Interview with Intel’s Dr. Ann Kelleher
The One In Charge of The Nodes
Companies mentioned: INTC 0.00%↑
If there’s one marketing message that Intel is drilling into the ecosystem about its capabilities, it’s the ‘five nodes in four years’ mantra. It’s hard to escape – for a company that sat at the peak of semiconductor manufacturing for many years, to be behind, the drive to retain that number one spot has been like no other. Insert a favoured CEO in Gelsinger, and then the need for a team in technology development ready to execute on the actual silicon. That team is headed up by Dr. Ann Kelleher, an Intel grandee who I’ve had chance to meet on a couple of occasions in the past few years. Ann’s history is that of fab management and execution – on our first meeting she made it very clear that she doesn’t accept anything less than on-time execution, and is putting in the process such that Intel will hit its milestones on time (and avoid what happened in the last 10 years with 10nm).
This is a video and transcript of my interview with Ann while at the opening of Intel’s Fab 34 at Intel Ireland. Fab 34 is Intel’s first high-volume manufacturing factory dedicated to the use of EUV and the Intel 4 process node. With the official opening of Fab 34, Ann is calling Intel 4 delivered, and we were both in town to see that opening with Intel CEO and the Irish Taoiseach. Ann is always a great person to talk to because she gets it – she understands the complexity and what’s needed to make the next generations of silicon hit new and exciting boundaries.
I last interviewed Ann in February 2022, which you can watch at this link. The video below is from this interview, followed by a transcript that has been edited for reliability.
Ian Cutress: We’re here because the new Fab 34 is Intel's EUV high-volume manufacturing facility. There is lots of excitement about Intel 4 and the use of EUV, but from your perspective, has Intel's development of EUV performed as expected?
Ann Kelleher: My simple answer is yes, it has been successful. I'll break it down into two pieces - our use of EUV in the Intel 4 technology has been very successful, and also our readiness in terms of manufacturability has progressed really well. We've been working very closely with ASML, who are the equipment suppliers, and their teams and my team have delivered really good uptime and really good productivity from the toolsets over the last 12 months.
IC: Would you consider it market leading use of EUV?
AK: Well, of course! I know from the benchmarks in terms of our manufacturing metrics that we are very well matched, if maybe slightly ahead of the rest within the industry, and I also believe that our use of it within the technology has leadership.
IC: One of the benefits of EUV is reducing layer counts, ultimately improving yield, but it also reduces cycle time and cost, initial outlay notwithstanding. When developing new process node technologies, does that cycle time and cost equation come into account?
AK: Yes it does, and each time we make a decision in terms of what type of lithography we use at a given layer, there is a cost question and cost equation. The cost equation takes into account the cost of running that section of the process flow - it comes into the cycle time, it comes into the footprint of the factory you use. So absolutely the analysis is done. And there is the technical capability and needs to be looked at, and then what is the cost.
IC: We're here in Ireland for the opening of Fab 34 - Intel's high volume manufacturing for Intel 4. And Intel 3?
AK: Intel 3 is coming, after Intel 4.
IC: You've told me in the past that nodes are often developed in Oregon first, and the technology is transferred to fabs like this one. Can you take me through that process, timeline, key milestones?
AK: Absolutely. So first of all, if you think about process development, I go back as far as when we do early research. Some of the early research can be 6-10 years before process gets into development. Then it goes from early research into what we called pathfinding, which is basically which portions of those early research will have the probability of making it into the final process node. Then we go from that into development, and that takes about 2-3 years depending on what we're developing. Then ultimately in Oregon we do pilot lines and we do a certain amount of manufacturing, and once we've proven that out, then we transfer it to the other high volume manufacturing factories. Today I'm very happy to be here, to formally pass the baton to Fab 34.
IC: When you've got it ready in Oregon, can you take me through the steps from there to deployment, the transfer.
AK: First of all, we match our factories. We match our factories - the equipment is matched to our factory in Oregon, the process flow is matched to the factory in Oregon. Then we qualify each section of the line to ensure that each section is matched, and then once we know we have a capability in the matching then the product goes in the line.
IC: Was that seamless for this EUV process?
AK: Actually it’s gone really really well. One of the things I was thinking about the other day, typically in the role I have, there would have been many escalations [of major issues] and in previous start ups and transfers. I've had zero escalations to me over the last 2 years, so that gives a measure of how successful [it has been]. It’s not easy, right, but it means, problems will happen, but the teams were working really well together to resolve the problems.
IC: Did you do anything specific to have zero issues, as it were?
AK: Well I think, first of all, there was a seed team [from Ireland] which was based in Oregon. That team knew my team very well, and also we resourced - we had very active transfer systems, and a group of people who were dedicated from my team to help the transfer into Ireland. I think that really worked well, knowing each other well, knowing and working together sometimes doing the same task forces together really helped to make it as smooth as it was.
IC: How much of that technology transfer is transferring actual physical hardware compared to just buying new?
AK: Well our process when we do it, we have a process called "copy exactly" - the majority of the equipment in here in Fab 34 is the same as the equipment that we have in Oregon. Some of that is reused equipment from previous nodes, which is qualified and we know it matched and will deliver what we expect. The rest of the equipment would be new purchases which would be the same, or maybe an upgrade to what we'd have in Oregon.
IC: Ireland has had a long manufacturing history with Intel, and you've played a key part of that being the literally factory manager for Fab 24 next door. What makes Ireland such a good place for Intel's manufacturing?
AK: I think there is a really strong workforce, there’s an educated workforce, there’s a dedicated workforce. And I think people are focused on the technical problems and all working together as a team to solve them.
IC: I know the Taoiseach is here for the opening today, and I'm sure there are some government in there as well.
