Intel Foundry’s Cultural Reboot

Rebooting from the outside in. Naga & Kevin showcase cultural progress at Direct Connect 2025.

I attended Intel Foundry’s Direct Connect in San Jose this week, where it was evident the company is shedding its old top-down, Intel-knows-best approach. I saw humility, acknowledgment of past missteps, and meaningful customer and partner engagement. Credit goes to the new leadership with external experience. Still, 18A remains a challenge.

Building The Foundry Muscle

One of Intel Foundry’s most significant yet underappreciated cultural shifts is its move away from the strict “copy exactly” practices rooted in Intel’s IDM legacy. Naga Chandrasekaran, Intel’s head of manufacturing, made this shift clear, referencing it several times in his keynote.

By the way, Naga’s title is “Executive Vice President—Chief Technology and Operations Officer—General Manager, Intel Foundry Technology and Manufacturing,” which is quite the mouthful 😊 Simply think of Naga as the head of manufacturing and supply chain operations.

From the keynote,

NC: As we move away from copy exact, we need Intel factories to work on improvement and that requires us to partner very closely with our ecosystem partners.

and

NC: We're ensuring that our systems and structure are changing to the new reality of serving as a foundry… The change from copy exact to continuous improvement mindset…

I’ll explain why this is so important, but it first requires a copy exact tutorial. (If you know this already, feel free to skip ahead.)

Copy EXACTLY!

As Intel explains, there was a time long-ago when R&D struggled to transition innovations from the R&D fab to the production manufacturing lines:

During the 80's, we were driven to find methods to improve our manufacturing competence. Our competitors were achieving higher yields, transferring and ramping new technologies faster than us, and, overall, running their factories more efficiently. To remain competitive, we made the basics of manufacturing technology a key piece of our agenda. One of the main issues we faced was the phenomenon of something called the Intel U. The Intel U was a predictable drop in factory performance every time a new technology or product was developed and transferred to manufacturing. This phenomenon, when plotted as a function of yield versus time, always showed a U-shaped curve. There were many approaches taken over the years to fix this problem but until the concept of Copy Exactly! was developed, we had limited success in eliminating the Intel U. Today, this phenomenon is non-existent, and Intel U stands for Intel University, not a yield problem.

Intel addressed these lab-to-production discrepancies with a simple solution: make the production environment mirror the lab precisely. And I mean precisely.

Copy EXACTLY! enables Intel to bring factories on-line quickly with high-volume practices already in place; hence, decreasing time to market and increasing yields. Copy EXACTLY! solves the problem of getting production facilities up to speed quickly by "copying" everything--process flows, equipment set, suppliers, plumbing, manufacturing clean room, and even training methodologies--from the development plant to the volume-manufacturing plant.

Same blueprint. Same paint. Same everything.

For context, you’ll often hear of Intel’s manufacturing R&D team as “TD” (Technology Development). As you would expect, they do the early pathfinding and R&D for the next-gen process (e.g. RibbonFET and PowerVia) and packaging (e.g. EMIB or Foveros). But this same team also develops the manufacturing process that sets the stage for copy exact. As explained in TD job posting:

Technology Development (TD) is the heart and soul of Moore’s Law at Intel. TD has enabled Intel to create world-changing technology that enriches the lives of every person on earth. TD’s more than 13,000 employees drive breakthrough research, develop next-generation process and packaging technologies, while also running high-volume manufacturing operations in its state-of-the-art facilities in Oregon and Arizona.

So, TD carries out copy exact; they develop a process, qualify it, and transfer it (EXACTLY!) to the corresponding fab.

Benefit

The key benefit of copy exact is that once the process is locked in, it enables a fast manufacturing ramp. Qualify it once, and quickly deploy and ramp it!

Back when Intel’s fabs existed to serve only Intel’s internal product teams, the rigid “copy exactly” playbook worked just fine. It ensured reliability and scale across a narrow product line, mostly CPUs. Sure, Intel manufactured other chips in smaller quantities (e.g. networking chips and FPGAs), but those teams were lower priority and had to conform to the established process.

Problems

Why ditch something that works? Copy exactly is inflexible and slow.

For the sake of illustration, imagine TD’s R&D occurs in Fab 1 and is then deployed to different production lines (Fab 2 and Fab 3). Imagine the Fab 3 team uncovers an issue. How does this problem get solved? It would seem the fastest and best approach is to empower the local Fab 3 manufacturing team to debug and fix the issue. But copy exact implies that any issue found by manufacturing must be handed back to TD in Fab 1 for troubleshooting, fixing, and then transfer back to Fab 2 and Fab 3.

