{"title":"A Multi-Objective Modeling Framework for Joint Performance and Economic Optimization in AMOLED Manufacturing Processes","authors":"Runqi Chai;Nan Liu;Jianhua Zhao;Hong Wang;Senchun Chai","doi":"10.1109/TSM.2026.3742564","DOIUrl":"https://doi.org/10.1109/TSM.2026.3742564","url":null,"abstract":"Process development in AMOLED manufacturing involves complex trade-offs between device performance and production economics under tight manufacturability constraints. Conventional process optimization approaches often rely on experience-driven parameter tuning or focus on isolated technical objectives, making it difficult to achieve systematic and reliable decision-making in high-dimensional and uncertain process spaces. In this paper, we propose a multi-objective modeling and optimization framework for joint performance and economic optimization in AMOLED manufacturing processes. The framework adopts a hierarchical parameter–performance–economics modeling structure to explicitly capture the interactions between process parameters, device performance indicators, and manufacturing economic metrics. Probabilistic safety constraints and uncertainty-aware surrogate modeling are incorporated to ensure reliable optimization under process variability. A safety-constrained multi-objective optimization strategy with trust-region control is developed to balance exploration efficiency and decision stability. A historically grounded case study shows that the proposed approach can generate feasible Pareto-optimal process configurations, improve constraint feasibility and normalized economic indicators relative to unconstrained search, and produce interpretable performance–economics trade-offs, while yield becomes more conservative. The results indicate that uncertainty modeling and trust-region control improve search stability in this evaluation.","PeriodicalId":451,"journal":{"name":"IEEE Transactions on Semiconductor Manufacturing","volume":"39 4","pages":"788-799"},"PeriodicalIF":2.5,"publicationDate":"2026-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149399508","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Guest Editorial Special Section on the 2026 International Conference on Compound Semiconductor Manufacturing Technology (CS-MANTECH)","authors":"Patrick Fay","doi":"10.1109/TSM.2026.3744074","DOIUrl":"https://doi.org/10.1109/TSM.2026.3744074","url":null,"abstract":"","PeriodicalId":451,"journal":{"name":"IEEE Transactions on Semiconductor Manufacturing","volume":"39 4","pages":"673-673"},"PeriodicalIF":2.5,"publicationDate":"2026-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11795524","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149399516","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Differentiable Etch Proximity Correction With Transformer-Based Optimization","authors":"Sooyong Lee;Guandao Yang;Suyeon Choi;Seongtae Jeong;Gordon Wetzstein","doi":"10.1109/TSM.2026.3724360","DOIUrl":"https://doi.org/10.1109/TSM.2026.3724360","url":null,"abstract":"We introduce a novel deep learning framework for etch proximity correction (EPC), termed differentiable EPC (DEPC), which enables end-to-end optimization of after-etch patterning prediction and subsequent semiconductor layout correction. Unlike conventional EPC methods that rely on local corrections, this approach enables collective pattern optimization. The framework utilizes a transformer architecture incorporating relative coordinate encoding and masked attention to process layout data represented as polygons, thereby eliminating the prediction model’s patch dependency to scales well below the database-unit level. Additionally, we propose an inverse differentiable correction algorithm that enables convergence of targeting-off values within a database-unit level threshold across the entire chip layout. Using synthetic data generated to represent real-world semiconductor fabrication scenarios, we evaluated the performance of our proposed method and found that it reduces prediction error by 6.6% compared to hybrid ML model-based EPC and lowers revision verification scenario error by 6.5%, while maintaining accuracy close to the stochastic noise limit. The embedded-space inclusion analysis indicates the usefulness of perturbative training samples within design-rule bounds, enabling coverage of the corrected-layout embedded space. By reducing variability in critical dimensions and improving targeting accuracy, this approach contributes to improved process margins and manufacturability in semiconductor fabrication.","PeriodicalId":451,"journal":{"name":"IEEE Transactions on Semiconductor Manufacturing","volume":"39 4","pages":"811-821"},"PeriodicalIF":2.5,"publicationDate":"2026-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149399505","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
