How ASML revived Moore's Law and remade chipmaking

Key Takeaways

[Music] Welcome to the transformation debrief. I'm Chris Hood, a digital strategist at Google Cloud and your host. In each episode, we aim to stir visionary thinking and share unexpected insights on transformation initiatives and lessons learned along the way. ASML is an innovation leader in the semiconductor industry. They provide chip makers with everything they need from hardware, software, and services to mass-roduce patterns on silicon through lithography. Founded in 1984, ASML now has over 32,000 employees at over 60 locations worldwide. In 2021, they closed the year with net sales of $18.5 billion. To dive deeper into this incredible, innovative company, I am joined by a co-host from Google and a wonderful guest from ASML. would you both mind introducing yourselves and sharing a little bit about yourself? Thanks Chris. Hi everyone. Uh my name is Simon Floyd and I'm director of manufacturing industry strategy and solutions at Google Cloud and my team is responsible for the overall strategy for executing within the industry and aligning our Google products to our um customer needs and customer problems. A real pleasure to be with you and um over to you Arno. Thank you Simon. My name is Arno Ibo. I'm a product cluster manager at ASML. I lead the delivery of solutions that maximize the the utilizations of ASML machines by chip manufacturers and minimize the resources needed to service those machines. In the intro, I shared just a little bit about ASML, but for the audience, could you share a little bit more about exactly what ASML does? So simply put, ASML makes a machine that makes the chips. So companies like Intel, Samsung, they design the chips they want to make and basically a chip is broken down into a 3D structure. So it's a collection of layers potentially. Now we are reaching hundreds of those layers. So each layer needs to be processed and we provide the machine that turn those different layers into a physical structure on silicon. So I know maybe a quick question for you just about um about your machines and about your work um and in in particular how are you working with Google Cloud today and what was the the overall opportunity that you saw and uh how has the work been going today? What we notice is as we transition also in the industry slight to link the two things together. You see that everything was very much hardware driven because to shrink we had to make better sensors, better actuators, better light sources. But what you see now happening is that those machines are good enough and you want especially with not the chip shortage that we all have to face. You want to optimize the utilization of those machines and the optimization of those machines is something that you typically drive with software and it requires massive data processing. So initially we actually started working with Google cloud I think about four years ago where we realized that our local environment the environment we had on prem was not sufficient to drive the analysis we wanted to do the product we wanted to build. So it actually started with one small team delivering one product as a pilot. Uh and we noticed quickly after six months that uh the performance of the team improved 10 times. So you may notice that using the cloud not only made us faster at building an ML model or machine learning model or even a statistical or physical model. the overall cycle of application development improved significantly because what is very important for us about the cloud because we move the whole environment the whole development environment there it's not just about building good models it's about building good solutions for customers. Did you see an increased demand from your consumers for that type of service or or was this solely something internally that you said we need to optimize how we're producing software and hardware and we need to look at the cloud as a as a solution. So the the main driver was time to market. So we needed to be faster at releasing our solutions and also improving the quality of our solutions. I expect that in the coming two to three years there will be a slow shift towards the cloud for the non-critical um processes and what I mean by non-critical pro processes if you look at the chip design there are processes that actually optimize the behavior of the machines and they do really runto- run optimizations those processes are so critical that they will I think never rely on an internet connection but other processes that are running basically outside of those production critical processes like diagnostic processes, monitoring prop pro processes that actually are very data intensive. Those I think will slowly move to the cloud because this is the only way you can scale uh those processes for the new needs of of the the production uh process. I know I'm very curious um just about the people aspect of um the work that you have done. Did you find that um you know bringing these types of solutions to them was um something that they could easily adopt? there were two profiles. So I think the engineers who had to deliver the solutions were super excited to on board. Uh people that were a bit more reluctant were um I think mostly on on the receiving end of those products because on the one hand there is of course this big concern around IP. IP in semiconductor is very critical. What does it mean? It means that you can go with your own chip design and they can make it for you. It means that the design that you uh ask Intel to to basically turn into a physical chip could actually be competing with Intel CPUs. So there they have to be very very very big walls and boundaries to protect the IP that is running in the tools and the equipment. So now you say we're not going to do the calculation, the modeling, all our IP is not going to be running on prem is going to be running in the cloud maintained by someone that we don't really know. I I think that's actually a really key point about transformation is um I think I I would call that building trust that everybody un understands how something works. Um you know to to the degree where they can feel very confident in it and they have that sort of that develop their own trust model around it. Um did you find that there were other sort of human factors that that played into this as well? Maybe not so much security or sort of the the the confidentiality part I think we understand but were there other attributes there as well which you had to work through as part of your transformation journey? I think one of the elements very often associated with the cloud is AI. uh because you see that in many cases the cloud is promoted as the enabler for AI and and you have some part of the people who just say AI is hype so I don't even want to hear about it and those are more skeptical and even when your solution let's say you're using machine learning comes with extremely high levels of accuracy they still don't want to use it because they don't trust the model and what you see today even the best methods to provide model explanability are still not good enough, not reliable enough, not trustworthy enough to be able to say yes, this is this module and this is the drift of this module that is causing this effect on the chip. And until we can make this link very clear, very explicit between the results of a prediction and all the causal relationship leading to that recommendation, people will just not trust it. I know. I'm I'm kind of curious. Um, you know, in in my role as an example, I I have the privilege of meeting with many executives and they always ask the question, why am I doing this? What's been your your experience there working with your own executive team and explaining to them why you wanted to take this particular approach and how at the end of the day, how's it really been measured your overall success? I think first it started by a leap of faith because uh they like SML is where it is today because it tried pretty much everything until it found the right way to do so. So there there is this kind of trial and error modes that is part of how we do things because it's so complex that you have to try and learn and try and learn and try and learn until you figure out the right solution. So so Chris um I read that not so long ago you were selling popcorn at the local movie theater. So uh a very important part of a movie is is the the story. So in a digital transformation, how important is the story that we tell? I think that goes back to even what you introduced yourself with, right? is that story line and and the ability to communicate what that story is or what that message is in the right way so that you can paint the picture of your transformation and be successful and bring others along for that journey. I think all of that is absolutely critical and really this is about understanding and and you touched on it a few times what consumers are looking for. What are your customers expectations and then being able to translate that value that you're talking about directly back to those customers and paint this picture. We are transforming for this reason. we are transforming for our customers so that they can receive this value from us. With that, thank you so much, Arnon. Appreciate your time and insights. It's been a fabulous conversation with you. My pleasure. And thank you, Simon, for being my co-host today. I appreciate it, Chris. Great to great to work with you again, Anna. And if you would like to learn more about ASML, we have two channels for you to explore. At cloud.google.com/transform, we dive deeper into this transformation journey and you can tune into that digital show to hear the extended podcast version of this conversation. And don't forget to hit subscribe and join us again for our visionary thinking and lessons learned on the next episode of the transformation debrief.