SemiWiki.com - Podcast EP346: How EMD Electronics Bridges the “Lab to Fab” Gap With Ganesh Panaman

Episode Date: May 15, 2026

Daniel is joined by Ganesh Panaman, the President of Intermolecular Services at EMD Electronics. In his current role, Ganesh is dedicated to accelerating product time-to-market, securing first-mover advantages on disruptive technologies, and actively engaging with the dynamic startup ecosystem in the Silicon Valley. … 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 Ganesh Panaman, the president of intermolycular services at EMD Electronics. In his current role, Ganesh is dedicated to accelerating product time to market, securing first mover advantages on disruptive technologies, and actively engaging with a dynamic startup ecosystem in Silicon Valley.

With over two decades of experience, Ganesh has held pivotal roles across various high-tech industries showcasing a strong aptitude for technology development and customer engagement. Welcome to the podcast, Ganesh. Thank you, Daniel. It's my pleasure being here. So first, I want to ask, what brought you to intermolecular? Intermolecular is a very unique company in the way where it operates at the cutting edge of technology, particularly materials innovation. and it's been around for about two decades now. And in many ways, it's super relevant in present-day discussion, simply because we see all these large language models

being so powerful at materials prediction and property production and such. And intermolecular is the experimental half of that. So the ability to operate at the intersection of the cutting-edge technology and commercializing technology is what really brought me to intermolecular. Interesting. You know, I wanted to have you on because I heard the term lab to fab gap. So can you tell me what does that mean in practical terms for semiconductor materials and manufacturing today? Yeah. Essentially, lab to fab refers to how materials perform in different stages of its development. So when you take how much look at how materials are developed, first it starts with, there's a, there's a,

There's a chemistry identification phase on what chemistries actually can deliver the necessary material. And that's developed with certain KPI. Then it goes to an equipment and it's developed in an equipment environment. Then it goes to performance on an A-B testing type of exercise where it's compared to an incumbent material or technology and then it moves on to the FAB eventually to the foundry. So in all of these steps, what happens is that these get developed in isolation. And eventually the gaps that build up builds into a tremendous chasm, if you will, then you start from the initial material to when it actually goes into product.

So bridging this gap is what is super critical in present day semiconductor technology. That's interesting. So when you look across the semiconductor industry today, what signals tell you that the lab to fab gap is becoming a bigger constraint on progress than maybe you know the traditional scaling challenges that we see every day yeah the one clear shift is is the complexity that is present today right but particularly the AI chips there are three pillars if you will of of ships that are happening the first one is that the unit process complexity if you build a device

the number of process steps that are involved in building a device has increased. And this is primarily because of the 3D device architectures where you see almost every technology is shifting 3D. Transistors are coming off the silicon plane when you look at CFET. DRAM is going 3D, NAND is already 3D, and that's one side. So the number of process steps are also increasing along with this. The second thing that is happening is the chip disaggregation. Now, the high value logic chip and the memory array, they're all built now separately and is integrated through wafer bonding and some of the advanced packaging techniques today.

There's the backside power delivery that is coming in, which is built on a different. So you can just imagine the disaggregation creating its own set of complexities. The third thing is that node after node, the yield is also progressively dropping down. So what does the order all of the signal? Higher complexity leads to higher gaps as you go into the development cycle for these materials. And I only expect that this increase.

And really co-optimization is the name of the game where you have to start from the first molecule all the way to device design and design and integration and optimize across this chain, entire chain. So how do companies typically work with you inside the center? You know, what does effective co-development look like when materials, processes, and device performance all have to be optimized at the same time? It starts with breaking probable silos, if you will.

We need standardized environments where you can have R&D shared. spaces where you're co-developing intermolecular is a great example for that IMEC is an example of that and how do we then share data across this ecosystem to now feedback and feed forward the challenges as you see through this development cycle so this kind of approach is really what it's not it's not a choice it's a necessity moving forward when it comes to comes to collaboration collaboration is the name of it So how does intermolecular help customers de-risk new materials before they reach, you know, full-scale manufacturing? Right. So intermolecular has some very unique materials development capabilities. It starts with our philosophy, which is we want to first take a cycle of learning and speed it up. And how do we do it? We have a fully integrated set of capabilities on site here at Intermolecular.

and it starts with first the deposition so we have very unique home-built tools that can do combinatorial deposition so you can really run a hundred experiments on a single wafer and with the wafer is electrically testable you run a hundred different unique device materials combination to then evaluate performance in a single shot and then we've also built deposition tools that can that can combinatorially deposit many different compositions so one of the unique things that we did here was a few years back on the Chalpergenite space when you think of the ovanic threshold switch it's a very

complex material seven-element system so the tools that we have allows you to deposit seven-element composition space with hundred different compositions in a single way for this way you can accelerate accelerate data generation. And then when you complement this with what I mentioned earlier about these large language models and AI being able to accelerate learnings through say Bayesian optimization and such, you can converge to your target space very quickly.