James C. Gallagher;Nadeemullah A. Mahadik;Alecsander N. Imhof;Robert E. Stahlbush;Alan G. Jacobs;Emma Rocco;Kasey Hogan;James R. Grandusky;Robert T. Bondokov;Karl D. Hobart;Michael A. Mastro
{"title":"Computationally Efficient Full Wafer Scale Mapping of Basal Plane Dislocations in SiC Using Machine Learning","authors":"James C. Gallagher;Nadeemullah A. Mahadik;Alecsander N. Imhof;Robert E. Stahlbush;Alan G. Jacobs;Emma Rocco;Kasey Hogan;James R. Grandusky;Robert T. Bondokov;Karl D. Hobart;Michael A. Mastro","doi":"10.1109/TSM.2026.3745221","DOIUrl":"https://doi.org/10.1109/TSM.2026.3745221","url":null,"abstract":"Rapid non-destructive evaluation of SiC substrates and epitaxial layers is highly advantageous for the development and manufacturing of SiC structures and devices. Various extended defects form during boule growth, substrate processing, epitaxy, and device fabrication. Basal plane dislocations (BPDs) are particularly problematic as these defects can function as a source of stacking faults, including generating stacking faults during device operation. X-ray topography (XRT) is effective at detecting extended defects; however, mapping them requires automated detection of microscopic defects on a macroscopic scale. This work presents a U-Net based neural network that can rapidly define the location and shape of BPDs in large-area XRT maps of SiC substrates and epitaxial layers. The model achieves a pixel-wise accuracy of 94% and an F1 score of 0.865, enabling computationally efficient quantification. The developed model can analyze a 150 mm wafer in minutes on a standard desktop workstation and has been shown capable of mapping similar dislocations in other materials such as AlN with an F1 score of 0.82.","PeriodicalId":451,"journal":{"name":"IEEE Transactions on Semiconductor Manufacturing","volume":"39 4","pages":"715-721"},"PeriodicalIF":2.5,"publicationDate":"2026-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149399506","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Comparison-Based Anomaly Detection of Semiconductor Devices With a Projection-Based Co-Attention Module","authors":"Seungmi Oh;Jeongtae Kim","doi":"10.1109/TSM.2026.3730269","DOIUrl":"https://doi.org/10.1109/TSM.2026.3730269","url":null,"abstract":"In this paper, we propose a domain-specific unsupervised comparison-based anomaly detection (AD) framework for semiconductor inspection. Due to the rarity of defects in semiconductor devices, it is impractical to collect sufficient labeled defective data for supervised learning. Therefore, unsupervised AD methods are promising alternatives that learn representations of normal data and identify anomalies as deviations from the learned representations. However, most existing unsupervised AD methods struggle to identify abnormal regions in complex and intricate normal patterns that are frequently observed in semiconductor products. To address this challenge, leveraging the unique repetitive structure of semiconductor devices, we propose a comparison-based AD system that compares a test device with its corresponding reference device. Specifically, our method utilizes a weight-shared twin encoder to compare test and reference images in a shared feature space. We then refine each feature map through a projection-based co-attention (PCA) module which estimates channel correlations and registration errors using horizontally and vertically projected feature maps to effectively suppress false positives arising from pseudo changes (<italic>e.g.,</i> misalignment or illumination variation). Based on these refined maps, we compute difference features that tend to concentrate near zero in normal regions while producing larger deviations in defective regions. Finally, these features are fed into a pixel-wise decoder for defect localization and into a normalizing flow (NF) network for density