Dragan Djurdjanović

The manufacturing of integrated circuits (ICs) is an enabler of technological innovation and is critical to the resilience of industry and national security. Four levels of excellence are necessary as the foundation for semiconductor manufacturing: ecosystem; fab profitability; IC design for manufacturing and research & development; culture and customer trust. In this paper, the culture, semiconductor manufacturing processes, innovations of equipment, removal processes, in-line metrology for process control, and data analytics are discussed. Future topics, including advanced packaging, sustainability, fab operation optimization, and in-space semiconductor manufacturing, as well as workforce policy and ethics, are elaborated.

Achieving resilient and high-quality manufacturing requires reliable data-driven anomaly detection methods that are capable of addressing differences in behaviors among different individual machines which are nominally the same and are executing the same processes. To address the problem of detecting anomalies in a machine using sensory data gathered from different individual machines executing the same procedure, this paper proposes a cross-machine time-series anomaly detection framework that integrates a domain-invariant feature extractor with an unsupervised anomaly detection module. Leveraging the pre-trained foundation model MOMENT, the extractor employs Random Forest Classifiers to disentangle embeddings into machine-related and condition-related features, with the latter serving as representations which are invariant to differences between individual machines. These refined features enable the downstream anomaly detectors to generalize effectively to unseen target machines. Experiments on an industrial dataset collected from three different machines performing nominally the same operation demonstrate that the proposed approach outperforms both the raw-signal-based and MOMENT-embedding feature baselines, confirming its effectiveness in enhancing cross-machine generalization.

Achieving resilient and high-quality manufacturing requires reliable data-driven anomaly detection methods that are capable of addressing differences in behaviors among different individual machines which are nominally the same and are executing the same processes. To address the problem of detecting anomalies in a machine using sensory data gathered from different individual machines executing the same procedure, this paper proposes a cross-machine time-series anomaly detection framework that integrates a domain-invariant feature extractor with an unsupervised anomaly detection module. Leveraging the pre-trained foundation model MOMENT, the extractor employs Random Forest Classifiers to disentangle embeddings into machine-related and condition-related features, with the latter serving as representations which are invariant to differences between individual machines. These refined features enable the downstream anomaly detectors to generalize effectively to unseen target machines. Experiments on an industrial dataset collected from three different machines performing nominally the same operation demonstrate that the proposed approach outperforms both the raw-signal-based and MOMENT-embedding feature baselines, confirming its effectiveness in enhancing cross-machine generalization.

Semiconductor front-end fabrication involves a highly complex and flexible job shop environment, prompting extensive research into modeling system performance and supporting efficient adaptive decision-making. Considerable attention has been directed toward photolithography, as it is considered the most critical processing step, with inspection processes directly influencing both the defect detection capability and overall productivity. Typically, to ensure precision, each wafer layer undergoes thorough inspection with measurement markers spread across the entire wafer surface, resulting in long inspection times. Recent studies have shown that model-based process control coupled with the optimal down-selection of measurement markers can significantly enhance system performance. Building on this insight, this study aims to expand the analysis of semiconductor front-end fabrication by proposing a novel analytical model for the evaluation of the steady-state performance of quality and productivity. The model accounts for the material flow split based on quality attributes: defective parts are either scrapped or sent to rework stations, whereas undetected defects continue through the line. In addition, it integrates the dynamics of the full front-end process chain, offers a comprehensive representation of the system, and supports informed inspection policy decisions. This model has been effectively used to optimize inspection strategies, enabling a balanced trade-off between quality control and system productivity. Furthermore, the approach is generalizable and applicable to other manufacturing contexts that face similar trade-offs between inspection effort, resulting quality levels and throughput.

In this paper, we propose a novel method which utilizes samples of measured product quality characteristics to efficiently estimate the probabilities of those quality characteristics being within the desired specifications and, consequently, the process yield. Specifically, when dealing with 1D Gaussian distributions, we formally prove that the proposed yield estimator asymptotically gives a lower Mean Squared Error compared to the best unbiased estimator. In order to enable maximization of yield, this novel estimator is incorporated into the framework of Bayesian Optimization which iteratively seeks controllable tool parameters under which the outgoing product yield is maximized. The newly proposed yield maximization method is demonstrated in an application involving high-fidelity simulations of a reactive ion etch chamber, a tool component commonly used in semiconductor manufacturing. The aim of these simulations was to rapidly and reliably determine tool parameters that maximize the probability of delivering desired plasma density characteristics under stochastic variations in chamber conditions. The novel yield estimation and optimization methods show superiority when the number of experimental observations is limited and the distributions of outgoing product characteristics can be approximated well by a Gaussian distribution.

