Extreme ultraviolet (EUV) lithography is a cornerstone of next-generation computer chip patterning technologies, yet continued improvements in photoresist sensitivity and pattern fidelity are required to enable high-numerical-aperture EUV exposure tools. Numerous studies have shown that underlayers can significantly enhance EUV photoresist performance, including reduced dose-to-size and improved pattern quality. The physical and chemical mechanisms underlying these improvements remain incompletely understood. We investigate one such mechanism: the transport of the photo-acid generator (PAG) in an EUV photoresist–underlayer system. Using time-of-flight secondary ion mass spectrometry depth profiling, we demonstrate that PAG molecules from an EUV chemically amplified resist penetrate significantly into a cross-linked siloxane underlayer. Through a series of experiments, we show that this penetration does not arise from secondary ion mass spectrometry knock-on artifacts, the underlayer already containing PAG-like additives in it or thermally enhanced diffusion. Instead, we find that PAG transport occurs during the spin-coating step at room temperature by the casting solvent. Complementary experiments further show that PAG initially loaded into the underlayer can be leached out by solvent during spin-coating. These results establish solvent-assisted small-molecule transport as a dominant mechanism governing photoresist–underlayer interactions and provide a mechanistic explanation for previously reported dose-to-size reductions for EUV photoresists.
The viability of next-generation extreme ultraviolet (EUV) lithography wavelengths is evaluated through photon–matter interactions across optical materials and photoresists, rather than by resolution scaling alone. Candidate wavelengths of 13.5, 11.3, 6.6, and 3.13 nm are examined within a unified materials-based framework to compare imaging capabilities, enabling assessment of wavelength-dependent tradeoffs. Photomask and optical multilayers impose primary constraints through refractive index contrast, absorption, dispersion, and achievable large-angle Bragg reflectance. Although 13.5 nm Mo/Si systems define the current baseline, 11.3 nm multilayers may provide improved index contrast, reduced absorption, and increased angular bandwidth, supporting more favorable optical performance. At shorter wavelengths, reduced index contrast and subnanometer layer thickness requirements may impose fundamental limits on reflectance, bandwidth, and manufacturability, which may challenge the practical implementation of 6.6 nm and water-window regimes. These limitations propagate to system behavior. Scatter-driven flare increases rapidly with decreasing wavelength, imposing stringent roughness requirements or motivating reduced-surface optical designs. Polarization effects may limit hyper-NA imaging at 13.5 nm, in which large incidence angles produce significant TE–TM imbalance and image contrast loss. Reduced numerical apertures at shorter wavelengths can mitigate polarization sensitivity, with 11.3 nm potentially providing a more favorable balance between angular performance and polarization behavior. Within the photoresist, increasing photon energy with shorter wavelengths produces higher-energy primary electrons and broader secondary-electron cascades, which can increase stochastic blur and offset optical resolution gains. This identifies new resist design opportunities based on core-level threshold matching, in which the exposure wavelength is aligned just above accessible core-level binding energies to constrain electron cascade transport. Wavelength–material pairings such as 11.3 nm with the Si L2,3 edge and 6.6 nm with the S L2,3 edge represent enabling photoresist opportunities to localize energy deposition and potentially reduce stochastic variability through control of electron dynamics rather than absorption alone. These results establish wavelength selection as a challenge of material convergence across multilayer optics, imaging physics, and photoresist response.
Photoresist patterning for semiconductor manufacturing is encountering new challenges with the move to extreme ultraviolet (EUV) lithography, driven by smaller pitches, thinner resists, and fewer photons. Obtaining information about the nanoscale distribution of chemical changes in the resist will provide key information to optimize processing and resist development. Soft X-rays provide sensitivity to chemical functional groups at key atomic absorption edges, including carbon, oxygen, and fluorine. By conducting specular reflectivity measurements at these energies, we can track the chemical depth profiles in a resist. This process is demonstrated by characterizing the optical constant profiles, which are proportional to the concentration of chemical bonds in a film at soft X-ray energies, of an EUV resist coated on a cross-linked polystyrene (CLPS) mat. The resist was exposed to blanket exposures with different doses and measured at different stages of the process, including post-exposure but before the bake, after the post-exposure bake, and after development. Segregation of resist components was observed in the as-coated case films, with a fluorinated component aggregating at the film surface. Stratification of the chemical changes that occur in the film is observed after exposure, but before the bake. The magnitude of this stratification is dose-dependent. After a post-exposure bake, the stratification is no longer observed, and the optical constant profile is uniform with the exception of the fluorinated components. These results are complemented by near-edge X-ray absorption fine structure spectroscopy obtained with an imaging microscope. Finally, we also examine the CLPS mat before and after the process and observe evidence for chemical changes in the CLPS, with the possibility of residual resist components residing in the film after development.