Think about how slow large companies can be. The Fab 3 team probably had to write a white paper describing the issue, submit it to a TD committee, wait for Fab 1 to replicate the problem, run tests, and find a solution. Only after committee approval would Fab 2 and Fab 3 get the fix. That’s the constraint of how Copy EXACTLY! operates. It’s slow, but it works if every fab mostly makes the same product. (Or if you rarely go back and make fixes, whether because you spent a ton of time perfecting it up front or because you can accept some distribution of lower quality results).

Note that even when copy exact works, there’s an unfortunate byproduct: the local manufacturing teams never develop the ability to experiment and innovate. This is often called “continuous improvement” whereby a manufacturing team identifies issues, figures out process improvements, implements them, and further optimizes as possible. It’s all in the name of monotonically increasing yield and decreasing costs.

Unfortunately, by design, Intel’s local manufacturing teams never develop this continuous improvement muscle. These folks closest to the problem identify it but have to toss it over the wall to TD for an approved fix. Worse, TD’s focus is split between R&D, qualifying the next process, and fixing HVM issues, so this local fab’s problem may not even be the highest priority for TD.

Copy exact can succeed within an IDM, but it’s poorly suited for a foundry that must serve diverse external customers.

Intel’s move into the foundry business introduces complexity its fabs never faced when focused solely on internal CPU product lines. Each Foundry customer brings unique design constraints, and as a result, individual fabs must support different products, which demands local process flexibility. And the best people to resolve yield issues and adapt in real time are these same local fab teams.

If Intel wants to match TSMC’s execution level, its fabs must be empowered to adapt and continuously improve. Retired TSMC co-COO Shang-Yi Chiang indirectly said as much in this 2022 interview with the Computer History Museum:

Intel decided they want to do everything, “copy exact”. This most important principle in their R&D and their manufacturing. What it means, they develop this technology in R&D, use this equipment, use this recipe. They check everything very thoroughly. Everything great. Go to manufacturing, you never change it. You just follow that. Don't do any change. It's great. You have much lower risk. But the one problem is a year later, the new equipment has a better efficiency. TSMC will adapt that, Intel would not. So, then TSMC would begin to have a lower cost than Intel. So, TSMC tried to be very flexible, which it adapted. And one reason is, obvious reason is Intel’s system. They can sell the wafer for $20,000 a wafer, because their CPU chip's very high price. And TSMC cannot sell the wafer for $20,000. We can only sell for $4,000. And we have to try to cut the cost.

Implicit in Pat’s plan for Intel Foundry to serve external customers is the need to shed “copy exactly”.

And that’s why I was so encouraged when Naga explicitly stated that Intel Foundry is doing just that. This requires change. Cultural change.

The need to shift away from copy exact isn’t rocket science, but if you’ve only known Intel manufacturing, it can be hard to see it and make it happen.

Naga has been at Intel less than one year and has 20+ years of experience at Micron (which has its own memory fabs).

Sometimes, a fresh perspective is all you need.

Naga is a welcome addition.

Shedding Copy Exactly Works!

Here’s Naga talking about Intel 4 & 3 in Ireland shifting away from copy exactly and building local continuous improvement capabilities:

NC: I want to highlight a few cultural transformations that we have made. Over the last couple of quarters, we successfully transferred this node [Intel 4/3] to our [Ireland] manufacturing facility. In the past, Intel would have waited longer to reach maturity.

In the past, TD would have spent longer perfecting this process before sending it out to Ireland. Not anymore!

NC: We have now started moving this end-to-end to get the nodes into our manufacturing facility earlier. By moving it earlier, we have enabled our manufacturing team to start focusing on continuous improvement in reliability and delivering it to our customers.

Now Intel transfers process from TD sooner and expects the local Ireland team to iron out the kinks as they ramp the process. Much faster, and building local continuous innovation muscle at the same time!

This is precisely what a true foundry should do.

Keep in mind: cultural shifts are challenging. Long-tenured TD veterans might say, “We can’t send this to Ireland yet. It’s not ready. That’s not how we do things.” However, the new Foundry leadership is succeeding in pursuing the right path, even if it requires changing established habits. And Lip-Bu’s been talking about flattening the org, removing layers that slow things down…

Naga’s quote continued,

NC: One of the key aspects of delivering to customers is quality. And in the past, Intel's focus was on using test methodologies to build a die and sell it to our customers.

Test methodologies refers to testing chips after fabrication. This feels like a diplomatic way of saying Intel used to ramp quickly, test after the chip was built, and handle any yield variation through product binning. For example, the lowest quality CPUs are branded Celeron, average chips are i3, good ones are i5, and i7 for the best performers.

But external foundry customers expect tighter process control and don’t accept this wide yield variability. That forces Intel to introduce quality checks earlier in the flow (“shift left”) through inline monitoring.

