Semiconductor front-end fabrication facilities (fabs) perform hundreds of manufacturing steps in sequence to produce power electronic devices using specialized equipment such as epitaxy, lithography, and thermal processing tools. Each step is governed by Unit Process Instructions (UPIs) that specify tool configuration parameters, required bill of materials, and operator actions necessary to achieve desired outcomes. Furthermore, UPIs are organized into sequences called process flows that describe all steps required to manufacture advanced semiconductor products. Optimizing UPIs for specific products and tools, and harmonizing UPIs across multiple process flows, is critical for improving yield, reducing energy and material consumption, minimizing waste, achieving sustainable manufacturing practices, and accelerating time-to-market for emerging power electronic applications. The ATRIA project addresses these challenges in gallium nitride (GaN) epitaxy, where UPI optimization is particularly challenging due to the intricate physics governing epitaxial growth and the high dimensionality of the UPI parameter space. To overcome these obstacles, the project aims to: (1) advance digital twin creation by formalizing domain knowledge and the semantics of physics-informed processes, and (2) develop novel optimization and planning methods to refine existing UPIs, design new ones, and harmonize them across process flows. In particular, we aim to research novel hybrid AI approaches that integrate heuristic search methods with physics-informed models based on deep learning and reinforcement learning to enable efficient UPI optimization and harmonization. The proposed methods will be validated using real-world data from an industrial GaN epitaxy tool.
As artificial intelligence (AI) regulations evolve and the regulatory landscape develops and becomes be more complex, ensuring compliance with ethical guidelines and legal frameworks remains a challenge for AI developers. This paper introduces an AI-driven self-assessment chatbot designed to assist users in navigating the European Union AI Act and related standards. Leveraging a Retrieval-Augmented Generation (RAG) framework, the chatbot enables real-time, context-aware compliance verification by retrieving relevant regulatory texts and providing tailored guidance. By integrating both public and proprietary standards, it streamlines regulatory adherence, reduces complexity, and fosters responsible AI development. The paper explores the chatbot's architecture, comparing naive and graph-based RAG models, and discusses its potential impact on AI governance.
The LongEval lab focuses on the evaluation of information retrieval systems over time. Two datasets are provided that capture evolving search scenarios with changing documents, queries, and relevance assessments. Systems are assessed from a temporal perspective-that is, evaluating retrieval effectiveness as the data they operate on changes. In its third edition, LongEval featured two retrieval tasks: one in the area of ad-hoc web retrieval, and another focusing on scientific article retrieval. We present an overview of this year's tasks and datasets, as well as the participating systems. A total of 19 teams submitted their approaches, which we evaluated using nDCG and a variety of measures that quantify changes in retrieval effectiveness over time.
The industrial landscape is undergoing a transformative shift towards Industry 5.0, a paradigm characterized by the convergence of sustainability, digital autonomy, and human-centric design. This article focuses on the adoption, enhancement, and implementation of AI-driven hardware, tools, methodologies, and semiconductor technologies in this progression. We present here a comprehensive strategy from the AIMS5.0 project with the objective of connecting academic developments with practical industrial use, fostering a harmonious relationship between humans and machines to improve efficiency, spur innovation, and enhance adaptability. Hence we show here our global vision, and examples of how the creation of AI-based industrial solutions is supported by novel AI-tool chains, advancements in hardware, and tools supporting human aspects.
Search engines and LLMs are increasingly being used in learning contexts to find and access learning resources. While conventional ranking mechanisms in general-purpose search engines are based on topical relevance, in learning contexts, pedagogical suitability plays a crucial role in addressing learners' information needs, i.e. how well a resource supports a learner in expanding their knowledge within a specific context. This paper conducts an empirical study, investigating how search engine rankings and LLM rankings compare to those of human experts and learners to determine which ranking approach best supports learning. Using statistical methods, we analyze agreement across rankings collected from seven experts, 60 learners, and five LLMs over four topics. Results show that LLM rankings align more closely with expert judgments than with search engines or learners. Both experts and LLMs exhibit moderate internal agreement but differ notably from search engine rankings, indicating that conventional search engines are not optimized for pedagogical effectiveness.
