SemiWiki.com - Podcast EP370: An Overview of PDF Solution’s New AI Platform and Upcoming CONNECT Event with Kimon W. Michaels

Episode Date: October 2, 2026

Daniel is joined by Kimon W. Michaels, a prominent figure in the semiconductor industry, currently serving as the Executive Vice President of Products and Solutions at PDF Solutions, Inc. He co-founded the company and has held various leadership roles, including Vice President of Design for Manufacturability since June 2007.… Read More

Hello, my name is Daniel Nenny, founder of SemaiWiki, the Open Forum for Semiconductor Professionals. Welcome to the Semiconductor Insiders podcast series. My guest today is Kim and Michaels, a prominent figure in the semiconductor industry currently serving as the executive vice president of products and solutions at PDF Solutions. He co-founded the company and has held various leadership roles, including Vice President of Design for Manufacturability and Vice President of Field Operations for Manufacturing Process Solutions. Welcome to the podcast, Kimmel. Thank you. I'm excited to be here.

So first, just to start off, can you tell us a little bit about your semiconductor journey and your current role at PDF Solutions? Of course. So PDF is going to be 35 years old in about a month. So this has been my personal semiconductor career. We are a spin out of Carnegie Mellon and really started looking at the interaction of product data and manufacturing data to understand the root causes of yield loss, ways of improving yield, growing at twice the rate of yield improvement. So we've, with the industry, grown through having many companies on the leading edge to the division into the Fablis and Foundry model, to now the resurgence

in leading edge silicon, and the growing importance of looking broadly across data sets to understand causes of yield loss, operation, and efficiency quality. So it's been quite a fun journey. Yeah, I know PDF Solutions, of course. I'm local Silicon Valley, and I've been in the industry for 40 plus years. And I am familiar with you guys, but you seem to keep evolving and adding more capabilities. And, you know, it's nice to get an update. Today, PDF Solutions is working with semiconductor companies in many different areas.

it seems from technology and yield ramp to test operations and equipment integration. How would you define what really brings all of these solution areas together? At the end of the day, it's data or the need today of using more integrated and aggregate data to understand what's going on. There used to be the famous book, It Takes a Village. And I think that also summarizes the way our industry is trending. The equipment manufacturers, the FAB and Gen. the product engineers, all have a component and a responsibility in how we make these advanced chips, work, yield, hit the price points we need to. Okay. And PDF solutions recently announced

the launch of a new architecture for your analytics solution called Accentio Aurora. What are the new characteristics for the new features of this new architecture? And what led to its development? Why do companies need it now? Good question. So this is maybe the third major instantiation of our Accentio product. The original was truly a workflow or worksheet-based approach. The second generation was a template-based approach where people could script analysis as they became more complex. And Aurora Exensio, this third major evolution, is the AI-first instantiation of the Accentio platform. Well, what led to its development or why did we invest in it? The trends in our industry are the need for broader and more data to understand root causes of

performance, yield, quality, operational improvements. Handling these large volumes of data, today requires AI. AI has gotten to the point of being a real value adder versus the hope of the future. And the platform is now established to handle that. What does that mean? Well, we've added scalable analytics. It's 25 times faster performance at the same hardware cost compared to using our second generation system. It incorporates a manufacturing data house, which you is really the evolution of Accentio data and our ontology, to be AI ready for the integration, not just the fab data and product test data,

but back end production data as well, particularly important with 3D packaging and the use of chiplets. With AI requires two things. One is a platform for the creation and integration of AI. models, but also the way to model the model life cycles themselves through revisions, through deployments to the edge across multiple facilities, to the scalability required and leveraging the

scalable analytics. And today, things have to be agentic AI first. It is not a single AI model. It is an LLM enabled agents that can call other agents while still monitoring what they are doing. and avoiding hallucination. So for us, how you avoid hallucination is through the use of workflows. Workflows for us are the language and the long-term memory of the Accentio Aurora platform. Even the AI agents create workflows as part of the ML pipelines. Why is this important? Well, it lets the engineers understand what the agents did. It lets people modify as necessary, Then lock them down so as they deploy them across more engineers, across more products, you have consistency and understanding of what your AI agents are doing.