In semiconductor manufacturing , photolithography represents the core process of frontend fabrication as the quality outcome in terms of overlay errors depends entirely on it. Hence, particular attention is devoted to the inspection of each wafer layer, having 100 % measurements of markers distributed across a wafer with subsequent long inspection times. At the same time, process control is based on each layer’s overall measurements, discouraging companies from improving productivity by reducing inspection time. As a consequence, in this context, the product, process and system are extremely inter-related. Recent developments in joint product-process modelling show that robust model-based control coupled with optimal down-selection of measurement markers enables improved process control without decreasing the quality. However, when considering the system level effects, new dynamics should be accounted for in order to make decisions about production system configuration and operations. This paper proposes a novel analytical model for the evaluation of quality and productivity performance in manufacturing systems characterized by propagation of quality errors, process adaptation and alternative inspection policies. The proposed model is general, but particularly useful for the semiconductor sector. Application of this method to an industrial-scale semiconductor manufacturing system shows that when product-process-system are considered together, global optimal solutions can be achieved.

This paper presents a metaheuristic optimization-based approach for selecting a pre-determined number of measurement markers from the set of available markers that optimizes the performance of the recently introduced robust <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${\mathrm { L}}^{\infty }$ </tex-math></inline-formula> norm overlay control algorithm, which robustly minimizes the worst overlay error across a given pattern layer. This optimization is then used in a Design of Experiments (DOE) setting to build a tractable regression model of a customizable objective function encompassing cost effects of quality losses and throughput benefits resulting from the down-selection of markers selected for robust overlay control. Using this model, one can rapidly determine the optimal proportion of markers for any set of cost parameters, and the optimal subset forming this proportion of available markers can be down-selected to maximize performance of the resulting robust overlay controller. Overlay data and models from a semiconductor manufacturing fab were used to evaluate the newly proposed inspection and control strategy. Results clearly indicate that the novel strategic down-selection of measurement markers coupled with robust overlay control could lead to vastly improved throughputs without decreasing quality relative to what can be achieved using traditional Run-to-Run (R2R) control. Feasibility of the novel DOE-based optimization was demonstrated for two scenarios of cost-effect parameters.

Artificial intelligence (AI) that is applied in the entire chain of the new value creation and product life cycle has become the most important part of the Industry 4.0/5.0 model. The history of AI is a little more than eight decades long, and in research and development for manufacturing, it has been applied since the mid-1980s. Expert systems (ES) were the first AI tools implemented in this field. The aim of this paper is to perform a systemic analysis of the state of development and application of AI in manufacturing, which is originally used as support to the engineer, planner, designer and manager of various mechanical products. It is also used to manage processes and systems in manufacturing engineering. The paper is structured in such way that it provides answers to the following questions: how did AI models in manufacturing systems come about and develop, what are today’s models and the perspectives of applying AI in them, and some directions of future research in this area. As a special point of this paper, a review of some of our research results in this area obtained in the last few decades are presented.

The present research explores the optimization of maintenance strategies for floating offshore wind (FOW) farms using nested genetic algorithms. The primary goal is to provide insights on the decision-making processes required for both immediate and strategic maintenance planning, crucial for the viability and efficiency of FOW operations. A nested genetic algorithm was coupled with discrete-event simulations in order to simulate and optimize maintenance scenarios influenced by various operational and environmental parameters. The study revealed that short-term maintenance timing is significantly influenced by wind conditions, with higher electricity prices justifying on-site spare parts storage to mitigate operational disruptions, suggesting economic incentives for maintaining on-site inventory of spare parts. Long-term strategic findings emphasized the impact of planned intervals between inspections on financial outcomes, identifying optimal strategies that balance operational costs with energy production efficiency. Ultimately, this study highlights the importance of integrating sophisticated predictive models for failure detection with real-time operational data to enhance maintenance decision-making in the evolving landscape of offshore wind energy, where future farms are likely to operate farther from onshore facilities and under potentially highly varying market conditions in terms of electricity prices.

