SemiWiki.com - Podcast EP368: An Overview of TSMC Open Innovation Platform (OIP) event with Aveek Sarkar

Episode Date: September 23, 2026

Daniel is joined by Aveek Sarkar, who heads TSMC’s Ecosystem and Alliance Management Division. Along with his team, he is responsible for driving semiconductor design enablement in collaboration with the partner ecosystem through TSMC’s Open Innovation Platform® (OIP). Prior to TSMC, Aveek worked on chip designs at Sun Microsystems… Read More

Hello, my name is Daniel Nenny, founder of Semaywiki, the Open Forum for Semiconductor professionals. Welcome to the Semiconductor Insiders podcast series. My guest today is Avik Sarkar. Avic heads TSM ecosystem and alliance management division. Along with his team, he is responsible for driving semiconductor design enablement in collaboration with the partner ecosystem through TSM's open innovation platform. Prior to TSM, Avic worked on chip designs at Sun Microsystems and held leadership positions in EDA at Apache Design Solutions, Ancys and Synopsis. Welcome to the podcast, Avic. Thank you, Dan.

Excited to be here. So first, what I like to ask is, what brought you to TSM? Can you share a little bit about your journey? Yeah, absolutely. I've been in TSM for over a year based out of the headquarters at Shinchu, Taiwan. This opportunity to be part of the team and to lead our OIP ecosystem was exciting for me in many different ways, right? First, you know, I've been both a customer and a partner in the OIP ecosystem for years. And I've seen how impactful and beneficial it could can be.

And over the years, it has grown in scope and impact, right? It's something that I have truly been impressed with. and when the opportunity came along to be part of it and to lead the next evolution, looking at AI and some of the goals that the TSMC leadership has, I jumped on it. Great decision. I think AI will be changing our whole industry. So we're on a big disruptive journey for sure. So let's talk about OIP. What are the key highlights or major takeaways from this year's TSM OIP ecosystem forum. Thank you for giving the opportunity to talk about what we are going to be showcasing from TSM and with our partners and customers in the North America

Symposium and also in a global events, right? So in that process of creating OIP, it reduced the barrier of creating designs for TSM's customers. So fast forward today, not only TSMC has built the PEOPlay Foundry model, but in many ways we have democratized silicon innovation for our industry through the ecosystem. And when I look at what we have today, about 90 members globally distributed, bringing different skills, different focus areas, both, you know, large and small companies, serving TSM's customers and enabling them to, you know, streamline, you know, streamline their design flows, accelerating their schedules, and overall de-risking the entire process to get their designs over to TSM. And this is a mission, right? Every member of the ecosystem is

relentlessly focused on. So when I look at all of this, we are obviously focusing on getting the job done today. But we call ourselves as a leadership ecosystem also for a reason. It's not just because of getting, delivering to our customers, the solutions, the tools, the IPs they need, which is today we are almost a 100,000 plus IP portfolio, six different alliances, serving different needs, not just for our largest customers, but also for all the startups and all the emerging customers. And they are the ones in many ways who benefit significantly from our ecosystems focus. But the leadership ecosystem is not just about doing the you know, job for and serving the needs of today. But it's also about looking around the corner

in comprehending and, you know, looking at what are some of the challenges that are coming for our customers, some of the things that we need to address for them in collaboration with the ecosystem. And we look at AI, the use of AI to drive design productivity. We look at system-wide optimization. And these are areas that we need to harness our collective attention, our collective strengths and in order for us to deliver the benefits of the leadership ecosystem to our customers. So that's kind of in a way, you know, of where we started from and where we are. Now going to your question about what are we going to do in the OIP forum? And this year, we have 16 different technical tracks. We have 100 papers of which about 70% of them talk about the use of AI tools or

AI use cases. We have guest speakers, including Greg Dix, who's a vice president of ASIC products group at Broadcom. He will talk about how Broadcom, his team at Broadcom, collaborates with TSM and the OIP ecosystem in delivering some of the most complex chips on the planet. So when we look at this holistically, right, this is an incredible forum for our customers and partners to come together, share insights, learn from each other, develop connections. This is also a great opportunity for networking, and this is one of the key feedbacks we get from our participants. When we look at our participant feedback over the years,

almost everybody says they will return. More than 90 plus percent give us very high scores in terms of the technical content that they get to listen to. But I'm also excited about two new things that we are doing this year. First is we are organizing what we call an emerging technologies alley. You know, given the fast pace of innovation that's happening in AI, we are seeing a lot of startup activity, new solutions that are coming out. We surveyed our customers.

