PDF Solutions redesigns Exensio around LLM agents and scalable analytics in its Aurora architecture

PDF Solutions (Nasdaq: PDFS) has detailed a fundamental redesign of its Exensio platform, branded Exensio Aurora, that moves the product from a data analytics tool into what the company describes as a semiconductor intelligence system. The new architecture claims 25x improvements in analytics loading times and 42x improvements for critical analyses such as test structure release (TSR), according to a SemiWiki analysis published in August 2026^[1]. Scalable analytics under the Aurora architecture became available to early adopters in early 2026, with broader rollout to Exensio Cloud customers planned for late 2026.

Why the existing architecture hit a wall

The core problem is structural. Conventional analytics architectures reach fundamental limits at around 20,000 parameters, while modern chiplet and RF products routinely generate millions^[1]. Standard business intelligence tools bring data to the client - a model that collapses when tables carry millions of columns and hundreds of thousands of rows. Engineers compensate by writing bespoke scripts built on summary statistics, disconnected from original data sources and difficult to share or reproduce across teams. The result, as PDF Solutions frames it, is a fragmented analytics environment that cannot keep pace with advanced-node manufacturing complexity.

The semiconductor yield analytics tools market was valued at $1.04 billion in 2025 and is projected to reach $2.29 billion by 2034, reflecting how urgently the industry needs scalable alternatives.

Three pillars of the Aurora architecture

PDF Solutions has structured Aurora around three reinforcing layers^[1]:

• Enterprise ModelOps (Exensio StudioAI). An ML lifecycle layer designed to train, deploy, and govern thousands of models against the manufacturing datasets already resident in Exensio. Intel has licensed its Tiber AI Studio technology to PDF Solutions and plans to deploy Exensio Studio AI across its own manufacturing operations. The platform supports bring-your-own-model workflows and a secure enterprise model registry with full traceability between training data, models, and outcomes.
• Semiconductor semantic knowledge graph. A domain-specific layer spanning lots, wafers, dies, packages, test structures, and yield metrics. The graph gives language models and analytics engines the contextual grounding to interpret semiconductor data correctly, and integrates with Siemens Tessent YieldInsight to pull in design-for-test diagnosis and layout context alongside manufacturing datasets.
• LLM-driven agentic workflow platform. Autonomous agents that can query, reason over, and act on manufacturing data. Agentic tasks execute through inspectable, editable workflows that serve as guardrails - any result is traceable to the workflow that produced it. PDF Solutions offers a fully air-gapped, on-premises LLM infrastructure option so that sensitive yield data and proprietary models never leave a customer's secure firewall.

What CEO John Kibarian said in May 2026

On the Q1 2026 earnings call, CEO John Kibarian confirmed that the company's "new AI-enabled Exensio analytics systems" remained on track and were expected to reach beta release in Q3 2026, adding that customer interest had been "very high." He framed the current period as consequential for the sector, arguing that AI is changing how engineering work is performed and that executives are increasingly focused on AI's impact in R&D and manufacturing.

The next concrete milestone to watch is the PDF Solutions CONNECT 2026 conference, scheduled for 15-16 October 2026 in San Francisco, where the company has said it will demonstrate Aurora capabilities live, including LLM-driven analytics and next-generation user experiences. Whether the claimed performance gains hold up under customer workloads - and how quickly Intel's enterprise-wide deployment translates into broader commercial adoption - will be the clearest early signal of Aurora's real-world impact.