NC: But now as a foundry, we have to shift left. We need to have more quality focus inline. Moving Intel 3 into our manufacturing facility earlier highlighted the need for increased investment in inline characterization and advanced capabilities that can drive improvements in end results. The team is able to accelerate this learning and demonstrate that we can actually conduct development and manufacturing. And one part of that is to shift from a copy exact philosophy to ensure we are innovating at every part of Intel. And now we want to make sure that innovation engine can deliver solutions faster to our customers.

This is encouraging!

Successful “shift left” is another example of cultural metamorphosis from IDM manufacturing to foundry.

Equally important, Intel isn’t simply attempting to build this Foundry muscle as it serves Intel Products teams, but rather while serving external customers who know what good (TSMC) looks like. Foundry customers expect inline monitoring and process control:

NC: We are also running Intel 16 (16nm legacy node) in Ireland and this node has taught us a lot as well. We are partnering with our first foundry customer MediaTek. And with their support, we are able to make significant learning advancements in how Intel behaves as a foundry. We want to thank our customer for teaching us about variation control, how to improve to deliver to the customers, how to respond to customers fast, and more importantly collaborating with customers by putting them at the center of success. And these learnings would not be possible if we did not have a customer working hand in hand at our factories with our [Ireland employees]. And I want to thank MediaTek for that.

Yes, we are all waiting on Intel to announce a major 18A customer (more on that later), but don’t miss how important this is: MediaTek helped Intel “learn how to behave as a foundry”.

MediaTek knows what good looks like — it’s a Taiwanese company after all!

Foundry is listening and responding. But don’t take it from Naga.

First, words from the local Ireland fab team building continuous innovation muscles. Here’s Neil Philip, Factory Manager of Fab 24 in Ireland

NP: Our foundry journey has been really exciting for Intel and Intel Ireland is proud to have run the very first foundry customer wafers. We have learned a huge amount about customer expectations and how we need to serve you. And based on this, we're making major transformations in our business, all of our systems, and our culture.

And Joe Bolger, Factory Manager of Fab 34 in Ireland, explains how they are undergoing metamorphosis:

JB: We're working hard to transform our systems, our job roles, and our skills to ensure we can perform an outstanding job for you, our foundry customers.

Cultural reboot.

It’s a good sign when the teams on the front line say change is actually happening. But it’s not real unless it comes from the customer’s mouth. Here’s Vince Hu, Corporate Vice President at MediaTek:

VH: I’m happy to share our experience working with Intel Foundry. Today, Intel Foundry’s expertise has been instrumental to enhance chip performance while minimizing power consumption and area. The Intel Foundry team has been a reliable partner, consistently supporting us in achieving our customers’ goals and pushing the boundaries of innovation. Their responsiveness to challenges has been remarkable. Whenever issues arose, they provided swift and effective solutions, helping us mitigate risks efficiently.

Not possible with copy exact. Vince continues,

VH: Given the global reliance on semiconductors, having a robust and diversified foundry supply chain is essential. MediaTek values this diversification for business resiliency and continues to evaluate expansion opportunities.

Intel Foundry’s capabilities and fab interactions and operations have been instrumental in processing our test chips. This collaboration has provided valuable insights, allowing us to refine our designs and achieve A0 first silicon success with our first product before moving to full-scale production.

Moreover, Intel Foundry’s commitment to maintaining high standards in manufacturing and reliability has been affirmed through our fab audits. Their comprehensive quality defense systems and operations disciplines ensure that MediaTek products are ready for the market.

In summary, our partnership with Intel Foundry has been positive. Their experience, reliability, and commitment to excellence have supported us in achieving several strategic goals. We look forward to continuing this collaboration and driving this innovation together.

This testimony is important because it affirms several important things:

1. The market wants a Western foundry.

2. Intel Foundry is increasingly capable of meeting that demand.

3. Mediatek is investing in that vision.

There have been questions about actual market demand for a TSMC competitor. Here’s a Taiwan-based fabless customer seeking manufacturing diversity. We tend to only talk about American fabless companies seeking American manufacturing, but here’s an Eastern designer looking for Western supply. Demand is real and bigger than most realize. When latent demand exists, supply rises up to meet it.

Intel Foundry is taking the necessary steps to meet this demand, and Mediatek is clearly invested in Foundry’s success.

There’s tension, though. You might be thinking: MediaTek’s testimony confirms the cultural reboot is real… BUT it’s on a legacy node…

I feel that tension too. Hang tight, we’ll get there.

But first: chips don’t design themselves; EDA tools do. But can off-the-shelf tools design for Intel Foundry nodes?

The Ecosystem Is Shaping Foundry

Customers want to use familiar EDA tools—Synopsys, Cadence, Siemens—and license common IP blocks to move quickly. However, Intel Foundry has been behind in tool readiness and third-party IP support.

Bringing EDA CEOs onstage early sent the message that Intel Foundry is taking ecosystem support seriously. Lip-Bu gets it, of course, given that he ran Cadence.

If Pat knew Intel, LBT knows the ecosystem.

