CD Error Research Articles

Regional dynamic point cloud completion network

Point cloud completion network often encodes points into a global feature vector, then predicts the complete point cloud through the vector generation process. However, this method may not accurately capture complex shapes, as global feature vectors struggle to recover their detailed structure. In this paper, we present a novel shape completion network, namely RD-Net, that innovatively focuses on the interaction of information between points to provide both local and global information for generating fine-grained complete shape. Specifically, we propose a stored iteration-based method for point cloud sampling that quickly captures representative points within the point cloud. Subsequently, in order to better predict the shape and structure of the missing part, we design an iterative edge-convolution module. It uses a CNN-like hierarchy for feature extraction and learning context information. Moreover, we design a two-stage reconstruction process for latent vector decoding. We first employ a feature-points-based multi-scale generating decoder to estimate the missing point cloud hierarchically. This is followed by a self-attention mechanism that refines the generated shape and effectively generates structural details. By combining these innovations, RD-Net achieves a 2% reduction in CD error compared to the state-of-the-art method on the ShapeNet-part dataset.

Integrated Test Pattern Extraction and Generation for Accurate Lithography Modeling

A set of good test patterns is a key for accurate lithography modeling. We address three sub-problems in test pattern extraction: (1) Each pattern is associated with one or more gauges, where CD is measured and modeling is performed. We perform gauge clustering in feature space, while the goal is to minimize the number of chosen patterns. (2) Sample patterns do not usually guarantee enough coverage, so once coverage of chosen patterns is identified, some synthetic patterns are systematically generated to fill the empty region in feature space. (3) Pattern representation defines the feature space. We propose to use some Gaussian convolutions (or a few of their principal components) together with image parameter set (IPS) to reflect exposure for optical model and both PEB and development for resist model. Experiments with 1xnm DRAM design demonstrate that lithography modeling through our test pattern extraction and generation is performed with 35% smaller number of test patterns while CD error is reduced by 55%, with pattern extraction in feature space of IPS parameters as a reference.

Towards point cloud completion: Point Rank Sampling and Cross-Cascade Graph CNN

The Point Fractal Network (PF-Net) is a seminal work with capability of completing the missing regions of point clouds. However, the multi-resolution structure of PF-Net neglects effective feature fusion between each resolution, which makes the resulting shape lack local geometric details. To tackle this problem, we design a novel shape completion network named PRSCN. We first present Point Rank Sampling to rate and sample feature points more objectively through local outline form. In this way, the sampled points can facilitate the downstream tasks. Subsequently, considering the correlation between features from different scales, we design a Cross-Cascade Module to combine features hierarchically. Moreover, we propose Leap-type EdgeConv to enlarge the receptive field while maintain the kernel size unchanged. These improvements together make our CD error 3% lower than that of state-of-the-art method on ShapeNet-part dataset.

Stochastic side-lobe printing in EUV lithography: a simulation study

Background: Extreme ultraviolet (EUV) phase shifting masks (PSMs) may be introduced at the N3 node as alternative to double patterning or high-NA exposures. To be optically advantageous, PSMs may have to come with phase different from 180 deg and high transmission. Aim: The main concern is how such exposures could be affected by stochastic side-lobe printing. Approach: We present a simulation study where we compare stochastic distributions for contact layers with different mask approaches: binary mask, PSM with and without assist features, and TriTone masks. The purpose of this work is to evaluate whether PSMs in EUV lithography may become a valuable option to further extend single-exposure EUV at NA = 0.33. Results: We identify the best exposure conditions for lowest stochastic defectivity for each of the masks considered and perform a large-scale study at these conditions. We report CD and placement error distribution parameters, along with frequencies of observed defects. Conclusions: We find that PSMs are more limited in range of CDs that can be printed without stochastic defects, compared with binary masks. When compared at target CD that is optimal for PSM, proposed phase shift masks did not exhibit a clear advantage, suggesting careful consideration of stochastic defects will be needed during future process development.

