Field report E-91 / M8 · sheet /operations

Operations & RestructuringFiled from the territory

PDF Solutions Launches Exensio Aurora, Claims 25x Faster Chip Analytics

PDF Solutions says its AI-first Exensio Aurora platform runs semiconductor analytics roughly 25x faster, targeting a disaggregated chip supply chain that runs on email and phone calls.

Grid
E-91 / M8
Surveyed
November 16, 2026
Reading
5 min read
Correspondent
Elena Vasquez
The AI-First Platform the Disaggregated Semiconductor Industry Needs
Fig. E-91 / M8The AI-First Platform the Disaggregated Semiconductor Industry Needs — ranjatm / pixabay

Field notes

  1. PDF Solutions claims Exensio Aurora runs analytics roughly 25x faster at the same cost or energy level, on the company's own benchmarks.
  2. Exensio Aurora was announced September 9, 2026, and shown publicly at PDF Solutions CONNECT in San Francisco on October 15–16.
  3. Advanced packaging is expected to grow at 26%; Qualcomm reported its data grew 600% since 2022.
  4. The first release ships as six solution bundles through an early-adopter program.
  5. Michaels said the semiconductor industry is 'not built on trust' and cross-company collaboration largely runs on email and phone calls.

PDF Solutions unveiled Exensio Aurora, an AI-first data platform for the semiconductor industry that the company says runs analytics roughly 25 times faster at the same cost or energy level, with the first release shipping through an early-adopter program.

Kimon Michaels, co-founder and Executive VP of Products and Solutions at PDF Solutions, presented the platform's rationale on October 13, 2026, at the opening day of SEMICON West, in an ESD Alliance program titled "Bridging Design and Manufacturing with AI." The platform itself was announced on September 9 and shown publicly for the first time at PDF Solutions CONNECT in San Francisco on October 15–16, 2026.

Michaels' argument rests on structural change: packaging now drives leading-edge innovation, manufacturing is spread across continents and companies, and system-technology co-optimization demands AI that no single-firm toolset can support.

Why does packaging change the data problem?

Geometric scaling in two dimensions "may not be dead," Michaels said, "but it is certainly not scaling at the same rate as before." The industry gets Dennard-style electrical scaling instead from new materials, gate-all-around devices and on-wafer 3D structures, with backside power as the most visible example. Because the reticle caps what fits on one die, the next step is disaggregating systems into chiplets and recombining them through heterogeneous integration.

Advanced packaging is expected to grow at 26% going forward. Judging the yield, performance and quality of a 3D-integrated package means aggregating data from every chiplet, design data at die and package level, back-end manufacturing data, and second-order effects such as interposer-induced stress. At PDF's 2025 Users Conference, Mike Campbell of Qualcomm reported that Qualcomm's data had grown 600% since 2022.

The data's shape matters as much as its size. An advanced part can carry 1 million to 10 million parameters across tens of test insertions, producing gigabytes of extremely wide but shallow data that shift as processes drift. Tools built for tall, narrow business data handle that shape poorly.

What does Aurora actually deliver?

Aurora sits on the Exensio Manufacturing Data House with five pillars: Scalable Analytics, Enterprise MLOps, Agentic, AI-First Interactivity, and Edge Deployments. Underneath sits FUSE, a compute engine PDF says it designed from scratch because "no distributed computation engine supports the specificity of semiconductor data." FUSE runs complex workflows, including custom Python scripts and custom models.

Michaels broke the performance claims down by use case at his CONNECT keynote, all PDF's own figures:

  • Data egress on wide data, 25K–250K columns: 18–40x faster
  • Compute vs. prior Exensio at cost parity, 5K columns × 500K rows: 16x faster
  • Compute at cost parity, 10K columns × 500K rows: 12x faster
  • Compute at cost parity, 1M columns × 60K rows: now feasible
  • NPI at a large fabless, 2M columns across a full lot: ~8 minutes to load and summarize
  • High-mix IDM, 5K–20K columns, hundreds of wafers: 15x faster for yield, Cpk, correlation
  • High-volume manufacturing at a large foundry, 2M columns × 60K rows across 1,200 wafers: ~10 minutes to load and summarize

Exensio StudioAI targets training and serving tens of thousands of models across an enterprise, with bring-your-own-model support and Kubernetes deployment. "Five years from now, no one is going to care about the user interface," Michaels said; what matters is whether the platform exposes what agents need to act on the data. Workflows serve as the system language, built from natural language through an LLM, with guardrails against hallucination. The architecture hosts an orchestrator LLM, PDF's own LLMs, public models and customer models side by side.

How does collaboration work without trust?

"If you are a Marvell or a Broadcom making an advanced chip," Michaels said, "that chip is never in a factory with your name on it." A 3D-integrated product passes through foundries, multiple OSATs, materials and tool suppliers and a system maker. Today most cross-company collaboration runs on email and phone calls. "We are not an industry built on trust," he said. PDF's portfolio pairs secureWISE for supply-chain connectivity, Cimetrix for edge equipment integration, and the Sapience Manufacturing Hub for links to ERP systems.

The target is a closed-loop feed-forward framework: automatically collect upstream data from your fab or a supplier's, transform it into model inputs, deliver it within each step's time constants, apply the model at package sort or test, and return results to the originating database. With chiplets, known good die is no longer enough. "You can't throw out a $10,000 or $20,000 part at the end if you can avoid it," Michaels said.

Who owns the data?

A Q&A exchange with a former Department of Commerce supply-chain policy official sharpened the sovereignty question: "ultimately data is gold that people monetize." Michaels agreed ownership and tracking go beyond secure transfer; permissions must travel with data and databases must enforce them, sometimes by expiring access. He predicted resistance akin to EDA's move to the cloud, and said data sharing beyond product qualification "is not something we have thought through collectively as an industry." Any system must be decentralized, "because no one party controls our industry." The goal, he said, is a "don't trust, verify" stage reached collectively.

The first release ships as six solution bundles: Scalable Analytics, Scalable Data APIs, Manufacturing Data House, Automated Analytics, StudioAI, and Advanced Feed Forward. PDF summarized the announcements as one new architecture, two major innovations, one agentic-AI approach focused on repeatability and predictability, and a global platform to deploy AI across the supply chain. Independent customer data will tell the fuller story.

via semiwiki.com (Source)

Share this article:

More from Elena Vasquez

Elena Vasquez

Show full bio

Staff writer covering companies, deals and strategy in corporate strategy and chief-executive leadership at CEO Frontier.

486 articles

Filed under

Adjacent sheets

« Previous articleNext article »

End of sheetE-91 / M8 · November 16, 2026