Via Framework employs multiple knowledge graphs and Digital Twin Definition Language (DTDL) for the implementation of digital twins within the semiconductor industry. Digital twins are virtual replicas of physical assets, processes, or systems that require the integration of real-time data to dynamically mirror their physical counterparts. Accurate digital twins rely on integrating multi-modal data sources, including sensors, enterprise systems, and equipment data, to create a holistic and up-to-date representation.
To support these requirements, Via has developed a fine-tuned Large Language Model (LLM) based on the latest open-source data, capable of ingesting diverse data sources such as PDFs, databases, log files, and SECS/GEM data. This data is then converted into a graph-based model using Neo4j, which facilitates seamless data integration and interoperability across various factory systems. By leveraging knowledge graphs in Neo4j, the AI is grounded through structured relationships between entities, creating a robust framework for explainable AI. Additionally, Via utilizes a mixture of experts and a large action model (AI agents) to handle complex decision-making processes. These AI agents dynamically interact with the knowledge graphs to extract insights, ensuring the digital twin remains accurate, context-aware, and within predefined guardrails. This approach enhances the digital twin’s dynamic and precise representation of its physical counterpart.
We will demonstrate the Industry CoPilot application across Maintenance use cases where a set of AI agents will operate on the digital twin models and extract past maintenance history involving alarms, work orders and manuals. We will also demonstrate a Recipe Digital twin which tracks multiple recipe runs and provides a generative AI based analysis on the performance and any fine tuning that could happen.
Via Automation is building an industrial AI copilot designed to revolutionize manufacturing operations. Unplanned machine downtimes cost manufacturers over $1.5 trillion globally each year—a problem that grows more expensive annually. In fact, the cost of a single hour of downtime has surged by 50% in just two years. Traditional solutions rely on dashboards and complex visualizations, but these don't solve the root problem—actions do. Via takes a different approach in building Automation Agents that are interfacing the process control systems. It predicts and prevents machine failures in real time, acting as an engineer would by diagnosing issues and taking proactive steps to address them. Via’s AI-powered agents leverage all available factory data—structured and unstructured—ranging from machine sensor readings to maintenance manuals and work orders. With its intuitive interface, manufacturing teams can build and deploy custom workflows, even without deep AI expertise. This empowers manufacturers to focus on what matters most: keeping their critical machinery running smoothly and efficiently.