And we've done this in many setups. We've done this for large integrated device makers. where fail fast is the name of the game where you don't want to work in spaces and invest too much in things that don't show promise and you want to get to the promising one fast and start-ups really and in the case of startups what typically happens is that they go develop something in a lab environment in in the university and then they want to take this all the way to the foundry and there is a big gap between these two in terms of quality in terms of repeatability of your producibility, itomolecular helps bridge these gaps for these companies as well. Yeah, you know, data and AI are often discussed at a high level, but where do they create the most, you know, the real impact today in shortening the path from materials discovery to manufacturable technologies? In a development cycle today, I think AI and the associated tools that come with it data and AI in general play a tremendous role right from in

our EMD electronics example we deploy it across across the entire R&D to manufacturing space so I'll give you an example in the early stage of development ideation ideation is where you want to figure out what molecules have a success has a chance of success at being able to able to produce the films of interest. And our smart chemists can draw out a molecule, the question then becomes can they synthesize it. Then there are retro synthesis tools that are improving every day

in terms of performance and able to, in its ability to predict molecules. Then there are DFT tools that have dramatically improved with the MLIPs, which are machine-learned intratomic potentials that accelerate DFT, which is an ability to understand simulated materials and surfaces. Then the next step in the integration would be, can you do a lab on a loop? This is having the right tool set, having the chemistries. Now can you collect, generate, formulate new materials, test them, and then feedback and let this run on a single loop

through completely with data control. And we've shown some examples of how to do as well. And then when you, once this is scale to a certain level, there is this need to bridge from this setup all the way to foundry, where anonymized data transfer between an equipment maker, a metrology company, the foundry, and a material supplier, all of it, you know, allows you to then look at data with different lenses and understand how to improve a performance. performances solve materials related problems, ultimately delivering higher performance for these materials. Right. So Ganesh, beyond AI and digital tools, what human expertise or cross-functional skills are critical to bridging the lab-to-fab gap effectively?

Yeah, so cross-disciplinary collaboration is critical to ultimately addressing the lab-to-fat gap. Materials, process engineers, device architects, manufacturing experts, all of them need to work hand in hand. And it's not just AI that is going to replace the way we work or is going to solve our problems. It's actually human intelligence guided AI that is going to make a dramatic difference. There was a use case that was actually that I came across recently, which was around trying to develop these high aspect-based.

issue edge processes and the experiment was very simple how successful is engineers by themselves in being able to solve this this problem with edge the problem is that there is biases that you see in the edge where the shapes don't all perfectly match across the entire 300 millimeter wafer how is AI good at solving this how is human intelligence good at solving this and And it turns out human guided AI solves it almost 60% faster than either of these cases. So it's just pretty dramatic improvement, I would think.

So I think we have to use all of these capabilities, a cross-disciplinary collaboration, the tools at disposal, AI and digital, and real expertise that really lies only in those subject matter experts to solve some of these problems. Interesting. Great conversation, Ganesh. I really appreciate your time. One final question is, how do customers normally engage with intermolecular? We have our website. A lot of our engagements is through reputation. I request listeners to please go to our website. The biggest,

why would anybody want to engage with us? People come to us because they have a fundamental materials problem that they need to solve. They need to bridge the lab to fat gap. We've had many success stories in helping startups go from investments to hundreds of million dollars in in money raised and being able to take their materials from lab to lab to fab. We've done that with the device makers as well. I would kindly request your audience to come to our website and understand what we do. Great. Thank you again and hopefully we can see each other sometime at one of the conferences. I guess you're here local in Silicon Valley.

Yeah, you have an invitation from me to visit our site anytime. Thank you, Daniel. Thank you, Ganesh. That concludes our podcast. Thank you all for listening and have a great day.