As semiconductor manufacturing systems grow more complex, the need for efficient, intelligent data handling has become paramount. From Root Cause Analysis (RCA) to Risk Assessment (RA), modern manufacturing depends on the seamless orchestration of vast, distributed datasets in different formats generated by sensors, machines, QA systems, and human operators. In order to make informed decisions and increase operational efficiency, it is essential to have in-depth expertise to correctly interpret vast amounts of data.
But here’s the challenge: today’s software tools and programming paradigms often fall short of handling decentralized environments effectively. The scale and diversity requirements of modern industrial systems outpace the capabilities of conventional architectures. These challenges align with Edge AI principles, which focus on decentralized data processing closer to the source, fostering efficient decision-making in decentralized systems. These challenges were addressed within the framework of a master’s thesis by combining Action-Oriented Programming (AcOP) with Automatic Agent Generation (AAG).
The result is a flexible, modular, scalable, and intelligent system that reimagines how decentralized data collection and analysis can be performed in semiconductor manufacturing and beyond.
Why Existing Systems Fall Short
Conventionally, software systems for RCA and RA in semiconductor environments are built using programming paradigms like Object-Oriented Programming (OOP), which continues to dominate software development across industries. While OOP excels at encapsulation and abstraction, it has inherent limitations in managing distributed concurrency, system modularity, and adaptability. These limitations become particularly pronounced in decentralized data environments.
In distributed environments, OOP’s reliance on highly structured, synchronous interactions becomes a bottleneck. The lack of native support for distributed concurrency, dynamic adaptability, and modular orchestration often results in rigid architectures that cannot scale seamlessly or adapt to evolving data flows. Furthermore, OOP systems typically require significant manual effort to handle fault isolation, data heterogeneity, and modularity in modern microservice-based architectures. These challenges result in higher development complexity, increased maintenance costs, and limited scalability, automation and fault detection-isolation in workflows.
This is where Action-Oriented Programming (AcOP) steps in, offering a paradigm shift that places actions rather than objects at the center of system design. By focusing on actions as the primary units of computation, AcOP provides a more natural and scalable approach to managing concurrency, modularity, and automation in decentralized systems.
A New Programming Mindset with Action-Oriented Programming (AcOP)
To address these challenges, AcOP was introduced into a data collection application designed to perform ETL (Extract, Transform, Load) tasks. Built on a microservice architecture, the application was used to conduct experimental scenarios and evaluate its performance based on several key metrics, including runtime efficiency, concurrency management, error and fault handling, modularity and adaptability, scalability, and implementation complexity.
Unlike conventional programming paradigms that center around objects or processes, AcOP operationalizes the concept of “actions” to drive system behavior:
- Actions: Defines the system’s functionality, containing a set of guards, and a body.
- Guards: Conditions under which actions can execute.
- Participants: Passive objects that store the data-state and play specific roles.
- Bodies: Logic that executes when triggered by actions.
- Scheduler: Coordinates and executes the actions.
Figure 1. State diagram of HPC Application with AcOP paradigm.
The AcOP structure effectively manages concurrency, ensures a clear separation between logic and data, and supports modular collaboration across distributed components.
The modularity of AcOP aligns with Edge AI principles by supporting localized, independent actions while ensuring coordination across distributed components. This approach makes the framework highly modular, scalable, and suitable for managing the complexity of decentralized data flows.
Embedding Intelligence with AAG
The second major investigation was AAG, a framework that uses large language model (LLM) based AI agents to dynamically generate agents that assist in complex tasks like coordination, data interpretation, and system monitoring with specific roles and domain knowledge.
To achieve this, two specialized AI-powered agents AAG Coordinator and AAG Analyze work together to enhance system intelligence and adaptability.
AAG Coordinator – an agent that supports the system’s scheduler and is responsible for:
- Overseeing the execution of actions in the system
- Ensuring coordination across multiple actions and participants (objects).
- Writing guards of the actions.
Figure 2. AAG Coordinator execution process in the AcOP-based HPC application.
The AAG Coordinator minimizes manual configuration by dynamically adapting the system to handle new data and workflows.
AAG Analyze – an agent that complements coordination efforts by providing advanced analytical capabilities. It is designed to:
- Perform preliminary RCA and RA analysis based on the automatically generated roles of AI agents.
- Highlight patterns, anomalies, or deviations
- Recommend next actions or signal the need for deeper analysis
Figure 3. AAG Analyze execution process.
Together, these agents make the system enabling intelligent bridging in data collection and analysis, where decisions need to be made quickly and autonomously.
The proposed AcOP-AAG framework offers several advantages, particularly for industries dealing with decentralized and distributed data landscapes for RCA and RA such as scalability, coordination of decentralized components, reduction in development and maintenance effort, as well as manual intervention, and automation of data interpretation
Industry Impact and Future Potential
While this research focused on the semiconductor manufacturing industry, the AcOP-AAG framework has broader applications across industries that face similar challenges with decentralized and distributed data. For example, sectors like IoT (Internet of Things), smart manufacturing, and autonomous systems can leverage the scalability, modularity, and adaptability of this architecture.
Modern distributed systems require local processing capabilities paired with global synchronization for effective operation. Integrating AcOP and AAG into Edge AI workflows could empower edge devices to scale dynamically, operate with greater autonomy and coordination. This research provides a foundational framework for building such intelligent, decentralized systems.
Conclusion
By integrating Action-Oriented Programming with Automatic Agent Generation, this work redefines how data collection and analysis systems can be built for the modern industrial landscape. With intelligent components like AAG Coordinator and AAG Analyze, the system is capable of managing itself, reducing developer burden, increasing agility, and enabling faster, smarter responses on the factory floor.
As the complexity of industrial data systems continues to grow, architectures that combine declarative action systems with AI-generated agents will become essential not just for semiconductor manufacturing but for any industry operating in a distributed, data-intensive environment.
The research of this master’s thesis highlights the significant potential of industry-related research and demonstrates how the insights gained can be successfully integrated in future into existing systems in industry, particularly at Infineon Technologies Austria. This presents a vast field for innovation and further development. The expertise gained in this thesis is intended to serve as a foundation for further in-depth research in a subsequent PhD, in order to build on the research findings and tackle new challenges in industry.
Blog signed by: IFAT team