As semiconductor patterning complexity continues to escalate, the early identification of latent weak patterns has become increasingly critical for maintaining yield stability and controlling metrology cost. Conventional identification strategies—primarily based on rule libraries, simulation heuristics, or limited sampling—often lack sufficient coverage and fail to provide statistically reliable assessments of defect vulnerability. The recent availability of production-scale wafer measurement data presents a fundamental opportunity to shift weak-pattern discovery from heuristic screening toward statistically grounded risk modeling. However, transforming such large-volume manufacturing data into actionable insights remains challenging due to feature heterogeneity, data redundancy, and limited interpretability of existing approaches. To address this gap, we introduce a wafer-driven vulnerability quantification framework that enables statistically robust weak-pattern discovery at production scale. Multidomain features derived from design geometry and computational lithography are integrated with over four million real wafer measurements to classify pattern populations and quantify defect risk at the cluster level. Unlike prior clustering-based layout classification studies, the proposed approach performs clustering using design-derived features, whereas wafer measurement data are subsequently integrated at the cluster level to enable statistically grounded risk-based prioritization. Evaluations conducted on production datasets demonstrate that the framework successfully identifies both previously known and undiscovered weak-pattern groups while substantially improving sampling reliability. To the best of our knowledge, this is the first production-scale study that integrates wafer measurement data with unsupervised clustering for statistically grounded weak-pattern ranking. Weak pattern discovery is inherently an unsupervised structural grouping problem as defect-prone configurations are not known a priori and labeled datasets are rarely available in production environments. In addition, a cluster-level visualization methodology enhances engineering interpretability and supports faster decision-making. Collectively, the proposed approach establishes a scalable, manufacturing-aware paradigm for data-driven metrology planning and weak-pattern discovery in advanced semiconductor processes. This design-first and data-augmented strategy ensures generalizability to unseen layouts while leveraging wafer data for statistically reliable vulnerability assessment.
Hyper-NA is the prospective successor to high-numerical aperture (NA) extreme ultraviolet lithography (EUVL). Hyper-NA presents a number of challenges on the mask side, including reduced depth of focus, amplified mask three-dimensional effects, and an increased mask-side angular range. The Sharp Hyper numerical aperture Actinic Reticle Review Project (SHARP) is a synchrotron-based EUV mask imaging microscope. SHARP is supporting mask-side Hyper-NA imaging since 2024. To accelerate research and development toward Hyper-NA EUVL, Fourier Ptychography (FP) has been added to the mask-side Hyper-NA imaging capabilities of SHARP. FP is a computational imaging technique that reconstructs the complex field of an object from a series of intensity images. FP offers pupil fill synthesis and quantitative phase imaging and enables the reproduction of wafer-side through-focus behavior and aspect ratio in SHARP image data. Through-focus image data are computed from FP reconstructions. We demonstrate agreement between NILS curves from FP image data and regular SHARP through-focus image data for a range of pitches, orientations, and pupil fills. The accurate reproduction of the best-focus shift in FP image data is shown for pitch 22.5 nm (1×) vertical lines and spaces. These results highlight the potential of FP as a complementary imaging mode, significantly enhancing the capabilities of the microscope for Hyper-NA EUV mask imaging.
Background The thermal deformation during dose accumulation shifts the image transfer path, leading to significant distortion in the local image profile. This phenomenon underscores the necessity of evaluating thermal deformation not merely as static placement errors but as factors affecting imaging performance. Aim This work aims to clarify how the thermal deformation occurring during EUV exposure directly influences image degradation and placement error. Approach A thermomechanical wafer heating model is established to calculate transient deformation under EUV exposure. The resulting deformation data is subsequently coupled with lithography simulation results to calculate the local image profile. Finally, the image profile is analyzed to quantify the image twisting effect and placement error. Results Placement errors exhibit spatially diverse distributions dependent on global field location and intra-slit coordinates. Furthermore, the dynamic variability of deformation during exposure was found to be a systematic source of image performance reduction, particularly through NILS reduction in fine-pitch patterns. Conclusions These results show that wafer deformation should be treated not only as a source of overlay error but also as a contributor to image degradation. This finding provides a basis for understanding deformation-induced placement error and image twisting in EUV lithography and for improving future overlay and imaging control strategies.