Ronald Reagan said, trust, but verify. I think that's the same for AI agents in semiconductor. You have to understand what they are doing in making sure they stay within the regions of validity as well. Interesting, trust but verify. That's a good one. That's a common mantra for semiconductor. people for sure. So AI is a big deal. Everybody's talking about using AI to help analyze semiconductor data and even design semiconductors. And some companies are directly using solutions like chat GPT or

Claude to do that. Is that a threat for PDF, you know, just to be blunt? And how do you see AI playing or enhancing your existing solutions? Another good question. I think it's more of an opportunity than a threat. I have to do a lot of meetings with investors for PDF solution. Of course, for any software company, one of the concerns are you going to be vibe coded out of existence? But I think in semiconductors, what AI does is brings increasing need for very broad data sets. It really opens up the capability of using the platform such as Accentio Aurora, our data sources, such as our e-be inspection equipment, to have a bigger impact in the company.

So why do systems like PDF solutions or platforms really being required? I think there's two interesting components about semiconductors relative to, you know, quote-unquote generic data. One is semiconductor data is physics-based, right? There is a relationship between the data, understanding what a signal off of a etch equipment means or what metrology or inspection data it may affect is knowledge that has to be incorporated into the models into the ontology of the semiconductor database to understand and make it useful for AI models. The second is you have to be able to collect this data, meaning the ability

to understand and connect to the disparate types of equipment and data streams inside semiconductor. Sometimes I like to say semiconductor was big data before big data was cool. We always had large volumes of data, but it was siloed to groups or organizations and historically was not made to work together. PDF through the Accentio Aurora platform has decades of experience of reading and aligning this data.

I think the second significant component that's unique to semiconductor is for, in particular, real-time or online analysis, you have to cross the customer vendor boundary. A large fabless company making a 3D integrated product has perhaps a six-month supply chain, taking chiplets from multiple front-end fabs through up to two dozen test insertions across maybe two OSATs, never in a facility with their name on the front door. So how do you get access to the data? you are allowed or required.

How do you move your AI models to these facilities and get them to the right piece of equipment or the right tester at the right time? This requires a trusted network. We have our SecureWise network that both you and your vendor or customer trust, not just for the secure transfer of the data, but also to ensure that you as the partners,

the partners are limited or maintain within the agreements you have as far as data access, data usage, who's allowed access to the data. The other critical requirement is your systems need to connect with the OSAT or FAB systems or test floor systems. To know how to push your AI model to the right tester at the right time requires integration with the systems within each of these facilities. And then, of course, you need to bring back all your results, integrate them into your master database in case you have an RMA in the future. So in Semiconductor, the network or the platform that is common across you, your customers, and vendors, and has the trust that enables all of you to integrate to an acceptable level within your systems is really key.

I got it. So PDF Solutions will hold its yearly conference called PDF Solutions Connect October 15th and 16th in San Francisco. I've been to it before. This is Union Square. It's a great venue. What should people expect to see at Connect this year? And why will they intend? Well, we're pretty excited about putting it on again. What we'll talk about is what we see as the industry's requirements going forward. In PDF, it's about having the right data, not just more data. And we'll talk about some of our differentiated data systems, particularly at Hart E-Pro. It's about having the platform for the integration of the data and making it AI ready.

Exensi Aurora, we announced much of the capability last year. We'll be announcing and showing examples of the product this year and how it can have impact on your business and really business outcome. And then the orchestrations of how you take analysis and turn it into business results. I think one of the thing that's unique about PDF's events is you don't just hear from PDF and you don't just hear about use cases from our customers, but we also bring in execs and other leaders to give their views of the needs of the industry and where the industry is going. So generally we get really good feedback for our events and looking forward again to this year's.

Great. Yeah, I'm definitely going to attend, so I'll see you there. So final question, how do customers normally engage with PDF solutions? I mean, you have services, you have product, you have a lot of things going on here. How do customers normally engage? A good question. And as you can imagine, there's variety across our customers base. But I think that the way to look at PDF is providing the enterprise platform or software, which enables their engineers to turn their breadth of data into results. That requires not just the IP and our software and our data systems, but the services to integrate them quickly with other systems within the FAB or product group company

and work into the solution processes that work for them. So it's predominantly a product company or an IP company where services help our customers quickly get to the solutions. Great. Thank you for your time. I hope to have you again as a guest, and I will see you next month at Connect. I hope they'll have a chance to talk to you there. Thanks for having me today.

That concludes our podcast. Thank you all for listening and have a great day.