We looked at some of these startups and based on the relevance to our customers' design flows, we invited six of them. So even though they are not part of OIP yet, we invited them to participate and to get exposed to TSM customers. and to TSM their needs at all. The second area that we are doing something new is in our Japan and Europe sessions, we are organizing what we are calling workshops.

Based on the feedback from our customers in these regions, they not only want to hear about what are the TSM technologies that are available to them, they also want to understand how do you get there. So we partnered with some of our EDA Alliance members and we actually have created a first-time workshop forum in which it's a combination of a TSM keynote and the partner, you know, demos and all that really ties an end-to-end workflow solutions that they can go and then

use into production. Yeah, I go to a lot of conferences and TSMC forums are definitely the number one networking conferences in the world for sure. The TSM OIP ecosystem is widely recognized as the industry's most comprehensive network of, you know, long-term design partners. What are some of the most notable recent collaborations or milestones you've achieved? The OIP ecosystem, the focus and effort started almost two decades ago. But the work with our design partners, design enablement partners, our IP partners and other constituents started many, many years ago before that. It was an important and required step

to enable this journey of DSMC to develop and deliver the pure play foundry model for our customers. The process that we went through in collaboration with our partners led to the eventual formation of OIP, which formalized the method and the process in which we interact with our partners and deliver the combined and join solutions. Now, if you look at some of the outcomes

that we have achieved throughout this process, we have now reference flows, certified tools as part of our EDA Alliance, 100,000 plus titles and IPs as part of our IP alliance, secure certified cloud environments as part of our cloud alliance, certify your design methodologies and services from our DC and VCA partners.

And then we are looking at the 3D Fabric Alliance, which has truly democratized, democratized in many ways the multi-di design process and brought in all these elements of OSAT, substrate, memory testing, EDA, and IP, all under the one tent to make sure that this AI journey can be sustained and accelerated. We continue to work with our partners, right, in a sustained manner on all the different things our customers need to be doing from a design flow enablement or the latest nodes, from making sure they have the IPs available, they have the design services available, and those happen on a continuous basis.

But what I'm really excited about is some of the new initiatives that we have started to roll out this year. As we look at the technologies that we are creating, the power efficiency gains that we are delivering, through DTCO and all these other optimizations that we do, it's not just about the technology. Our customers need to be able to use them effectively and efficiently. efficiently. And this is where we have expanded our collaboration with the ecosystem partners in standardizing some of the reference flows, some of the training methodologies, and in enabling our customers with the access to those, especially the startups and the emerging customers who are coming up with their solutions and we want to accelerate that significantly.

One of the newer things we did also this year was brought some of our alliance members to collaborate with each other, in which our EDA Alliance members actually conducted advanced no training for our design service partners. What that does, it upskills and elevates the knowledge and the learning the design service community have, so that when they go and they work on our customers' projects, they're able to deliver much more effective outcomes for them. Now, that's, you know, when we look at compute and, you know, how do you get our customers going on that. The second is our focus on design productivity. Obviously, with AI, you know, this has been a multi-year journey.

We started with AI-based, you know, optimization flows using reinforcement learnings. Those we started working several years ago, and those have become mainstays. And today, we are seeing quite significant benefit, right? You look at analog migration, design search, space exploration. Then with generated AI, obviously, that took us an step forward. But then as agentic AI and the workflows start to become more of available technologies, and this is where we are really excited in how do we work with our partners to deploy these into our customer workflows as well.

