Name: Quality & Test Under Pressure
Uploaded: 2025-10-17T19:00:00.000Z
Duration: 20 min 53 s

Clips
More from theCUBE + NYSE Wired: GSA Series - Women in Leadership

Leslie Averbeck

Senior Director, Quality

Cisco

Sarah McGowan

Sr. Director, Design Technology Enablement

GlobalFoundries

play_circle_outline Navigating Market Pressures: The Crucial Role of High-Quality Design and Testing in Chip and Integrated System Development

play_circle_outline Enhancing Quality Assurance and Process Optimization in Manufacturing through Data Feedback Loops and Advanced Process Control in Silicon Production

play_circle_outline GlobalFoundries’ investment in advanced packaging and hybrid bonding technologies for innovation.

play_circle_outline Challenges posed by latent failures like ESD damage in ASIC components identification.

play_circle_outline Predictive models developed from field data to accelerate root cause analysis of failures.

Info
Transcript

Quality & Test Under Pressure

Leslie Averbeck

Senior Director, Quality Cisco

Sarah McGowan

Sr. Director, Design Technology Enablement GlobalFoundries

In this GSA Series – Women in Leadership interview, theCUBE’s John Furrier sits down with Leslie Averbeck, director of quality at Cisco, and Sarah McGowan, senior director of design technology enablement at GlobalFoundries, to unpack quality and test under pressure in the AI infrastructure era. McGowan explains how GlobalFoundries drives “first-time-right” silicon through robust design collateral that mirrors the process, volume-ramp readiness and stable models for reuse. She details how process design kits (PDKs) pair SPICE modeling with OPC, where machine l... Read more

explore Keep Exploring

What are the current challenges and priorities in chip manufacturing and design quality at GlobalFoundries and Cisco? add

What is the role of feedback data in semiconductor manufacturing and design processes? add

What are the recent investment areas for GlobalFoundries, particularly regarding advanced packaging technologies? add

What are the major concerns regarding quality in the production and reliability of ASIC components? add

What is the approach being taken to utilize field data and genetic AI models to improve the identification and prediction of customer issues? add

bolt Powered by CUBE AI

Quality & Test Under Pressure

>> Welcome back everyone. I'm John Furrier, host of theCUBE here in our Palo Alto studios for a special series with theCUBE and the NYSE Wired programming community featuring the GSA series with future of semiconductors and hardware that's powering the infrastructure, of course, the software. This next generation infrastructure is the hottest area and there's a lot going on too. We've been featuring all the leaders. We've got two leaders here, Leslie Averbeck, director of quality at Cisco. Thanks for coming on theCUBE. Sarah McGowan, senior director of Design Technology Enablement at GlobalFoundries. Thanks for coming on.

Sarah McGowan

>> Thank you.

>> I mean, this is quality and test under pressure. So I'll start with you first. The pressure in this market right now is at an all-time high. The demand for systems, not servers, servers are now many servers connected together. New software stacks, the demand for chips, demand for integrated systems is in all-time... Just share your perspective and scope from your perspective how rapid is the pace of play right now in tech?

Sarah McGowan

>> Yeah, so at GlobalFoundries, we're manufacturing essential chips and we cover a wide range of technologies to apply in any kind of a system. So for sure, we're always trying to go faster to offer what the customer needs, what is that special sauce that we have to create into their designs to win in their marketplace.

>> And everything's got to be high-quality. Leslie, the shift at design quality, and Cisco's very much known for the high-quality product working all the time. What's your perspective on the shift and testing under pressure?

Leslie Averbeck

>> The shift in quality and to this next generation of products is just it's really kind of shaking up everything and helping us really think about how we design the next generation of systems. Cisco is very unique in that we have the full stack. With the Silicon One launch of ASICs and integrating those into our next generation of AI products, we really have a unique opportunity to really help respond to the market and be at the forefront of a lot of these new build-outs.

>> And the ramp confidence these days, people want to know supply schedules, so volume ramp confidence, Sarah, is something that you pay attention to. What does that mean? What does that mean for the industry?

Sarah McGowan

>> Yeah, so for a foundry for volume ramp is when we think of the product's been qualified, the customer design has been qualified and we're ramping the number of wafers. So we actually spend a lot of time, at least in my area of expertise, on making sure that the customer design is first time right so they can get as fast as possible to finish the qualification when they need to ramp. We're spending a lot of time upfront in creating the design collateral that mimics the silicon exactly. And they can have confidence in their design that when it hits the silicon, it's going to work right.