It also felt like a quiet admission that EDA wasn’t prioritized enough before.

Just 10 minutes into the keynote, Synopsys CEO Sassine Ghazi took the stage. I recommend watching his full remarks (~10:30 mark), but here are a few key takeaways:

SG: Customers are, I want to say, desperate for optionality and for technology that they can bring in. There's huge silicon demand, as you know. And having an option like Intel Foundry and you mentioned it earlier in terms of a North America technology foundry, that's something that's very important for our customers.

It’s one thing when a chip designer like MediaTek shows interest. But when an EDA vendor says there’s demand for a Western foundry, that reflects signals from many customers, not just one.

Yet even if the demand is unquestionable, Intel Foundry’s ability to supply it was the concern. As Synopsys voiced on behalf of their customers:

SG: What they are saying initially, what they used to wonder—what is the difficulty to ramp up on an Intel technology versus what they're used to with other foundries? And I believe that the gap of design difficulty has absolutely shrunk to give you just the perspective.

There was genuine concern that Intel Foundry’s design process carried Intel-specific baggage. Ghazi says those concerns are now fading as Foundry adopts more standard industry practices. If this claim was coming from Intel, we would take it with a grain of salt. But hearing it from a top EDA vendor carries real weight.

And to Naga’s credit, he doesn’t shy away from acknowledging past issues, as he mentioned some design challenges customers have had with Intel during this keynote:

NC: On Intel 3, there are some ease-of-design-use challenges, but we want to partner with customers to develop their offerings on this node…

Lastly, Sassine emphasized that Intel can only win when it partners with the ecosystem:

SG: You cannot only win with technology; you need to have the whole process of enablement ready in order for the customer to see it as viable.

True end-to-end enablement means working closely with the ecosystem, and that only works if Intel humbles itself and drops its arrogant, insular mindset. With leaders like LBT, Naga, and Kevin O’Buckley now in place, that shift is underway.

Kevin is another veteran with a strong fab background. He joined Intel a year ago after stints including GlobalFoundries and IBM.

Anirudh Devgan, CEO of Cadence, reinforced the point that Intel must move away from its inward-facing habits. In response to Lip-Bu’s question, he emphasized the need for more humility and a sharper focus on partner and customer needs.

LBT: With what do you think of Intel Foundry and how we can improve --- what are the changes?

AD: I'm so delighted that Lip-Bu is there at Intel because it's an national asset… Now, Intel, through its rich history, needs to do more, listen to customers. I think that's a big change…I think Intel a great company, but can't do it by themselves, need the partnership of the ecosystem, which I think is even more important for the foundry… So I'm glad you are doing that, too.

Sure, EDA CEOs saying “work with us more” might sound self-serving. But seeing them publicly remind Intel that the ecosystem matters was refreshing. And to Intel’s credit, the partners’ presence on stage showed that LBT and Intel Foundry are listening.

Later in the day, Cadence’s Boyd Phelps, SVP and GM of the Silicon Solutions Group at Cadence, gave an engaging presentation detailing Cadence’s partnership with Intel Foundry.

First, Cadence is enabling Foundry’s roadmap of 18A, 18A-P (🤔), 14A, and 14A-E (🤔). Cadence will support Intel’s RibbonFET Gate-all-around transistors and PowerVia backside power delivery network and is shipping IP libraries based on 18A/P including SerDes, LPDDR5, UCIe 1.0, with more coming soon.

Slide courtesy Cadence.

Cadence supports Foundry’s advanced packaging too:

Slide courtesy Cadence.

Intel Foundry is the only place to get both backside power and GAA today, and Cadence is fully supporting that stack along with advanced packaging for chip-to-system DTCO.

Cadence is also working to enable 14A’s second-gen backside power and GAA, both improved.

There’s an interesting argument that isn’t circulating broadly: Why wait for the competition to enable backside power when you can get on it today and get the second-gen improvements tomorrow?

Relatedly, Intel showed an interesting slide arguing PowerVia lets you simplify the frontside by using wider pitches and cheaper patterning methods. That cost reduction helps offset the complexity and expense of implementing backside power. So again, why wait for the benefits of backside power?

The point of this ecosystem exposition is to emphasize that Foundry can’t win without EDA enablement. And it was very clear that it’s happening.

But yeah, you’re wondering what that 18A-P is. And 14A-E. And was there any 18A customer announcement? I skipped those as we focused on the positives, but now let’s address it.

So… what about 18A? 18A-P? 14A-E?

So we covered all of the positives. Foundry is maturing. Standard EDA and IP support are real, and MediaTek’s success on a legacy node demonstrates that loosening copy-exact is starting to pay off.

But where’s the 18A customer announcement? Where’s the proof that Intel can enable leading-edge silicon customers?

And what’s 18A-P? Is there something wrong with 18A?

We’ll dive in behind the paywall.