So we have identified few areas, and this year in OIP we are highlighting them. In digital workflows, for example, we are seeing through the combination of the agentic framework, and what we are calling the TSM AI design kit, certain elements that are geared towards enabling an AI-first design methodologies. We're seeing like 3 to 5x better productivity in PPA gains for digital designs or 6x productivity improvements for analog circuit migration. Of course, we will continue and the support and maintain our robust library of PDKs, right? But we believe this new framework, this TSMC AI design kit or ADK in short, right, will start to play a key role in enabling and, you know, driving faster adoption of agentic flows for our industry.

The other area, you know, we are paying close attention is listening to the voice of the customer and understanding some of the needs that they have, particularly around IPs. And this is where we started to have a lot more shift-left initiatives, especially for advanced nodes. And this year in OIP Forum, we have papers from our partners on, say, 224 gig-30s, or latest generation PCIE, or some of the IPs that they have developed on N12E for Edge and IOT applications. So these are, again, new initiatives that we have started to launch to get them accelerated. The latest focus area for us that, you know, we will start to talk about more going forward is our attention and focus on system level optimization. And this is where we are expanding our tent in some ways.

We are bringing ODMs who work on creating, you know, these AI server racks and other AI systems and bringing them closer to our OIP ecosystem and making sure that we start to create best practices and shared methodologies and to accelerate and reduce some of the challenges that they face. So with that, looking forward to all of the papers and all these other announcements that we will be having. Great. You know, energy efficiency is arguably the biggest challenge facing AI expansion today. How are OIP partners leveraging TSM's advanced processes packaging and COOP technologies

to meet the massive performance and power demands of AI infrastructure? Energy efficiency, right, or it's clearly the dominant parameter, right? In some ways, you can also say it's the passport to enabling AI. In last year's OIP, we had shared a metric in that, you know, the power per packaged part was increasing by almost 3x. But then the number of these package parts that are getting deployed in an AI, you know, in a data center was significantly more like 8x or so. So if you look at the compounding effect, the impact and the challenge of power is quite significant

and it's not just from a point of delivering the power to the transistors, but also how do you take out the heat, right, from a cooling point of view. So our focus on energy efficiency, both at TSM and through our collaboration with our OIP ecosystem members, are on three vectors, compute, integration, and connectivity. On compute, we work on our advanced logic and our DTCO focus, right? And this is we concurrently optimize the technology, the library, the IPs, the EDA flows, so that we can deliver consistently note-to-note power efficiency outcomes, right? But this effort doesn't stop at PSMC.

This is where we operate as one team with our ecosystem members. From very early stage of the PDK development, we bring in. our alliance members in which we collaborate and tune, you know, in optimizing both our technology and their workflow so that when it gets to our customers, they're able to realize the PPA entitlement very effectively. But we don't do just only for, you know, digital, even for analog, right? The complexity for analog circuits for the larger connectivity interface IPs continues to grow. So this is where we are also enabling our IP providers, giving them earlier access.

We are collaborating with them to understand some of the key care abouts they have and so that we can move forward together. Now when I shift from compute to advanced package, right, or talk about that, the advanced package has in many ways, right, co-was and now increasingly SOIC, has enabled this AI journey. Now look at SOIC specifically, right? It has 5x better energy metrics like 56x, 56x better interconnect density. So when you look at the combined benefit of the increased size of the COVOS solution and SOIC integration and the ability to integrate on SOIC more and more of our advanced technologies,

we're seeing that 48X increase in compute transistors, right? So now we are able to pack more and more transistors. But not just that, right? We are able to co-locate memory. We are able to support next generation memory architectures. And this is where our collaboration with our 3D fabric partners. And we will recognize three of them, Ska Hynix, Micron, and Samsung memory in OIP this year on the collaboration that we did with them on testing out HBM5 on co-os and, you know,

establishing best practices on thermal, mechanical stress and other areas. Coop, obviously, you know, with co-package optics is going to be a key enablement function for scale up and scale out. And this is where it's not just a technology that TSM enables to get to the 4x power efficiency and 10x latency reduction versus copper wire. But it requires significant effort and collaboration with our ecosystem partners to design these systems. You have to really co-design the electrical part and the optical part. You have to look at thermal, electrical, mechanical, all of this simultaneously. You have to have high fidelity modeling technologies like electromagnetic extraction and all. But then when you look at the advanced package overall, you need to be able to integrate all of this.