>> One of the things I'm observing from this series is that theme of doing the work upfront is also translating and crossing over into other areas in the industry, and a lot of people talk about machine learning and AI, how they're using it to help them get better, are doing a lot of that and then finding out that there are other ways once they're in doing that. What are some of the layers you guys are using AI machine learning to make the process go better?

Sarah McGowan

>> So in our process design kits, we're providing SPICE models. We're modeling the silicon process. We also provide OPC, which is what we use during the tape outflow to convert the design into the reticles. And so for long, that technology has been around a really long time and both technologies are using what we call compact models to try to mimic the physics and to speed up the simulation, basically have something that run fast. So I feel like that's the original creation of machine learning models is you make some assumptions and you want to make sure it's accurate enough and it can extrapolate. Where we see the advantages of using machine learning is how can we create accurate models as fast as possible, coming back to that topic about time to market and how can we deliver the right models to the customer as quickly as possible. We're also using gen AI similar to other software companies because the process design kit is like software. So things like a Copilot ways that we can generate the collateral as fast as possible.

>> And are you guys using the data feedback loops? Because this is again feeding back into the data, can you comment on that piece?

Sarah McGowan

>> Yeah, so semiconductors for a long, long time, we've always used advanced process control where we're feeding back data in the factory to adjust the recipes to keep the silicon on target. I would say in the design space, once we've frozen the process, have a qualified process with the design kit, actually designers don't want us to feed back data and to change the design models. They want it to be very stable. They want to reuse blocks from a previous design. Once it's committed to silicon, they know some circuit works, some IP works, some block works. So actually I would say the feedback that we do is just to make sure that the model that we have is constantly... The silicon is constantly matching that frozen model. So we have advanced test chips where we can quickly collect the data on a regular monitoring basis.

>> And that speeds up the process, and that's more quality and speed?

Sarah McGowan

>> That's more of a quality from the perspective of the process is running smoothly, day in day out, the customer's going to get what they need.

>> Leslie, talk about AI for quality signals because I know Cisco, we cover them deeply, and I know AI have been infused from the top down. Jeetu Patel is always talking about we're going to be AI first within Cisco, I'm sure there's practices that you guys have already had. Can you share how you guys are looking at using AI for some of the quality?

Leslie Averbeck

>> Yeah. So again, going back to the fact that we can see the full manufacturing process from end to end, we've already built a pipeline where we're getting data coming from the wafer build and then getting test data along and manufacturing process data along the way all the way to our system build. And once you have that set of data, you can really build a full end-to-end model to figure out and optimize your test process. And what we have found is in some cases there are very risky areas that we need to increase our testing so we can improve our quality out in the field with our customers. And in some cases we have opportunities to reduce our testing because either we were doing duplicate testing or we find that the correlation between what we're finding in the manufacturing process and what correlates to the field just isn't there. So in those cases we can just eliminate tests altogether. So what we're really seeing is that as a first step is that if we can tie this data end to end and build models on it, there's such an opportunity to really just improve our quality and optimize our operations, which is the panacea right there.

>> Yeah. And that's a great metric. I know Cisco's customer focused, how are they part of this customer requirements, how are they feeding into that quality decision0making or quality enhancement?

Leslie Averbeck

>> Yeah. This set of hyperscale customers has really been an interesting opportunity for us to look at how do we define our quality. So first of all, if you think about the overall systems that we're building, they are in the products, they are getting more complex, they're getting larger in form factor and heavier. And so we do everything from early engagements with FMEAs and defensive designs where we're trying to figure out how do we first of all make sure that if a product is going to be misused or mishandled, how do we prevent it from failing? And if we can't prevent it from failing, how do we make sure that it fails in a safe mode where we can mitigate the impact to our customer's network? And then we also want to get into how do we not have to take the entire unit out of the network, that is a huge undertaking for any customer, that we are really looking for how do we have repairability and serviceability in the field? Now, on top of that, what we have found with these large data center build-outs is that the installers, the conditions, some of these data centers are hot, they're humid. The beginning of the day is very different than the end of the day. So we've really had to bring in and think about human factors. So a lot of human factor testing has been built into our early design process, and that's starting to make a huge difference in our designs overall.