Background Pattern placement accuracy in high-numerical-aperture (NA) extreme ultraviolet (EUV) lithography is constrained by mask-induced image split in the aerial image, resulting from mask 3D effects under off-axis illumination (OAI). This image split stems from pole-dependent phase asymmetry between diffraction orders generated by the thick, reflective EUV mask stack, with increased severity for fine-pitch patterning in high-NA systems. The conventional approach of targeting zero phase delta (Delta phi) between the zeroth and first diffraction orders is suboptimal (could not suppress image split under realistic OAI conditions). Source averaging over finite illumination angles disrupts the symmetry assumed in simplified interference models, rendering Delta phi=0 ineffective for image alignment. Aim This study proposes and validates a design paradigm, termed active phase engineering, which treats Delta phi as an actively optimized parameter. Rigorous coupled-wave analysis for 0.33 and 0.55 NA systems demonstrates that the phase delta minimizing image split is generally nonzero and depends on pattern pitch. Approach By analyzing a broad absorber parameter space, we derive a quantitative design curve that directly relates the optimal Delta phi to the target half-pitch. For a representative 12 nm half-pitch, tuning the mask phase to this predicted target value suppresses image split, reduces the normalized image log-slope penalty, and yields a sharper aerial image. Results Actively controlling mask-induced phase-rather than enforcing a zero-phase condition-preserves image fidelity and pattern placement accuracy in high-NA EUV lithography. Conclusion The active phase engineering framework therefore provides a practical and quantitative guideline for next-generation EUV mask design across 0.33 to 0.55 NA systems.
Background: As extreme ultraviolet (EUV) lithography advances toward sub-40-nm single-pattern pitches, stochastic defects have emerged as a critical barrier to pattern manufacturability. These rare patterning failures, such as line bridging or contact not-open, cannot be fully captured using conventional metrics including critical dimension (CD), line-edge roughness, and line-width roughness (LWR). Aim: We aim to establish a quantitative framework for predicting stochastic defectivity by applying an extreme value statistical model to locally narrow space variations observed in EUV line-space patterns. Approach: Minimum space critical dimensions were extracted from critical-dimension scanning electron microscopy (CD-SEM) images and modeled using the Gumbel distribution derived from extreme value theory. The resulting stochastic defectivity D0 (defects/cm2) was compared with predictions based on the normal distribution. To evaluate the physical relevance of the model, Gumbel-based D0 values obtained from various published EUV resist images were correlated with pitch, dose-to-size (DtS), unbiased LWR, and Z-factor. Results: The Gumbel model predicted extreme tail defect probabilities several orders of magnitude higher than those estimated using a normal distribution, showing improved consistency with observed manufacturing limits. Pearson correlation analysis revealed strong dependence of D0 on pitch (r=-0.88) and unbiased LWR (r=+0.85), whereas Z-factor and DtS exhibited weak correlations. Conclusions: These findings indicate that stochastic defectivity in EUV lithography is governed by extreme spatial gap fluctuations rather than by average process variation metrics. The proposed Gumbel-based D0 estimation provides a physically meaningful and practical indicator for assessing EUV pattern manufacturability, although further validation using large-area measurements is required.
Background: With the continuous scaling down of integrated circuit feature sizes, the mismatch between the resolution of lithography systems and the feature dimensions leads to undesired shape distortions in wafer imaging, which can cause circuit malfunctions. These distorted regions are referred to as lithography hotspots (LHSs). Therefore, hotspot detection is crucial for ensuring the manufacturability and yield of integrated circuits. Aim: We aim to develop a high-performance hotspot detection framework that enhances feature representation, reduces false alarms, and accelerates inference. It addresses the limitations of existing methods in detecting complex and fine-grained hotspot patterns under advanced process nodes. Approach: We propose a re-parameterizable deep layer aggregation network that employs deep layer aggregation (DLA) to iteratively and hierarchically fuse multi-level features for improved hotspot recognition. The backbone integrates re-parameterizable convolutional blocks to speed up inference, whereas a multi-scale triple fusion (MSTF) module is introduced to further strengthen multi-scale feature integration. Results: We evaluated our method on standard LHS benchmarks. It outperforms state-of-the-art models in accuracy, false alarm reduction, and F1 score while cutting end-to-end runtime. Compared with prior approaches, our method shows measurable gains across all key metrics. Specifically, it delivers notable improvements in both detection performance and efficiency. Conclusions: The proposed DLA-based re-parameterizable architecture offers an effective and efficient approach for detecting LHSs, improving both accuracy and computational efficiency. We point to a practical direction for design that considers manufacturability and could be extended to future studies on hotspot prediction in emerging technology nodes and more complex patterning processes.
