roughness-measurement-of-2d-curvilinear-patterns-challenges-2qsldsmoly.pdf

Roughness measurement of 2D curvilinear patterns: challenges and advanced methodology

In the effort of continuing improving patterning strategies for increasing circuit density while reducing dimensions, several challenges regarding patterning fidelity emerge. In recent years, stochastic effects had their relative importance increased, and therefore the need for closely monitoring those effects is also increasing [1]. Among other stochastic effects, within-feature roughness is significant as it can impact circuit electrical behavior, decreasing time and power performance, and even lead to failures. The workhorse method of the industry for measuring roughness is based on top-down CD-SEM (Critical Dimension Scanning Electron Microscope) image. In recent years, methods have been proposed as a way to improve and standardize the roughness measurement [2, 3]. Those methods rely on the obtention of the power spectral density (PSD) from the detected edges of the features in SEM images, in order to determine their roughness. However, one important aspect is the impact of the CD-SEM image acquisition conditions on the limitation of the observed PSD. As the acquisition parameters changes, different frequencies may be more or less observable in a SEM image, potentially leading to errors in the metrology evaluation [4].

The goal of this study is to first, present the impact of the CD-SEM image acquisition conditions in the roughness measurement, and, second, propose a method to determine the validity domain of the roughness measurements as a function of the acquisition conditions. 

The proposed method relies on a compact SEM image model. For each acquisition condition, the model is calibrated based on experimental SEM images, from several design samples. Using this calibrated model, synthetic SEM images are generated with a known sample, including its programmed roughness signature (input-PSD), which can be a white noise (defined by a constant PSD). The next step relies on a robust-to-noise edge detection algorithm [5], which is then used to compute the PSD by applying the method proposed in [4]. As the input-PSD is known, it is possible to compute the transfer function of the acquisition system [6], for each of the evaluated acquisition conditions. We call ‘limit-PSD’ the transfer function which may be considered as the signature of the acquisition conditions in the frequency domain. It can be seen as a low-pass filter and it defines the validity domain of the roughness measurements. For each input-PSD (simulated or experimental), if it is ‘below’ the limit-PSD (within the low-pass filter), the measurement is within the validity domain. If the resulting measurement reaches the limit-PSD, it is not possible to know if some roughness information was lost due to the acquisition conditions. Such relationship is illustrated in Figure 1, presenting one case where the input-PSD is inside the validity domain and a second case where the actual roughness is underestimated.

Thanks to the produced results, the information of different acquisition conditions and detected roughness can be obtained and stored in what we call condition tables. For each acquisition condition, the limits of frequency range and roughness parameters range (ξ, H) of the measurable roughness are stored (Figure 2). These condition tables are very useful for assisting metrology specialists in choosing the most suitable acquisition conditions for the CD-SEM, if prior knowledge about the expected roughness is available, or in accounting for the validity domain of the roughness measurement. 

As a final step, the proposed method is applied to experimental data – an example of observed PSD over an experimental SEM image is shown in Figure 3. The experimental dataset is composed of SEM images acquired under different conditions (notably a large variation in landing energy – 300eV to 4000eV), and different sample materials (post-etch and post-litho samples). The obtained results confirmed the applicability of the proposed method in a real environment, and will be fully demonstrated in the final work.

/pdf/determining-the-validity-domain-of-roughness-measurements-as-3kv1w66nrl.pdf

Patrick Schiavone

Papers

Investigating SEM-contour to CD-SEM matching

Roughness measurement of 2D curvilinear patterns: challenges and advanced methodology

Determining the validity domain of roughness measurements as a function of CD-SEM acquisition conditions