Modeling high-efficiency extreme ultraviolet etched multilayer phase-shift masks

Achieving high-throughput extreme ultraviolet (EUV) patterning remains a major challenge due to low source power; phase-shift masks can help solve this challenge for dense features near the resolution limit by creating brighter images than traditional absorber masks when illuminated with the same source power. We explore applications of etched multilayer phase-shift masks for EUV lithography, both in the current-generation 0.33 NA and next-generation 0.55 NA systems. We derive analytic formulas for the thin-mask throughput gains, which are 2.42× for lines and spaces and 5.86× for contacts compared with an absorber mask with dipole and quadrupole illumination, respectively. Using rigorous finite-difference time-domain simulations, we quantify variations in these gains by pitch and orientation, finding 87% to 113% of the thin-mask value for lines and spaces and a 91% to 99% for contacts. We introduce an edge placement error metric, which accounts for CD errors, relative feature motion, and telecentricity errors, and use this metric both to optimize mask designs for individual features and to explore which features can be printed on the same mask. Furthermore, we find that although partial coherence shrinks the process window, at an achievable sigma of 0.2 we obtain a depth of focus of 340 nm and an exposure latitude of 39.2%, suggesting that partial coherence will not limit the feasibility of this technology. Finally, we show that many problems such as sensitivity to etch uniformity can be greatly mitigated using a central obscuration in the imaging pupil.

Effect of Particle Contamination on Extreme Ultraviolet (EUV) Mask and Megasonic Cleaning Process for Its Removal

Particle contaminants on EUV mask surface can cause CD error during EUV lithography process. These contaminants should be removed from EUV mask. The critical size of defect which can cause 10 % CD error is about 23~38 nm with various materials. Among them, critical defect size of SiO2 is about 30 nm. To remove this 30 nm fine silica particle, megasonic cleaning was adapted with various megasonic frequencies and with various cleaning solutions such as dilute alkali solution and gas dissolved water. At high megasonic frequency (2 MHz), though bubble cavitation effect is decreased, cavitation threshold value is higher even with the addition of gases into cleaning solution. As a result, higher particle removal efficiency of 30 nm silica particles and lower damages on PR pattern is achieved at 2 MHz megasonic cleaning in a cleaning solution of H2 gas and dilute NH4OH as compared with 1 MHz megasonic cleaning.

Effect of Particle Contamination on Extreme Ultraviolet (EUV) Mask and Megasonic Cleaning Process for Its Removal

Extreme ultraviolet lithography (EUVL) is the most promising lithography technique for the 22 nm half-pitch and beyond. The EUV patterned masks work in a reflective mode and it has 40 to 50 pairs of MoSi multilayer to reflect EUV radiation. This multilayer is capped by a 2.5 nm Ru thin film to protect the MoSi multilayer. Since EUV mask does not have a pellicle which protects mask surface from the contaminants. Therefore, EUV mask is more vulnerable to particle contamination during lithography processing which is a major yield loss in device fabrication. No study has been reported on the effect of particle size on CD variation if they exist on EUV mask. The critical defect size, which can cause 10 % CD error, has to be removed from the mask surface. When the defect is located between line and space patterns on the mask, the reflectivity will be reduced due to defect absorption. Simulation predicted that 30 nm sized silica particles can cause 10 % CD error on 16 nm line and space pattern. Thus, these contaminant particles have to be removed from EUV mask surface for preventing CD variation. In order to evaluate the 30 nm sized silica particle removal test, we adopted 30 nm standard spherical silica particles. Actually, it is a challenge to evaluate 30 nm particles removal test without inspection tools. So, we used atomic micro scope (AFM) for the evaluation of these particles. The silica particles were intentionally deposited on Ru surface using spinning method. Cleaning test was performed with twin type megasnoic. 30 nm silica particles removal efficiency was drastically decreased as compared with 137 nm silica particles at 1 MHz megasonic cleaning. Because the megasonic cleaning force was reduced to ultra-fine particles due to boundary layer thickness. Therefore, higher megasonic cleaning force is required to remove ultra-fine particles to overcome boundary layer thickness. It is well known that the large cleaning force is induced, not only particle removal efficiency but also pattern damage will be increased. Thus, megasonic cleaning process without causing pattern damage has to be optimized.In this study, we shall discuss the effect of megasonic frequency on the removal of ultra-fine particles and pattern damage. Also, gas dissolved water and dilute alkali cleaning solution are adopted to high frequency megasonic cleaning process to achieve ultra-fine particles removal without pattern damage effectively. Figure 1