estimation of normal data.","PeriodicalId":451,"journal":{"name":"IEEE Transactions on Semiconductor Manufacturing","volume":"39 4","pages":"867-878"},"PeriodicalIF":2.5,"publicationDate":"2026-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149399543","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Drift-Aware Offline Reinforcement Learning for Run-to-Run Control in Chemical Mechanical Planarization","authors":"Jongkwan Choi;Chang Ouk Kim","doi":"10.1109/TSM.2026.3720649","DOIUrl":"https://doi.org/10.1109/TSM.2026.3720649","url":null,"abstract":"Classical run-to-run (R2R) controllers for chemical mechanical planarization (CMP)—the double exponentially weighted moving average (D-EWMA) and the Kalman filter—share a reactive structure: they observe removal-rate (RR) error after each wafer lot and issue a corrective recipe for the next. As wear shifts the process gain between lots, each correction is mismatched to the current state and recipe oscillation grows as consumables age. This paper reformulates CMP R2R control as an offline reinforcement learning problem and proposes <sc>SARC</small> (State-Adaptive R2R Controller), an offline RL framework that avoids online exploration by replacing reactive error correction with proactive wear adaptation. A <italic>Drift Encoder</i>, pre-trained via an auxiliary RR-prediction task, distills wear telemetry into a 2-D context vector; a conservative Q-learning (CQL) policy regularized toward the behavioral distribution then maps the state–context pair to zone-pressure recipes. We evaluate <sc>SARC</small> on a CMP drift simulator (mild and medium drift), the public Prognostics and Health Management (PHM) 2016 benchmark, and a 23,802-wafer industrial fab dataset. On both fab datasets, in model-based evaluation, <sc>SARC</small> attains the lowest RR tracking error among the controllers compared without inflating control effort, improving the accuracy–stability trade-off of reactive R2R control. Ablation studies identify the Drift Encoder as the source of the recipe-stability gains.","PeriodicalId":451,"journal":{"name":"IEEE Transactions on Semiconductor Manufacturing","volume":"39 4","pages":"846-853"},"PeriodicalIF":2.5,"publicationDate":"2026-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149399523","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Physics-Guided Masked Transformer for Read Reference Voltage Prediction in 3-D NAND Flash","authors":"Won Jun Hwang;Chang Ouk Kim","doi":"10.1109/TSM.2026.3733667","DOIUrl":"https://doi.org/10.1109/TSM.2026.3733667","url":null,"abstract":"Read reference voltages (<inline-formula> <tex-math>$V_{mathrm {ref}}$ </tex-math></inline-formula>) are critical to the reliability of 3D NAND flash memory. However, the optimal <inline-formula> <tex-math>$V_{mathrm {ref}}$ </tex-math></inline-formula> continuously shift due to program/erase (P/E) cycling, retention loss, and process-induced variations, leading to reduced first-read success probability (<inline-formula> <tex-math>$P_{mathrm {1st}}$ </tex-math></inline-formula>) and more frequent read retry (RR) operations, which significantly increase read latency. Accurate prediction of <inline-formula> <tex-math>$V_{mathrm {ref}}$ </tex-math></inline-formula> is therefore essential, yet it remains challenging in practical solid-state drive (SSD) environments where only limited observations are available. In this work, we formulate degradation-aware <inline-formula> <tex-math>$V_{mathrm {ref}}$ </tex-math></inline-formula> estimation under partial observations and propose a Physics-Guided Masked Transformer (PGMT) framework. By integrating physical degradation priors into a structured latent representation, the model enforces physically aligned trends and enables robust inference with partial observations. Experimental results demonstrated enhanced prediction accuracy and robustness under out-of-distribution conditions. An approximate system-level analysis further indicated that the improved <inline-formula> <tex-math>$P_{mathrm {1st}}$ </tex-math></inline-formula> can substantially reduce RR overhead. These findings demonstrate the potential of combining physics-guided representation with sparse observations for robust <inline-formula> <tex-math>$V_{mathrm {ref}}$ </tex-math></inline-formula> estimation under complex degradation