The Future of Chip Making: Using AI to Minimize Testing and Maximize Throughputby@alexlash
224 reads

The Future of Chip Making: Using AI to Minimize Testing and Maximize Throughput

by Alex LashkovJanuary 30th, 2024
Read on Terminal Reader
Read this story w/o Javascript
tldt arrow

Too Long; Didn't Read

David Meyer, Lynceus co-founder, discusses revolutionizing semiconductor manufacturing with AI, focusing on process automation and output optimization. Despite initial industry skepticism, Lynceus demonstrates significant efficiency gains. Meyer's background in scaling operations aids their growth amidst a challenging fundraising and competitive environment. Lynceus eyes expansion into other industries, emphasizing the importance of persistence and adaptability for success.
featured image - The Future of Chip Making: Using AI to Minimize Testing and Maximize Throughput
Alex Lashkov HackerNoon profile picture

The semiconductor industry, despite its innovative core, is traditionally known for its conservative approach, particularly in adopting new manufacturing processes and technologies. This cautiousness stems from the high stakes involved: even minor errors in chip production can lead to significant financial losses and delays in product launches. However, now this sector is undergoing a pivotal transformation.

My recent conversation with Lynceus's co-founder, David Meyer, revealed the startup’s innovative strategy: leveraging AI not as a tool, but as a transformative force to optimize processes and amplify output in chip production.

How did you come up with the idea of applying AI to semiconductor manufacturing? What was the initial goal and idea behind Lynceus?

My co-founder, Guglielmo Montone, spent 15 years as an AI researcher where he was looking at different techniques that enable the training of AI models on very dynamic and unstable data—that is, something that you find a lot in manufacturing.

We started the company by applying this to semiconductor manufacturing with the idea to automate all processes from the beginning to the end. This means that you're going to have an algorithm to continuously calculate the optimal parameters to put in the machine in order to obtain the best quality.

But initially, this was very hard to sell because you're telling manufacturers that you have a black box to run their machines. We understood that if you're going to change the parameters of a machine, you need to be able to anticipate the specific impact on the quality of the output. And so there is a quality prediction element in the core of our technology.

How do you sell your product to manufacturers now? What results can you guarantee?

We are showing the specific tangible impact that your tool can have if implemented. And at the same time, you have to show them how this is going to scale because if we're just going to deploy one model somewhere, it's not very interesting.

What we're doing is establishing a new way of managing the performance and the productivity of this type of factories, which is based on simultaneous optimization. This includes products’ quality, volume and costs - the three main KPIs of manufacturing. Traditionally, manufacturers can only optimize one of those at a time.

Our models also generate data that gives real-time visibility on what is happening inside the machine. Manufacturers don’t have it today, and it can feed a number of other systems in the factory. Maintenance scheduling or automation systems can both benefit from more granular and real-time data on how a process behaves.

This sounds like it can also be used outside the semiconductor industry. Do you have plans to scale your business in another direction?

Absolutely. Initially, we started Lynceus to optimize the production of vaccines, where you have to grow cells in a bioreactors, and you don't know how many cells you're growing. If there’s not enough, you have to throw everything away, so we have built an algorithm to predict if the cell culture is going to be successful or not. It worked quite well, but then the traction picked up on the semiconductor side of things.

Another market could be all the industries similar to semiconductor manufacturing but less sophisticated. LEDs, building the screens for the iPad, TVs, etc. Solar panels could also be a good one.

There's also battery manufacturing, where we would like to expand on in the future, but we haven’t investigated it so far. 30% to 35% of what they produce is not good enough, and there's a lot of complex chemical reactions to make sure that their metal has the right properties, which can be optimized with data.

How does your experience in managing transportation companies like Uber and then Circ help you grow Lynceus?

Even though I do not come from semiconductor manufacturing, I have experience scaling operations. I know how to go from zero to one, start a project, make people work, structure, set objectives, and make sure that we hit them. And I think not coming from semiconductor manufacturing is actually something that helped me in this venture because it's a closed and conservative world, and I wanted to disrupt it.

When we started, everyone told us that what we wanted to do was impossible. And so if we listened to people who've been in the industry for 20 years, we wouldn't even have tried and started what we did. I think arriving with another perspective and just continuously trying, helped us to get to where we are.