>> And that's something that AI could do. I mean, it's pick up some basic stuff that might not have been on the radar had surfaced.

Leslie Averbeck

>> Yeah. With this, we're starting to see not just how do we model that differently early on in our design process, but it's also a great example of the AI world and kind of the virtual world really interfacing with the physical world too. And how do you really make that work?

>> How do you work at foundries to handle the scale on the supply chain side for making sure the quality's there?

Leslie Averbeck

>> So for us... Is that to me or to Sarah?

>> Yes.

Leslie Averbeck

>> Okay.

>> You first, then we'll go next.

Leslie Averbeck

>> For foundries, we really go back in terms of guaranteeing the quality at our foundries and at our manufacturing sites up and down. The biggest way that we make sure that we don't have any surprises or any escapes is we really partner with our manufacturing partners to make sure that they have a solid quality management system. And all that's just from us making sure that they have proper statistical process control, that we have a trust but verify method that we're going in, they have the proper fixes in place. That's really our way that we're able to make sure that we're doing things right upfront.

>> Since I brought supply chain, that's a perfect one for you. You have customers. What's the perspective that because the scale is huge because the orders are coming in?

Sarah McGowan

>> Yeah, of course. And at GlobalFoundries we offer both kind of consumer grade products and automotive products. So depending on the class of product, they're going to have different quality standards that we're going to follow and meet. We get audited on a regular basis to answer those questions. I think from my team's perspective, what we're focused on is having a clear mechanism for change control and maintaining backward compatibility, demonstrating the benefit of the change, going through FMEA to know the impact of the change. And so once we get to that frozen process, it becomes... That's when we're locked in and we really think about what is the impact to the customer if we need to add something to it.

>> I'm sure you guys get this question a lot, but it's not obvious, but I'll ask it anyway. What's the biggest change? If you go back say 8 years, 10 years ago, that decade seems like yesterday, we've been doing theCUBE for 16 years, remember when big data was hot and servers, top of rack switches, the glory days. Certainly now that's all shifted. What's changed the most in your jobs in terms of is it speed, has it been mechanisms, has it been automation? What's been the big shift?

Sarah McGowan

>> I think in my space is on those things about automation and reusing what we did before. The things that we did 18 years ago, we still actually use them. And so those have to be robust and we need to be able to add to it to create new value for our customers, whether it's adding a new device type or a new type of memory, but we don't want to lose that history and we don't want to spend a lot of time kind of updating it, so to speak.

>> Leslie, how about you? What's the big change?

Leslie Averbeck

>> I think it's really the development cycles. The development cycles have gotten so fast, and not just so fast with kind of stable technology, but the technology is also getting more complex and the expectation that we need to deliver a quality process or a quality product right from the get-go and ramped very quickly. So a lot of times in this market and with these customers, we might have only had the unit, the new unit in our lab and powered on for a few months before we're expected to start shipping at volume for early field trials with our customers, and they're expecting that they will have good quality right out of the gate.

>> Yeah. And the community is interesting because you've got more co-design going on, a lot more ecosystem partners. Has the partner mix changed at all for you guys, Sarah, in GlobalFoundries?

Sarah McGowan

>> Mostly it's still the same. We use the kind of big three EDA vendors, so we're providing software across all of that. So that part hasn't changed so much, but certainly the end customers who's designing chips into our products, usually when we first introduce a node, it's maybe some special thing in your cellphone, and then as you move on, it's IoT, internet of things in your home or something like that. So the customer base is changing and the application.

>> And the design, Leslie, for you guys has been big. Any other changes on your side in terms of partners?

Leslie Averbeck

>> So pretty much in terms of partners, the partners have stayed the same. Our engagements with some of the partners have changed. So for instance, there might be third party distributors that we have our own relationships with, but then our customers also are building relationships with them and they're kind of this in-between sort of entity that we need to figure out how do we use our relationship with them to make sure that we're understanding what's happening to our products before they go out to the end customer.

>> Awesome. Sarah, I guess the question on investment that comes out, a lot of CapEx being thrown around. Where are you guys investing? You got a lot of gen AI, multi dye, a lot of things are going on. Where is GlobalFoundries investing?