AK: In any place we have our factories, having an environment which is pro-business is absolutely critical. Also the ecosystem is very important, so Ireland has done a really good job in working with us - having us as part of their community, as well as having the pro-business approach of enabling us, as well as the ecosystem enabling us.
IC: We had Intel innovation a few weeks ago, and the CEO Pat Gelsinger talked about this roadmap, the sort of "5 nodes in 4 years" - the thing that’s been drilled into our skull every time we have had a presentation from Intel, which is good marketing. At Innovation we saw upcoming product on Intel 4 and Intel 3, and then wafers of 20A and 18A, Last time we spoke, you highlighted the need for time to market, and the ability to, when you have new features, have backups. How much of that has gone to plan of late?
AK: So overall, we have pulled in our backups and our contingency plannings as we were looking at our new process nodes - anywhere we thought there was a higher risk we went and we built a contingency plan. One of those very key examples, which we've spoken about publicly, is we did a test node that enabled us to test out and develop our PowerVia technology. So to be ready for 20A and 18A, and do that independent of Ribbon-FET, I think that’s an example of something that has done really well.
IC: Not a lot has been said about that but I think it’s been really important. Now there is some confusion about whether the jump from 4 to 3, and the jump from 20A to 18A is what we’d traditionally call ‘a node jump’. How do you define that and measure that from your side of things?
AK: In terms of when I stood up in July of '21 and gave our roadmap, I said that the performance per watt improvement was going to be about 18% going from Intel 4 to Intel 3. We're working on delivering that. Same when we said about 10% going to 20A from Intel 3, and another 10% going to 18A. We've been working on delivering that.
IC: All going well?
AK: All going well.
IC: One of the biggest questions I get from the investment community is "Is Intel still on track for 18A?"
AK: The answer is yes, we are.
IC: We saw the v0.9 of the PDK being announced, and that’s all in the hands of everybody making chips?
AK: In the next couple of weeks it'll be in their hands, yes.
IC: So not long until V1.0 I guess?
AK: Well that'll come on the standard schedule that the industry normally delivers it on. That’s been a key thing we're working on - ensuring our schedules are lined up with what you'd typically see from a foundry.
IC: So what has the feedback been from the foundry customers?
AK: In recent months, the feedback from those that have been engaged with us on 18A has been positive.
IC: Ready to put in orders soon, maybe?
AK: Well you've heard Pat publicly announce we had a pre-payment, so I think that’s my answer for that one!
IC: One of the questions I had yesterday from an investor is "what does that pre-payment enable from a TD perspective?"
AK: I think one of the things it enables is the overall investment on any unique aspect that the customer wishes from that. It also enables us to start getting going on the capacity.
IC: So, based on that answer - we've seen other foundries develop customised nodes, or at least variants of their traditional processes for specific customers, those willing to pay. Can you speak to what Intel's direction or roadmap is for that? If a customer comes along and wants a specialised process, how that comes about?
AK: I think it very much depends on the particular ask, but where possible, we strive to accommodate the customer.
IC: So the idea is "if it’s possible, we can supply"?
AK: If it’s possible, and it also depends on the amount of volume a customer wants. If it’s a very low volume and at a high development cost, you'd have a long conversation with the customer before you do that. But if it makes a lot of sense, then absolutely - my team would be working it.
IC: This is where I ask the question of ‘well how much volume does a $1bn pre-payment get you’ but I know you won’t answer that! [chuckles]
AK: [chuckling] No, I won’t!
IC: Another one of the questions I get is about Intel's manufacturing for different types of IP, and the process research required. For example, developing GPUs and AI chips on a TD side is perhaps a little bit different from developing a process for CPUs. Can you talk about Intel's developments therein?
AK: One of the things that we're doing right now is working on our 18A foundry node that is our leading edge, and working on our definition of Intel next, and Intel next beyond that, and we're having internal discussions right now in terms of which aspects of the market are we going to do. So if you ask me this question in about a quarter's time, when we've worked our way through that, I'll give you a better answer.
IC: Let me ask it in a slightly different way. Typically with a GPU or AI chip, the thermal density of the compute units is a lot higher, and that obviously has to be built into how you do the manufacturing. Intel has historically been a very CPU focused company, and a lot of the process nodes have been CPU focused. So pivoting a GPU first, or GPU like mind set for a process node - how does that change things?
AK: One of the things that we need to work through as we move to PowerVia. There are a lot of people in the industry who had concerns about power from the backside and managing thermals. We, over the last 2 years / 3 years, have done a lot of innovations on how to manage the thermals using PowerVia, then we have to overlay how we apply that into the GPU world. So I think that’s an engineering activity that we need to work through.
IC: One of the questions is ‘one of the promises of PowerVia, potentially, given what we've seen other people do in the market and some of the research, is vastly increased frequency of the chips’. Is there anything to add to the other benefits of PowerVia?
AK: Well it gives increased speed, but also there is potential down the road in terms of how to utilise it in advanced packaging. I think we've just seen the first early steps of it, because now you have metal on the backside of your die.
IC: So you say we’re seeing early steps, so I’m assuming this has been in pathfinding for a while?
AK: Yes, it has.
IC: Given everything with TD, and the process node development, and being here at Fab 34, what should be the key takeaway that people watching this interview should come away with regards to the future of Intel's process node roadmap?
AK: In terms of our 5 nodes in 4 years, we are progressing very well. I will go and say Intel 7 is done. Intel 4, I'm now calling it done, Intel 3 is coming rapidly and progressing very well. 20A and 18A is coming, progressing well and progressing to what we're committed to.
IC: What if Pat turns around and says "I want another 5 nodes in 4 years"?
AK: He better talk to me when I've done this one!
IC: Good answer, good answer! Thank you so much Ann for your time, and good luck with the opening of Fab 34.
More Than Moore is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.