Process Optimization Using OPE-Master and CDU Master

OPE matching and CD-Uniformity control have become inevitable steps for most of high volume manufacturing(HVM) of a state-of-the-art devices. OPE matching is a step to match CDs of various patterns from different scanners or steppers. CD-Uniformity control is to stabilize CDs across exposure fields, across wafers and across lots, for HVMs. At a current OPE matching, one usually collects tool properties from each matching tool and import them into simulation software they are using to improve its prediction accuracy. However, generally speaking, this is a time and human resource consuming process, because data acquisition from scanners usually take a lot of time of both human and machine. Furthermore, what makes it more difficult is that an OPE matching, in most cases, requires several iterations. In this paper, we will introduce our latest software named OPE-Master. By employing the software, we can expect following merits. Firstly, with the help of the software, one can be able to automatically import machine properties. This means we can expedite the iterations without sacrifice of the accuracy. Secondly, one can also be able to get better solutions, because it uses all possible adjustment knobs, some of which are ones that unable to be accessed by a third party softwares, with the help of newly developed optimization algorithm that uses a hybrid method of nonlinear optimization and rigorous simulation. Finally, intuitive interface of the software makes it possible for an engineer without much optical simulation experiences to do an OPE matching simulation. In this paper, we will also introduce one of our latest software, called CDU-Master, which will be used to control CD uniformities for a HVM. CDU-Master can analys CDs from various metrology tools, such as, CD-SEMs, OCDs or a macro inspection tools(AMI, for example). CDU-Master can decompose CD errors into focus and dose errors and then derives parameters to correct CDs. One can expect extremely high accuracy of the decomposition, because, CD-SEM as an example, it uses all possible information such as CDs and pattern 3D structure information. The correction parameters, which may include very high order coefficients to be able to deal with very high spatial frequency distribution of errors across a field or across a wafer, will be converted into a format that can be recognized by scanners and then can be exported to scanners automatically, by the software.

Limitation of Layout Correction on Improvement of OPC Model Calibration

OPC models are fundamental for accurate OPC. With critical dimension shrinking from generation to generation, the accuracy requirement imposed on OPC model is getting more stringent. From a regression point of view, in order to acheive a good model, uncertainty in calibration data should be minimized throuhg averaging, while errors in control variables should be corrected as much as possible. In OPC model calibration, the deviation of mask CD from design CD constitutes the errors in control variables. It is perceived that OPC model quality might be improved by eliminating this error source in OPC model calibration. Based on this line of reasoning, we have carried out experiments, setting up two OPC models; one model with desgin GDS as control variables, and the other with measured mask CD as control variables. The model accuracies from these two OPC models are compared. It is found that the model accuracies form these two models are statistically insignificant. This unexpected result was analyzed, and the causes are mainly ascribed to (1) small mask CD errors, (2) nearly symmetrical mask CD error distribution around zero and (3) uncertainties in measurements, (4) models form's capability to absorb small mask CD errors.

CD errors lead to pharmacist’s three-month suspension

CD errors lead to pharmacist’s three-month suspension

OPC Accuracy and Development Time Improvement for Advanced Technology Nodes

Double Dipole (DDL), Double Patterning (DP), Spacers and other advanced process technologies require an enormous amount of CD and tool data collection and development time for OPC modeling. In this paper we studied a way to reduce the overall OPC preparation time and increase the OPC accuracy by early and accurate collection of the OPC measurements, automatic CD-SEM recipe generation and CD data analysis and model calibration with Brion's Tachyon software. We reviewed the design of the OPC test chip, OPC exposures, data analysis and OPC modeling on the advanced and accurate Tachyon T2.0 OPC platform and ways to improve modeling cycle time. Another critical parameter of the OPC modeling is accuracy as one of the main parts of the CD error budget. We investigated approaches to OPC modeling accuracy and achieved RMS below 1nm for critical features by using module data of the respective ASML TWINSCAN XT:1900i scanner.