This paper proposes a novel method for Virtual Metrology (VM) in plasma etch processes based on analysis of all time and wavelength samples of Optical Emission Spectroscopy (OES) signals. The new method flattens each OES signal into a single vector, after which Singular Value Decomposition (SVD) is performed on the matrix formed by vectors of flattened OES signals in the training dataset. Low rank SVD projections of flattened and standardized OES recordings served as inputs for Ridge Regression, Artificial Neural Network, and Random Forest based VM models. A VM study is then conducted on a dataset gathered from a major 300 mm wafer fabrication facility, showing that the use of newly proposed SVD-based OES features consistently outperformed benchmark VM model features. Additional analysis of feature importance performed based on the analytically tractable Ridge Regression VM model form demonstrated distinct time-frequency patterns of OES signal portions that were highly informative for prediction of relevant Critical Dimensions, clearly justifying the need to use the entire OES signals for VM.

was selected as the Best Paper for 2023 by a team of Associate Editors.This paper proposes using the Dirichlet Process -Naive Bayes model (DPNB) to simultaneously impute missing values and address classification problems in semiconductor manufacturing data.The DPNB is based on the use of infinite Gaussian mixture models to model complex data distributions and estimate missing data records and variables, which is of utmost importance in semiconductor manufacturing.The DPNB performs well for high missing rates since it uses all the information from observed components.Experiments on various real datasets including semiconductor manufacturing data show that the DPNB has better performance than stateof-the-art methods in terms of predicting missing values and target variables as percentage of missing values increases.Two other papers were recognized with an Honorable Mention.a) "

The Industry 4.0 model of manufacturing transformation based on the application of various communication technologies, models for the informational connections between machines and parts, with the possibility of big data analysis facilitated by Artificial Intelligence (AI) tools and techniques. These technologies provide the users with powerful data-driven tools to create, control and improve flexible and profitable manufacturing processes. In order to truly pave the way forward to Industry 4.0 and beyond, manufacturing must evolve into a Smart Manufacturing (SM) concept, as a base model, which is fundamentally monitored and managed by AI. This paper presents a case study from a real factory in which elements of Industry 4.0 paradigm were applied, in the construction of a digital manufacturing model, as a basic step in the development of the SM model.

Fabrication of nanostructures on sapphire surfaces can enable unique applications in nanophotonics, optoelectronics, and functional transparent ceramics. However, the high chemical stability and mechanical hardness of sapphire make the fabrication of high density, high aspect ratio structures in sapphire challenging. In this study, we propose the use of optical emission spectroscopy (OES) to investigate the sapphire etching mechanism and for endpoint detection. The proposed process employs nanopillars composed of polymer and polysilicon as an etch mask, which allows the fabrication of large-area sapphire nanostructures. The results show that one can identify the emission wavelengths of key elements Al, O, Br, Cl, and H using squared loadings of the primary principal component obtained from principal component analysis of OES readings without the need of domain knowledge or user experience. By further examining the OES signal of Al and O at 395.6 nm, an empirical first-order model can be used to find a predicted endpoint at around 170 s, indicating the moment when the mask is completely removed, and the sapphire substrate is fully exposed. The fabrication results show that the highest aspect ratio of sapphire nanostructures that can be achieved is 2.07, with a width of 242 nm and a height of 500 nm. The demonstrated fabrication approach can create high sapphire nanostructures without using a metal mask to enhance the sapphire etch selectivity.

Physical fatigue accounts for many injuries in the workplace, sports arena, or battlefield. The traditional approaches to monitor fatigue rely on detecting and measuring shifts in the person’s muscular surface electromyography (sEMG) signals. However, assessing neuromuscular fatigue based purely on sEMG signals fails to account for the changing muscle dynamics during long dynamic physical tasks. To combat this dilemma, a system-based methodology has been recently developed and applied to several upper-extremity tasks. In this paper, we validate the efficacy of this novel methodology on the lower extremities during a dynamic activity. Specifically, the system-based monitoring methodology was applied to a cycling endurance task. It was statistically demonstrated that the system-based methodology resulted in a more-sensitive and less noisy metric, in comparison with an EMG-based methodology. The efficacy of the methodology was further illustrated by analyzing the inter-segmental recovering and fatiguing trends, which aligned with each muscle’s expected inter-muscle synergistic relationship.