And this is where the work we did with 3D blocks language construct, which we have now contributed to I3PILE for standardization, has played a, a seminal role along all of this, right? So in summary, when we look at energy efficiency, or power efficiency or thermal and all of these, these are indeed complex challenges. And no one company can do it alone, right? This is where the OIP ecosystem and the leadership aspect of it truly shines, right?

In harnessing, I would say, the collective expertise and the knowledge that all of us bring. ring to drive in many ways repeatable first time right reality outcomes for our customers. Interesting. So AI has really increased the traffic on semi-Wiki. You know, we hooked Semawiki up to Open AI and they go through all of our content and we're actually up 250% traffic this year. But it's all about AI and we have a lot of new AI companies in the EDA section and even the existing EDA companies are talking about AI. and their tools. So, Sarkar, from your perspective, how is AI adoption progressing within design

tools and workflows in, you know, in what ways is AI reshaping the tools that OIP partners use to keep pace with this rapid demand? Yeah, it's a great question, right? Something, you know, that we always think about in terms of there are always new technologies that are coming along, right? We have to look at it in a totality, right? It's not just doing one thing. We have to comprehend what are the step one, two, and three. How do essentially work backwards from our customers' benefit scenario? How do they actually get to use it? What are the different things we need to be doing so that they actually see the benefits of all the work that we do? So our AI journey, it's not just a new

phenomena. We started this several years ago, working with our partners as the first generation of, I would say AI tools that we're looking at reinforcement learning technologies for optimization or for design search space exploration, driving, you know, faster convergence on digital design or analog design migration. We got great learning along the way, right? The collaboration starts early. We align with our partners or what kind of support they need. What do we actually need to do to meet our customers' goals? We talk to the customer, get their feedback. And through this process today, we have an established work through, let's just talk about analog migration. You can go from 40 nanometer to 22 nanometer, or you can go from our N2 to A14.

So this is the broad spectrum of technologies that you can actually do automatic design migration, right? But this new era of chip design, where a multi-agent framework will orchestrate an entire design flow and self-learning and autonomous manner solve this increasingly, you know, complex and challenging problems, this is really an exciting next step. But then we have to look at in a way that how do we make sure that it gets deployed at our customers, right? And this is where we are taking the first step towards enabling our customers and ecosystem through this vision of what we are calling the TSM-AI design kit or ADK in short. Through this, we want to enable, an AI first or an agent native design flow, right?

But we are balancing, obviously, the need for AI while balancing, making sure that the TSM IP, the customer IP, and all of this is not compromised in any way. So security and sanctity is obviously an important step in all of these discussions. But we are really excited, right, in the progress. Our partners are making and overall industries making, the collaboration we have with our customer, directly and also in conjunction with our ecosystem members. We are also staying in touch, very close touch, with a lot of the newer companies that are coming along,

both on the design side, because they have also new ideas in terms of how they want to get their designs completed, and also on the EDA side, and this is where, going back to the first time ever the emerging technologies Alley Forum that we are hosting in OIP this year to introduce some of these companies and their technologies over to do that. over to our customers. Yeah, you know, Avik, at the end of the day,

how is this collaboration driving real-world success for customers? You know, can you share an example where this ecosystem collaboration solved a particularly complex design challenge? Yeah, so, you know, this is a great question in some ways, right? Going, you know, back to some of the focus areas that I talked about in terms of compute, advanced logic, integration, which looks at advanced package, memory architectures, and connectivity, right? We have sustained focus and collaboration with our customers along all these different vectors.