High-accuracy correction of critical dimension errors appearing in large-scale integration fabrication processes

In our previous paper, we proposed a new method to obtain accurate pattern dimensions for correcting global critical dimension CD errors, which are defined as errors of CD uniformity in a region of several millimeters to several centimeters. The method is based on the pattern modulation method a method of correcting CD errors by controlling figure sizes of large-scale integration LSI patterns. An essential point of our method is to take into account the difference between the pattern density of the original LSI pattern and that of the corrected pattern which has a pattern dimension different from original one after pattern modulation for correction to provide an accurate correction. In this paper, we apply the proposed method to correct CD errors caused by flare and microloading effects. It is shown from numerical calculations that our method can suppress the correction error to less than 0.1 nm for both cases by three iterations. It is strongly recommended that our method be used for a wide range of applications to provide the necessary CD accuracies of the future.

Evaluation of producer’s and consumer’s risks in scatterometry and scanning electron microscopy metrology for inline critical dimension metrology

Measurement characteristics in scatterometry and critical dimension-scanning electron microscopy (CD-SEM) for lot acceptance sampling of inline CD metrology were investigated by using a statistical approach with Monte Carlo simulation. By operation characteristics curve analysis, producer's risk and consumer's risk arising from sampling were clarified. Single use of scatterometry involves a higher risk, such risk being particularly acute in the case of large intrachip CD variation because it is unable to sufficiently monitor intrachip CD variation including local CD error. Substituting scatterometry for conventional SEM metrology is accompanied with risks, resulting in the increase of unnecessary cost. The combined use of scatterometry and SEM metrology in which the sampling plan for SEM is controlled by scatterometry is a promising metrology from the viewpoint of the suppression of risks and cost. This is due to the effect that CD errors existing in the distribution tails are efficiently caught.

Possible Origins and Some Methods to Minimize LER

Line-edge roughness (LER) & line-width roughness (LWR) poses serious problems for the device performance and CD error budget for sub 100nm design rules. Illumination conditions, pitch, resist formulations, resist process conditions, substrates have shown to be the possible origins of LER. With in the resist formulations, polymer composition, dissolution characteristics of polymer during the development process, acid diffusion influence the roughness of the resulting patterns. ITRS (2003) roadmap requires LWR of 2nm (3 sigma) for 65nm DRAM half-pitch for 2007 year of production. Resist formulation optimization can minimize LER but this can not be achievable simply by the resist formulation modifications as the radius of gyration of polymers itself is in the order of 2-5 nm. It is therefore important to study (1) device performance compromises if the required 2nm is not achieved (2) methods to improve the LER after patterning (litho). As for methods to improve the LER after patterning viability of use of rinse solutions, hard bake, RELACS process and some etch process are reported. Possible origins of LER and some attempts to minimize the LER after patterning such as hard bake and RELACSTM process are discussed.

Critical dimension control in optical lithography

The exposure tool used for integrated circuit (IC) fabrication is critical to improving the packing density and transistor speed of the circuits. In addition to increasing resolution, which improves packing density and transistor speed, the exposure tool is also expected to provide tight linewidth control across the chip. Across chip linewidth variation (ACLV) has a significant influence on circuit speed. The allowed ACLV is usually assumed to be about 10% of the nominal linewidth. Therefore, just a few nanometers in linewidth variation may significantly impact IC performance. Contributions to the CD variation across chip and wafer due to lithographic sources of error are discussed in this paper. CD control afforded by future optical lithography tools is estimated using Monte Carlo aerial image simulations by making reasonable assumptions about the performance of the future tools and mask CD control. The impact of reticle enhancement technologies on ACLV is evaluated. The main sources of CD error can be identified. This approach will help define the path to improving CD control. The technique described was tested using data from the current generation of technology, and reasonable agreement between predicted and observed CD variation was obtained.

