A couple of weeks ago I attended SEMI’s AI Techniques in Semiconductor Manufacturing workshop in Milpitas CA, where semiconductor manufacturers, equipment suppliers, EDA vendors, and software providers shared their experiences applying AI to real manufacturing challenges.
While the use cases varied, a consistent theme emerged across nearly every presentation: the industry is moving beyond isolated AI models toward connected, governed, and increasingly agentic systems that can reason across manufacturing data and help drive action.
The Semiconductor Industry’s Biggest Challenges and Opportunities
One of the strongest messages from the workshop was that semiconductor manufacturing does not have a lack of data problem. It has a problem of fragmented data, disconnected workflows, and difficulty turning information into actionable knowledge.
Manufacturing knowledge is spread across metrology systems, equipment sensors, process recipes, engineering reports, MES systems, and the expertise of subject matter experts. As technology nodes become more complex, connecting these pieces becomes increasingly difficult.
Trust was another recurring theme. Manufacturing organizations require explainability, governance, repeatability, and human oversight before AI-generated recommendations can be operationalized.
At the same time, the industry is shifting beyond predictive analytics toward systems that can understand relationships across data sources, recommend actions, orchestrate workflows, and ultimately reduce the time between insight and impact.
Solution Approaches Demonstrated Across the Workshop
Agentic AI Built on Manufacturing Context
Multiple speakers demonstrated agent-based systems that combine analytics, machine learning, and LLMs with manufacturing-specific knowledge and workflows.
A recurring theme was the importance of manufacturing context. Several presentations highlighted the use of semantic models and knowledge graphs to connect manufacturing data, process knowledge, and engineering workflows. Rather than operating on isolated datasets, these approaches enable AI systems to reason across relationships between products, processes, equipment, metrology results, and historical engineering knowledge, producing more relevant and explainable recommendations.
Predictive Maintenance and Root Cause Analysis
Equipment monitoring and predictive maintenance was one of the focus areas. A presenter showcased an architecture combining forecasting, anomaly detection, and autonomous agents capable of generating operator-ready recommendations.
Other presentations focused on causal AI and root-cause analysis, emphasizing the need to move beyond detecting anomalies toward understanding the underlying drivers of manufacturing issues.
The common goal was clear: reduce investigation time, improve decision-making, and enable proactive intervention before quality or productivity are impacted.
AI Builds on Existing Manufacturing Tools
Another observation was that AI is not being positioned as a replacement for existing manufacturing and engineering tools. Traditional analytics, statistical methods, process control systems, and domain-specific applications continue to provide the foundation for decision-making. Instead, AI is increasingly being used to connect these capabilities, accelerate investigations, surface relevant knowledge, and orchestrate workflows across multiple systems. In many cases, the value comes less from a new algorithm and more from reducing the effort required to move from data to understanding and from understanding to action.
What Does this Means for PDF Solutions?
The workshop reinforced the relevance of the three themes that underpin PDF Solutions’ vision: Integration, Amplification, and Extension.
Integration was a consistent topic across the presentations. Whether the use case involved predictive maintenance, yield improvement, or agentic AI, success depended on bringing together data from multiple sources and providing the context needed to understand it. The recurring discussion around semantic models, knowledge graphs, and connected manufacturing data highlighted how important it is to establish a common foundation before AI can deliver meaningful value.
Amplification emerged through the industry’s focus on scaling engineering expertise. Many presenters were not looking to replace engineers with AI, but to make existing knowledge easier to access, apply, and reuse. The emphasis on explainability and trust reflected the reality that AI adoption in manufacturing depends as much on supporting human decision-making as it does on improving analytical capabilities.
Extension was reflected in the growing interest in agentic workflows. Some of the most compelling examples demonstrated how AI can help connect tools, data, and processes, reducing the effort required to move from insight to action. Rather than replacing existing manufacturing applications, AI is increasingly being used to orchestrate them, helping engineers navigate complexity and execute workflows more efficiently.
What stood out to me is that these three themes are not independent. The industry appears to be moving toward a model where connected data enables AI to amplify expertise, and amplified expertise enables more effective automation and workflow orchestration. The challenge is becoming less about building individual AI models and more about creating an environment where data, knowledge, analytics, and actions work together.
This direction closely aligns with the vision behind Exensio® Aurora, PDF Solutions’ new AI-first architecture. Many of the challenges discussed throughout the workshop were not fundamentally AI problems. They were data and operationalization problems. Manufacturing data is fragmented across tools, organizations, and supply-chain partners. Valuable engineering knowledge is often hidden inside reports, workflows, and the experience of subject matter experts. And even when insights are generated, turning them into trusted, repeatable actions remains difficult.
Exensio Aurora is designed specifically to address those challenges. It combines a semiconductor-aware data foundation, scalable analytics capable of operating on manufacturing data at petabyte scale, integrated MLOps, workflow orchestration, and agentic AI grounded in decades of manufacturing domain expertise. Equally important, workflows serve as a mechanism for capturing engineering best practices, preserving context, providing explainability, and creating the guardrails required for trusted AI adoption in manufacturing environments.
In many ways, the workshop reinforced that the future is not about deploying more AI models in isolation. It is about connecting data, analytics, domain knowledge, and operational workflows into a common system that can accelerate decision-making while maintaining trust and governance. That is precisely the direction Exensio Aurora is intended to support.
We will be discussing many of these concepts in greater detail at PDF Solutions CONNECT 2026, including scalable analytics, manufacturing-aware data models, workflow-driven automation, and the practical application of agentic AI across semiconductor manufacturing operations and the broader supply chain.
To experience Exensio Aurora first-hand, to meet with the team that developed it and to hear from early adopters of some of its capability register and attend PDF Solutions CONNECT October 15–16 in San Francisco, CA.
Visit PDF Solutions CONNECT 2026 conference website to learn more about the agenda, speakers, location, logistics and registration: https://events.pdf.com/connect2026/