Background Stable and reproducible extreme ultraviolet (EUV) light generation is essential for advanced semiconductor lithography and actinic inspection. Conventional plasma-based EUV sources, such as laser-produced plasma and discharge-produced plasma, provide high output power but suffer from high power consumption, mirror contamination, electrode erosion, plasma instabilities, rapid optics degradation, and system complexity. Aim We (part I) focus on the design, long-term reliability, reproducibility, and operational stability of a vertically aligned carbon nanotube (CNT)-based cold electron-beam (C-beam)-driven EUV light source. The physical emission mechanism and source efficiency analysis are presented in the companion paper (part II). Approach We demonstrate a CNT cold cathode electron-beam (C-beam)-irradiated EUV source based on surface-excitation-driven EUV emission. The system operates at an optimized irradiation condition using a tin anode mounted on a copper jig, enabling controlled EUV generation while suppressing anode melting and cathode degradation. Results Compared with our previous lab measurements, the proposed C-beam EUV source exhibits a significant improvement in temporal stability. Short-term stability improved from 2.0% to 0.6%, whereas long-term stability improved from 4.0% to 2.5%. The source demonstrates durable and reproducible EUV emission without observable degradation of either the cathode or anode. Conclusions The CNT-based C-beam EUV source offers a compact form factor, low operating power, debris-free operation, and high temporal stability. These characteristics make it a promising candidate for actinic metrology and semiconductor inspection systems that require highly reproducible EUV emission and long operational lifetimes.
Background As integrated circuit technology nodes advance, optical proximity effects distort printed patterns and degrade yield. Existing inverse lithography technology (ILT) struggles with multi-scale feature imbalance, sub-resolution assist feature (SRAF) incompatibility with main patterns, and the trade-off among optimization efficiency, printing accuracy, and mask manufacturability. Aim This work proposes an efficient gradient-driven ILT framework to address the above issues, improving printing accuracy, optimization efficiency, and mask manufacturability for advanced technology nodes. Approach We propose MF-ILT, a framework synergizing multi-scale adaptation and frequency-domain segmentation. It integrates two core modules: (1) a statistical multi-scale soft attention (SMSA) mechanism that adaptively fuses multi-scale features via feature variance metrics, avoiding manual weight tuning, and (2) a frequency-domain split SRAF (FS-SRAF) module that generates SRAF in high-frequency edge regions and suppresses isolated artifacts. The framework also adopts an improved hybrid loss function (Huber loss + process variation band regularization) and end-to-end differentiable design to ensure efficient gradient propagation. Results On the ICCAD13 dataset and large-scale layouts, MF-ILT outperforms state-of-the-art (SOTA) methods: compared with gradient-driven methods, the average L2 error achieves a maximum reduction of 35.3%, and edge placement error (EPE) violations decrease by up to 48.6%; compared with deep-learning-based methods, the L2 error drops by 31.4% to 33.4%, and EPE violations decrease by 73.2% relative to the optimal deep-learning-based SOTA. The average optimization time per test case is only 3.1 s, with a peak memory consumption of 3.9 GB. Conclusions MF-ILT provides an efficient and robust ILT solution for advanced technology nodes, with its core modules (SMSA and FS-SRAF) effectively solving multi-scale imbalance and SRAF incompatibility issues while balancing optimization efficiency and mask manufacturability. It lays a foundation for subsequent extensions to large-chip optimization.
One of the challenges with high numerical aperture extreme ultra-violet lithography (high-NA EUVL) is the limited depth-of-focus (DOF). In this study, we demonstrate how an illumination source with the majority of pixels in the central obscuration region of the source can increase the process window of contact printing cases, especially in dynamic random access memory (DRAM) but also in logic applications, except in configurations with a center-to-center pitch smaller than similar to 30 nm. We demonstrate this through full-resist model rigorous lithography simulations. For hexagonal contact configurations such as those used in DRAM, a simple qualitative explanation of the simulated DOF increase is also presented. An experimental validation of the simulated results will be presented in future work.
Background Metal oxide resists (MORs) have demonstrated exceptional performance in high-resolution extreme ultraviolet (EUV) lithography, primarily due to their high EUV absorptivity and minimal blur radius. Innovative solutions involving dry deposition and dry development processes, tailored for MOR, significantly enhance resist compositional control and mitigate pattern collapse. However, photon shot noise and process variations continue to pose substantial challenges in meeting key performance indicators for defectivity and line-edge roughness. Aim To address these stringent requirements, a comprehensive understanding of the multistep MOR patterning process flow and the dry processing mechanisms is essential. Approach We introduce a bottom-up modeling approach for MOR, enabling the construction of a physically grounded, module-based digital twin for the photolithography patterning process flow. This platform captures the effects of various process steps on the deposited MOR system, including photon absorption derived from aerial images, secondary electron-induced blurring, post-exposure bake cleavage and cross-linking reactions, progressive dry development characteristics, and scanning electron microscope-based postprocessing insights. Each module is built upon key mechanistic insights and is mathematically formulated using probabilistic algorithms. Model parameters are directly extracted from meticulously designed blanket and checkerboard exposure experiments, ensuring physical relevance without relying on empirical fitting to patterning data. Results The end-to-end model demonstrates strong predictive capability for dose and roughness metrics, including frequency decomposition, and shows good agreement with experimental wafer benchmark data. Conclusions The validated model brings mechanistic and physical insights for resists and processes and may help reduce the complexity in MOR optical proximity correction applications.