Simultaneous dose modification for balanced E-beam proximity correction minimizing CD error

Many proximity effect correction schemes employ certain iterative procedures and often correct circuit features sequentially within each iteration for faster convergence. Such sequential correction procedures, whether dose or shape modification, tend to lead to unbalanced results due to the recursive effect especially for high-density fine-feature circuit patterns. In this paper, a dose modification scheme which performs correction non-recursively within each iteration is proposed in order to obtain spatially well-balanced correction results, minimizing CD error. A common drawback of such a non-recursive or simultaneous scheme is its long computation time required for convergence of the solution (correction result). In the proposed scheme, a scaling factor is used between successive iterations to avoid oscillation of the solution. Also, it partitions each rectangular feature into regions judiciously such that the total number of regions, on which correction time mainly depends, is not too large. Performance of the scheme has been analyzed through an extensive computer simulation.

High-throughput high-density mapping and spectrum analysis of transistor gate length variations in SRAM circuits

High-throughput high-density mapping of gate length variations using static random-access memory (SRAM) as electronic test structures is reported. In the experiments, direct measurements of bit-line currents revealed the individual transistor gate length variations within every memory cell. With SRAM and its fast addressing circuits we can measure CD variations with measurement time as fast as 5 /spl mu/s per data point and spatial periodicities down to 6 /spl mu/m. Layout-dependent periodic errors were found to take up 30% to 90% of the total observed error variance, depending on the spatial frequency range and specific measurement grid used. Peaks in the error spectrum were found to be related to the periodicities existing in the circuit layout. Lithography simulations were done as efforts to identify the periodic error sources. It was found that proximity effects and pattern-dependent coma effects contributed to a large percentage of the high spatial frequency errors observed. By independent optical measurements of the poly mask, it was found that the CD error contributions from the mask are very small, and are negligible when compared to the stepper-lens-induced CD errors.

Analysis of a SCALPEL™ multi-pass writing strategy

In SCALPEL™, the microcircuit pattern information is written on the wafer in a series of stripes, typically 2.75 mm wide with 0.24 mm pitch. To distribute the heat in the wafer as evenly as possible, each stripe may be written with multiple passes of the sub-field illumination along each stripe, and a blending scheme will be used to improve the dose uniformity and reduce CD errors near the boundaries at the ends and sides of each stripe. Our software has been extended to simulate the performance of SCALPEL™ including multi-pass writing strategies and blending regions. The method is described and illustrative results are presented. The results show that there is no significant loss of performance due to the multi-pass and blending schemes.

SCALPEL mask defect imaging analysis

Parametric image-forming models were used to demonstrate the imaging of potential mask defects over ranges of descriptive parameters, either isolated or coincident with a critical object. Beyond basic defect printability effects, CD errors due to defects near CD objects cause serious constraint on allowable pattern defects or contaminants, and define post-repair requirements. The models can provide scaling laws based on defect type, size, contrast, and proximity to a CD object, as well as rules describing tolerable defects using a threshold value for the local CD anomaly they cause.

Practical approach to separating the pattern generator-induced mask CD errors from the blank/process-induced mask CD errors using conventional market measurements

With the continuing reduction of minimum feature sizes for semiconductor wafers, the manufacturing tolerances for photomask CD errors were reduced proportionately. As a result, it has become increasingly important to separate the major sources of mask CD errors so that they can be quantified and appropriately addressed. In this article we describe a reliable, convenient and inexpensive technique for separating the blank, and process CD errors from the pattern generator CD errors using conventional market measurements on cross arrays and two-dimensional spatial frequency Fourier analyses. By taking advantage of the fact that the blank/process-induced CD error spatial distributions are known to be fixed relative to the mask blank, while the pattern generator CD error distributions are known to be fixed relative to the pattern generator, a spatial frequency domain analysis of multiple sets of CD measurements on the masks, each set printed by displacing the mask relative to the previous set, is shown to allow a simple and unambiguous separation of the two types of CD errors. These measurements also directly verify the expected result that the blank/process CD error contributors tend to have mainly low-spatial-frequency components, while the pattern generator tends to mainly have higher-spatial-frequency components. Furthermore, by utilizing this result, it is shown that a simple low-pass filter applied to a single set of market measurements provides a surprisingly accurate estimate of the relative sizes of these two error contributors. Results are presented from masks printed on two commercial mask-making systems, one optical and one e-beam.