Now, when you look at compute, some of the real tangible effect that we are starting to see are as we roll out these AI-based methodologies, we are seeing our customers broadly adopt the work that we have done, and the AI-based solutions are becoming more and more standard. More recently, we had one of our IP partners, for example, share information with us that they, when they migrated from N2 to A14, the migration methodologies, the PDK constructs, and all the other elements that we had put in

significantly accelerated their migration journey for them, going from our first generation nanosheet to our second generation nanosheet for some of their most complex interface IPs. On the digital front also, we are pretty much seeing that the workflows that we have put in in terms of search-based exploration, those have become now regular part of the design process. Now, as we look at the advanced package, more specifically, let's say, TSMC coupe, this is where the co-design element of electrical optical, looking at the multi-physics elements of kermal, electrical analysis, transient analysis, all of these are again the solution. that we have worked within our different teams at TSMC and also with the ecosystem partners,

we are rolling out as part of design methodologies. Now, going back to memory, this is really an exciting year for us because we work with Skihynex, Micron, and Samsung Memory on validation of HBM5 on COWS. And this is something that we are recognizing and giving them an award in our OIP forum for this particular collaboration.

This work was early co-os level validation of all the elements that need to be considered, like thermal and mechanical stress, as HBM5 technology becomes mainstream. Great. Final question, Avic. Looking ahead, what's next for the OIP ecosystem? How will TSMC and its partners continue to collaborate to enable customer design success amidst the accelerated pace of the AI era? Our OIP ecosystem has served our industry very well over the last 20 years of its existence, right?

It has truly lowered the barrier for our customers, especially, I would say, for the emerging companies and startups who want to bring new solutions and new designs and new architectures into the market. We have democratized, I would say, the silicon innovation process along the way. But as the pace, the scope, and the breadth of innovation accelerates, we are working with our partners to help us evolve and uplift

our overall ecosystem. And we are doing this both at an individually within an ecosystem, an alliance, or also collectively across the alliances driving cross alliance initiatives. One such example is we are harnessing the collective experience, and the expertise among all the Alliance members for better advanced node design success for our customers. Paying closer attention to the voice of the customers in terms of what are some of their emerging needs, both from our design flow, IP and other areas that we have and then aligning our

focus and roadmap of our partners to that. Looking at Agentic AI clearly see significant benefits, But one of the interesting things that we see from Agentic AI is as people start to use this multi-agent orchestration framework, these agents tend to be a lot more, I would say, impatient compared to the human engineers, right? They want answers and feedbacks to some of the steps that they are taking. So the newer engines that are coming up to support the Agentic framework is also becoming very interesting. So we are seeing new. class of engines, simulators, solvers, and all that are leveraging some of the heterogeneous compute like GPU CPU combination or GPUs to deliver significant speed up, or in some cases, even using custom trained models. And this is a key focus for us is how do we enable and accelerate this particular class of, you know, new generation of engines, solvers and methodologies while making sure, you know, our customer needs are met. But then we're also looking at not just at the chip and the advanced package, we're also broadening our focus to look at the system. You know,

when you look at data center, the problem of power or thermal doesn't stop at the boundary of the chip. It doesn't stop at the rack. It's a board level problem. It's a data center level problem. As you look at the next generation of human robots, for example, these are systems that are dynamically adjusting, that integrate and interact with our environment autonomously. So these are extremely complex systems with advanced sensing, advanced thinking capabilities that we have to go in with a system-first focus. And this is where our expanded focus and collaboration with these partners who are developing these systems and tying them to our OIP ecosystem so that their journey can be accelerated through shared learning and best practices.

I think these are, you know, a lot of work that needs to be done, obviously, and this is where I want to underscore my appreciation and big thank you to all our ecosystem partners around the world who focus day in and day out relentlessly to make sure that our customers' missions are served, their tapeouts happen, and their designs realize the success that they are targeting. So with all of that, we are really excited, not just for the work that we are doing today, but all the work that we'll be doing in the coming days as well. Great conversation, Avic. And congratulations on the continued success of the TSM OIP. Thank you, Dan. Couldn't be more excited and privileged to be part of this organization and of this ecosystem. So looking forward to continuing this journey.

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