conditions.","PeriodicalId":451,"journal":{"name":"IEEE Transactions on Semiconductor Manufacturing","volume":"39 4","pages":"760-767"},"PeriodicalIF":2.5,"publicationDate":"2026-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149399504","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"SEM Image Denoising Using Noise-Step Correlation-Based Reverse Diffusion Process","authors":"Ruipeng Xu;Keye Wang;Kai Zhang;Yifei Wang;Sihai Zhang","doi":"10.1109/TSM.2026.3740902","DOIUrl":"https://doi.org/10.1109/TSM.2026.3740902","url":null,"abstract":"Diffusion models have achieved remarkable performance in image denoising through iterative reverse processes. However, conventional fixed-step inference assumes uniform noise levels, an assumption that is often violated in Scanning Electron Microscopy (SEM) imaging, where varying acquisition conditions lead to significant noise fluctuations. This mismatch may result in either residual noise or excessive structural smoothing. To address this issue, we propose a diffusion-based denoising framework that determines the reverse inference steps according to the estimated noise level of each input image. By analyzing the progressive denoising behavior of the diffusion process, we reveal a stable and approximately linear relationship between the effective noise level of the input and the optimal inference step. Based on this observation, a wavelet-based noise estimation method is employed to quantify the noise intensity, and a Noise-Step mapping model is constructed to guide the reverse diffusion process. Experiments on an industrial semiconductor SEM dataset demonstrate that our method significantly improves restoration stability compared with conventional fixed-step strategies, effectively balancing noise suppression and structural preservation. Furthermore, ablation studies show that weak denoising preprocessing during training can reduce noise uncertainty and enhance the robustness of the learned diffusion prior.","PeriodicalId":451,"journal":{"name":"IEEE Transactions on Semiconductor Manufacturing","volume":"39 4","pages":"768-776"},"PeriodicalIF":2.5,"publicationDate":"2026-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149399509","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Experimental Evaluation of Polarization-Graded AlGaN/GaN HEMTs With Electric Field Management and Improved Breakdown Performance","authors":"Yu-En Jeng;Weifeng Wu;Sothio Suzue-Pan;Pengcheng Xu;Juncheng Xiong;Zhongtao Zhu;Wesley Turner;Patrick Fay","doi":"10.1109/TSM.2026.3732609","DOIUrl":"https://doi.org/10.1109/TSM.2026.3732609","url":null,"abstract":"Polarization-graded AlGaN/GaN high-electron-mobility transistors (HEMTs) are experimentally evaluated to assess their potential for improving device breakdown voltage and linearity performance. Incorporation of a 10-nm polarization-graded AlGaN transition layer between the channel and barrier increased the breakdown voltage from 50 V for the conventional abrupt-interface device to 70 V; with a 2-<inline-formula> <tex-math>$mu $ </tex-math></inline-formula>m source field plate, 90 V was obtained. The graded devices exhibited a 53% reduction in the transconductance second derivative (g<sub>m</sub>”) and a 54% reduction in the gate-source capacitance second derivative (C<sub>gs</sub>”) relative to the abrupt-interface devices, corresponding to an estimated 15.5 dB reduction in intermodulation products. The graded devices exhibited slightly lower f<sub>T</sub> and f<sub>max</sub> (16.6 GHz and 46.3 GHz) compared to abrupt-interface devices (17.5 GHz and 56.5 GHz), attributed to the higher sheet resistance of the graded heterostructures. The increased breakdown voltage nonetheless yields a Johnson’s figure of merit of 1.16 THz<inline-formula> <tex-math>$cdot $ </tex-math></inline-formula>V, 33.2% higher than that of the abrupt-interface device. TCAD simulations calibrated to the measured DC characteristics indicate that the redistribution of polarization charge in the graded structure reduces the peak lateral electric field in the gate-drain region.","PeriodicalId":451,"journal":{"name":"IEEE Transactions on Semiconductor Manufacturing","volume":"39 4","pages":"702-707"},"PeriodicalIF":2.5,"publicationDate":"2026-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149399520","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}