What are some significant milestones you've achieved so far?

I think the main achievement we have is that we deployed the first AI model in production in our customer’s fabs, and broke into the semiconductor industry, which is notoriously conservative, in three years. I think the challenge here is just the sheer kind of ramp-up that you need to have in order to be credible in any discussion with PhDs who work on the same machine for 20 years.

You have raised $10M for Lynceus. What would you define as your major drivers to success? Was it hard to talk to investors given that you're working in a complex industry?

Yeah, it is hard to find the right investors. We work with big semiconductor manufacturers, so we end up having discussions with a lot of investors. But not many of them eventually have an appetite for this type of product.

When we started raising, it was the beginning of COVID. We talked to around 85 investors and ended up with three offers. The first thing for us is to make them understand what we do in a good level of detail. Investors see companies every day so they tend to have these kinds of boxes like this predictive maintenance or this is AI for manufacturing - to see whether we fit there or not. But we need to get deeper and explain exactly what we do and the ROI of our solution.

What's the difference in the fundraising process right now given the AI hype?

COVID was hard, especially because it was the first time I was doing a fundraising myself. So it was much harder to build any kind of connection with investors at this time. And I think it's an important part of the journey. We have investors who liked us and ended up working with us.

The appetite for risk is now lower than before. Now it's all about these questions: How much revenue do you make, are you profitable, can you guarantee that the market is huge and there's not going to be any legal risk and competition?

You need to understand which firms are interested and make sure that you screen the time to  spend with each of them.

You mentioned the competition. What does the landscape look like in the AI industry for semiconductors?

The good thing about manufacturing is that you have an infinite amount of use cases once you contract a big factory. We're starting to see some competitors emerging, but those startups are all spin-offs from big players existing in the industry.

So you might see a company in which all of the execs come from a major chip manufacturing company, and this company invested in their series A. Or there’s another one that is half acquired by a major manufacturer of machines for those factories. I think being independent is our great advantage.

How do you see the current landscape of AI in the manufacturing sector evolving, and what trends do you find the most impactful?

As we started having discussions with 15-20 major semiconductor manufacturers, the first takeaway is that everyone takes our business seriously. We haven’t seen a single major company that has never tried to implement AI in some form so far.

We are also reaching maturity in terms of the data infrastructure that would support the deployment of this kind of solution. This required a lot of investment over a long period of time. There’s finally something that produces payback for this investment now.

The third one is that we're seeing companies trying to do this themselves and then coming back from this decision because they don't want to do it on their own anymore. That's how we gained one of our first customers.

This solution can be applied across different factories, machines, and teams in the same factory that needs to be implemented and maintained in very close collaboration with IT and their quality team, because every new tool or new production process you introduce needs to be vetted there. To execute this, you need to have a transparent and efficient communication and sharing of information between those teams who work a little bit in silos. Companies that try to do this on their own often have difficulty breaking those silos, but we can provide a solution for that.

What do you think are the most useful qualities for an entrepreneur in this space? What personal skills do you believe helped you to grow that fast and to that scale?

The first one is to be stubborn and tenacious. We are talking about people who have a day job, where they need to run a factory with 3,000 people. So if they get on a call about something they're going to be skeptical of, you have to keep trying and also be open to understand what works and what doesn’t and adjust your message based on that.

Another one is having an ability to structure things quickly, even though you're working in a very unstructured environment. So maybe you just developed one part of your product and you don't exactly know how it works, but you need to be able to go in front of a customer and present a structured deployment process and say how it’s going to work and what the process is.

The last one is most relevant to working at Lynceus. Whether in consulting, or at Uber and Circ, I worked with like-minded people - very operational with the “let's go and tackle the project” approach. Here, working at Lynceus, I need a different set of skills.

My co-founder is an AI researcher who's worked in a lab his whole life on a different type of scale and timelines and everything. And so it's crucial to be able to open up and listen to people and accept other ways of working, but at the same time be conscious of the strengths and the skills that come from your experience. It's keeping the best of what you learned while being able to adapt it and make sure it fits with the other types of skills that you need.