This paper introduces stretchable, long-term wearable, tattoo-like dry surface electrodes for highly repeatable electromyography (EMG). The tattoo-like sensors are hair thin, skin compliant and can be laminated on human skin just like a temporary transfer tattoo, which enables multi-day noninvasive but intimate contact with the skin even under severe skin deformation. The new electrodes were used to facilitate a system-based approach to tracking of long-term fatiguing and recovery processes in a human neuromusculoskeletal (NMS) system, which was based on establishing an autoregressive moving average model with exogenous inputs (ARMAX model) relating signatures extracted from the surface electromyogram (sEMG) signals collected using the tattoo-like sensors, and the corresponding hand grip force (HGF) serving as the model output. Performance degradation of the relevant NMS system was evaluated by tracking the evolution of the errors of the ARMAX model established using the data corresponding to the rested (fresh) state of any given subject. Results from several exercise sessions clearly showed repeated patterns of fatiguing and resting, with a notable point that these patterns could now be quantified via dynamic models relating the relevant muscle signatures and NMS outputs.

Abstract It was recently demonstrated that scatterometry-based metrology has the capability to perform high-throughput metrology on large-area nanopatterned surfaces. However, the way this approach is currently pursued requires an a priori generated library of reflectance spectra to be simulated for an exhaustive set of possible underlying critical dimensions (CDs) characterizing the measured nanopatterns. Generating this library is time consuming and can be infeasible for complex patterns characterized by a large number of CDs. This article addresses the aforementioned drawback of optical inspection of CDs of nanopatterned surfaces through the use of an inverse problem-based optimization methodology coupled with a recently introduced approach for efficient organization of the library of previously simulated reflectance spectra. Specifically, for each physically measured reflectance spectrum, the best matching simulated spectrum is sought in the initial incomplete library in order to serve as the initial guess for the inverse problem optimization process. Through that optimization process, further refinements of the best matching simulated spectra are conducted to obtain sufficiently accurate estimates of the CDs characterizing the inspected nanopattern geometries. Capabilities of the newly proposed approach are evaluated through inspection of semiconductor wafer samples with hourglass patterns characterized by eight CDs. It was observed that one can obtain significantly faster measurements of CDs compared to inspection times associated with scanning electron microscopy, while at the same time not deteriorating the corresponding Gage Repeatability and Reproducibility. In conclusion, this method enables real-time, accurate, and repeatable metrology of CDs of large-area nanostructured surfaces with complex nanopatterns.

Roll-to-roll (R2R) nanofabrication processes are recognized as key enabling-technologies for many next-generation applications in flexible electronics, displays, energy generation, storage, as well as healthcare. However, R2R processing techniques reported in the literature currently lack a scalable method of performing high-throughput nanoscale pattern transfer of geometry requiring a high degree of fidelity in terms of critical dimension resolution, etch uniformity, and aspect ratio. Reactive ion etching (RIE) addresses the need for sub-10 nm pattern transfer with large-area uniformity in wafer-scale semiconductor manufacturing, but adapting plasma etch systems for use in R2R nanopatterning has proven to be nontrivial. Moreover, robust models for simulating R2R RIE do not exist, which is an obstacle to the creation of computational approaches to design, control, and scale-up of nanoscale R2R equipment and processes. To address these challenges, we demonstrate a process flow for fabricating Si nanopillar arrays utilizing a combination of nanoimprint lithography and RIE with all pattern transfer steps performed using a R2R plasma reactor system. Specifically discussed are process development details for etching imprint resist and Si including etch rates, cross-web etch uniformity, etch directionality, and etch selectivity at varying gas chemistries, powers, and pressures. 2k full-factorial Design of Experiments (DoEs) and ordinary least-squares regression analysis are also employed to study influence of process parameters on multiple outgoing etch quality characteristics and generate stochastic models of the R2R RIE pattern transfer process into Si. Utilizing these DOE-based models and desired targets for etch quality characteristics, we describe a bounded multivariate inverse-optimization scheme for automated etch process parameter tuning. The culmination of these efforts, to the best of the authors' knowledge, is the first reported RIE-based pattern transfer of 100 nm-scale features performed in continuous R2R fashion with control of feature geometry over large area. The methodology employed herein may be applied similarly to additional materials and geometries for future applications.