Sarah McGowan

>> One of the new things that I'm excited about that we're working on is advanced packaging. So we're not playing in the space of the advanced GPU or HPM memories, but where can we play in that advanced packaging ecosystem in things like silicon interposers, silicon bridge, even providing differentiation for our customers by combining two wafers in maybe a different use case. We have processes where we actually have a single device that's on two different wafers and it's connected after hybrid bonding. So we're looking at how do we enable our customers to design that particular device, how do they follow the usual verification, how do they test it? So those hybrid bonding 3D packaging are very exciting for us right now.

>> Awesome. And Leslie, for you, what do you see as two to three years out key breakthroughs that you'd like to see on the silicon and the yields and things of that nature?

Leslie Averbeck

>> Yeah, I think that there's probably two big things. So first of all, I live in the world of quality, so I just think of I'm always looking around corners, waiting for the next bad thing to happen. That's kind of my life.

>> Hope for the best, plan for the worst.

Leslie Averbeck

>> Yeah, exactly. Yeah. But in terms of our ASICs, our biggest nightmare is this idea of the walking wounded. So these are essentially components that have maybe gotten some sort of ESD damage during the manufacturing process and you're not able to detect them. They're still functioning during the test process, but they have some level of damage that will essentially cause a latent failure and premature failure out in the field. And so one piece that I'm really looking for is how do we tie that test data that I was talking about end to end, and even using some of the field data to be able to model and understand when there's been small shifts in the performance of our products. So that's a huge piece. The other piece is looking at the field back end. So we already have deployed a genetic AI model where we are using field data, customer failure data to identify known issues that we have root caused in our processes. What we're really looking for is how do we use this data that's tied end to end and be able to almost predict or start to form hypotheses about the unknown failures that we don't quite have characterized yet. And the whole goal would be as customers are taking the time to get the units out of their network, which sometimes can take a full month or more, by the time the units arrive at Cisco in one of our labs, we already have a set of hypotheses that we're ready to test out and we can very quickly figure out what the root cause of an issue is for our customers.

>> That root cause acceleration is a huge thing. And then using the data, you're not in the vacuum staring at just data, you can see things like that drop in performance actually might be a signal, that comes back to the high quality signals.

Leslie Averbeck

>> Yeah, yeah. That's going to be game changing. Being able to tie that data end to end, build the models, get the signals from the field incorporated in that, that will really help us change our game in terms of quality management.

>> Well, it's great to have you guys on, experts in the field. I guess I'll throw a wild card question out there, curve ball. A lot of people are coming in and trying to learn the industry. I mean, if you go back five years ago, chips and design, and if you were inside the ropes, it was all great, GSA, the community is awesome. A lot of people are learning about the chips because they're seeing all the news. What's the one thing that you'd say about the industry that people might not know about that goes on that's super innovative? It's more of a high level question. I mean, -

Sarah McGowan

>> I'll let you go first.

>> She's like, "No, don't pick me." I'm sure you get questions all the time at parties and the family events, "What do you do again?"

Sarah McGowan

>> Yeah, yeah, yeah.

>> I mean-

Sarah McGowan

>> Yeah, no, I think what's maybe different about the semiconductor industry compared to other industries is that it's always changing, but it's kind of always the same. Whether you're on the path of chasing Moore's Law, okay, every two years, you're doing a different technology node, but you're kind of doing it the same way. If you're adding evolving technologies to capture more and more customers as they need semiconductors in their applications, you're looking at what is their specific problem and how do I have a device or a capability in my process that will ?

>> And so the game has change and still the same.

Sarah McGowan

>> Yeah, yeah.

>> I mean, pretty much, because the cadence of... I mean, you're talking about design cycles that are pretty significant.

Sarah McGowan

>> Right.

>> And they're getting shorter, but not that much shorter, not like agile coding. What about you, Leslie? What's the-

Leslie Averbeck

>> I think they interface with our customers and just watching our customers evolve their networks, evolve their data centers, evolve their requirements back to us, that is just fascinating to see how that affects our product portfolio, what we're designing and delivering to them. It's just a whole new world in some ways for us.

>> Yeah. And the innovation creates a better practice, eliminates a lot of inefficiencies, and gets into new signaling, new data, and obviously new customers. Well, thanks for coming on. I really appreciate it, being part of theCUBE.

Leslie Averbeck

>> Thank you.

>> Thank you so much. I'm John Furrier with theCUBE here, part of the GSA series. We're featuring the leaders in AI infrastructure really who are building the chips, the systems, the software that's powering the large scale AI build-out that's happening. It's just the early days. We're doing our best part to bring that to you. Thanks for watching.