INTERNATIONAL JOURNAL OF INNOVATION AND INDUSTRIAL REVOLUTION (IJIREV)

EVOLUTION OF WAFER DEFECT DETECTION IN PHOTOLITHOGRAPHY: FROM CONVENTIONAL IMAGE PROCESSING TO DEEP LEARNING

2026-11-11T18:26:10+08:00

The increasing complexity of semiconductor manufacturing and the continuous miniaturisation of integrated circuits have intensified the need for accurate, automated, and robust wafer defect detection in photolithography. Conventional image processing techniques have provided the foundation for automated inspection; however, their dependence on handcrafted features and predefined rules limits their adaptability to complex defect patterns, process variations, and high-volume manufacturing conditions. This study presents a chronological review of the technological evolution of wafer defect detection from conventional image processing to advanced deep learning approaches. A systematic literature search was conducted using the Scopus database, applying predefined search terms, inclusion criteria, and exclusion criteria to identify relevant primary research articles. A total of 37 studies published between 1989 and 2026 were selected and analysed according to their chronological development, inspection methodologies, computational approaches, and technological contributions. The literature was organised into three temporal phases: Phase I: Pioneering and Foundational Research (1989–2019), characterised by knowledge-based inspection, conventional image processing, handcrafted features, and early machine learning; Phase II: Transition toward Intelligent Defect Detection (2021–2022), marked by the increasing adoption of computer vision, machine learning, and convolutional neural networks; and Phase III: Rapid Advancement and Deep Learning Expansion (2023–2026), characterised by lightweight networks, attention mechanisms, transformer architectures, anomaly detection, synthetic data, multimodal learning, and real-time inspection. The findings demonstrate a clear shift from handcrafted feature engineering toward adaptive, data-driven, and integrated intelligent inspection systems. Nevertheless, challenges remain in dataset availability, model generalisation, false-positive reduction, computational efficiency, explainability, and high-volume manufacturing deployment. This review provides a chronological perspective of technological development and identifies key research gaps that can guide the development of robust, scalable, and industry-ready wafer defect detection systems for next-generation semiconductor manufacturing.

COMPOSITE FERTILIZER PRODUCED FROM DECANTER CAKE AND PALM OIL MILL EFFLUENT (POME) AEROBIC SLUDGE

2026-11-11T21:34:06+08:00

Composite fertilizers offer a sustainable approach to improving crop productivity while reducing reliance on chemical fertilizers. This study explores the potential of converting Palm Oil Mill Effluent (POME) aerobic sludge and decanter cake (DC) into a composite vermicompost-based fertilizer for Capsicum annuum cultivation. The fertilizer was produced using a 40:60 ratio of POME sludge to DC, followed by vermicomposting with Eisenia fetida for 60 days. Nutrient characterization revealed that the composite fertilizer (PDC) contained higher concentrations of macronutrients (N: 2.56%, P: 7.56%, K: 9.07%) and micronutrients (Fe: 16.73%, Si: 15.86%) compared to commercial fertilizer. Plant growth performance was evaluated using five fertilizer dosages (50–250 g) and compared with a control and commercial fertilizer treatment. Results showed that a 50 g application of PDC achieved 102.96% stem elongation, equivalent to the 100 g dosage of commercial fertilizer (118.53%), indicating greater nutrient use efficiency at lower input levels. The findings demonstrate that POME–DC composite fertilizer provides balanced nutrient availability, improved soil conditioning, and enhanced plant growth potential. This study highlights the feasibility of valorizing palm oil industry waste into an eco-efficient fertilizer, contributing to circular agricultural sustainability and reduced environmental impact.

DEEP LEARNING FOR WAFER DEFECT DETECTION AND CLASSIFICATION IN SEMICONDUCTOR PHOTOLITHOGRAPHY: A STRUCTURED REVIEW

2026-11-11T21:49:12+08:00

The increasing complexity of semiconductor manufacturing and photolithography processes has intensified the need for accurate, efficient, and scalable wafer defect detection and classification. Conventional inspection approaches may face challenges in identifying complex, low-contrast, mixed-type, and previously unseen defect patterns, creating a growing need for deep learning-based solutions that can support automated and intelligent inspection. This systematic literature review examines the development of deep learning approaches for wafer defect detection and classification in semiconductor photolithography, with particular emphasis on emerging architectures, learning strategies, and industrial implementation. The review was conducted following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) framework. An advanced search strategy was applied to the Scopus and IEEE Xplore databases using the core keywords defect, detection, semiconductor, deep learning, and photolithography, covering publications from 2019 to October 2026. Following identification, screening, duplication removal, eligibility assessment, and quality evaluation, 15 primary studies were retained for qualitative synthesis. The findings were systematically organized into three themes: Theme 1, Lightweight Network Topologies and Real-Time Object Detection, highlighting computationally efficient architectures and real-time inspection capabilities; Theme 2, Self-Supervised Representation, Vision–Language Alignment, and Data-Efficient Adaptation, emphasizing approaches that address limited labelled data and improve model adaptability; and Theme 3, Industrial AOI Hardware Integration, Multi-Scale Segmentation, and Benchmark Evaluations, focusing on deployment-oriented inspection systems, precise defect localization, and standardized performance assessment. Collectively, the findings indicate a research trajectory from increasingly efficient deep learning architectures toward data-efficient learning and industrially integrated inspection frameworks. The review provides a structured synthesis of current developments and identifies key methodological directions for advancing reliable, scalable, and intelligent wafer defect inspection in semiconductor photolithography.