IIoT Comes To Chip Manufacturing

Nicholas Ward, director of marketing for the services group at Applied Materials, sat down with Semiconductor Engineering to talk about how data needs to be shared in semiconductor manufacturing and why it’s so slow to happen in this industry. What follows are excerpts of that conversation.

SE: Where are we with the IoT and the IIoT?

Ward: According to Gartner, the IoT is right at the top of their hype cycle. That doesn’t mean we can discount it, though. A lot people are putting a lot of money and commitment into the IoT. It’s real and things are going to change. But Gartner’s adoption cycle says it’ll take between 5 and 10 years before we see the actual benefits, the so-called “Plateau of Productivity.” However, we can’t simply wait for it, especially as semiconductor manufacturing is extremely difficult. It’s not like consumer or less-complex industrial scenarios. Ours is a highly complex manufacturing environment where just collecting data is not going to be enough. Understanding the data, the context of the data and when the data can be used is a big deal. There’s a lot more we need to understand to get the benefits.

SE: So where does Applied Materials get involved with the IIoT? It’s a very broad spectrum.

Ward: The sensors on the tools are already connected. We already have a lot of information coming in from the IC factory. We’ve had connected tools since 300mm, which was a huge changeover for that. We went to full automation. The data we’re getting off those tools is going into databases, and we’re seeing the size of those databases increase dramatically node to node.

SE: We’re into terabytes now, right?

Ward: We’re approaching petabytes. And how we deal with this volume is going to be a challenge. We’re already pulling data into databases using Big Data techniques supported by equipment engineering systems and we’re figuring out how to use that in a Hadoop environment. The big question really comes when we connect lots of other things—people, knowledge sets, actual production processes and what we know about our tools. Bringing these viewpoints together will provide the opportunity for new insights.

SE: This is the next mashup for you, right? Functions connecting with other functions?

Ward: Yes. It’s getting people’s knowledge into systems so that we can automate it. Right now we pull in data from our equipment with no problem, but sharing data between Applied Materials and its customers is really challenging.

SE: So you’re thinking about it from the standpoint of your tools communicating with other tools that may have been developed in a proprietary way?

Ward: That’s correct. It’s about how we can improve manufacturing. What’s really crucial is having a complex manufacturing environment that is fully instrumented and the people and their knowledge are accessible. That’s where we can truly start getting insights into how to improve manufacturing. We’re not necessarily as advanced as some other industries. Even a vending machine down the hall has a level of sophistication. The person who comes to service it arrives with an iPad. He or she can know about every candy bar that went out of it, whether there were any error codes on that machine. It’s just a candy machine. Our tools can be connected like that, but there is a barrier around them because of concerns about sharing data.

SE: There’s been talk about this with Industry 4.0. How far along is that?

Ward: We think of it as Manufacturing 4.0, and my view is that we absolutely need to get there. All industries are going to be moving toward it. But our industry needs to develop standards for protecting the data that’s shared. We must fix this over the next 5 to 10 years. The first challenge is to understand all of the data that’s in the factory and not just the data coming off the tools. This involves process recipes, design intent for the device being manufactured and how that impacts the processes that can be tuned. We need a semantic map. Understanding that data and why it’s necessary will be essential if we’re going to have automated protection and analysis of data.

SE: How does this play out in the real world?

Ward: A good example is to track how much gas is going into a chamber. We can do that now. But is that gas flowing because the tool is in idle, or because process A, B or C is running, or is it an engineering experiment? It’s necessary to understand the full context for the data to be valuable, and a map is required to do that. If you ask the process engineer on the tool, the answer may be they’re running a test wafer. But that’s not enough information to enable big data analysis to work. Only information in computerized form can identify exactly what’s happening. The semiconductor fab has many computer systems and many, many sources of information. We’re bringing it all together so you can understand everything there is to this gas flow.

SE: What are you hoping to get out of this?

Ward: We’re trying to optimize manufacturing and provide customers with exactly the solutions they need. From a maintenance perspective, we are focused on being predictive. Our supply chain needs to be aware of what’s going on in a fab so we can route parts in advance of failure as part of a service contract. It’s all about efficiency.

SE: How much do you have to work with your customers and your competitors?

Ward: It’s about bringing our views and insights together with those of our customers. We need to work with them. But working on a customer-by-customer basis will only get us so far. We need to work with multiple customers in some kind of collaborative sense.

SE: That’s a big challenge, isn’t it?

Ward: Yes, it’s a very big challenge, but we all want to collaborate. Everyone wants to benchmark their data against the other company’s data, but they’re reluctant to be first to share. They’re very cautious. We need Internet of Things types of standards. Is anyone else looking at semantic models? Can we work with that? And then there’s the whole security aspect.

SE: Security is a big worry for the IoT and the IIoT.

Ward: Yes, and we spend a lot of time talking about this with customers. We have to put walls in place. There are best-known practices for securing and sharing company information, but most of the time it’s still a very close-to-the-vest type of interaction. There are lots of companies working on generic security access, among other things. However, security in terms of data context—the meaning of the different categories of data—is another matter. For example, it’s okay to share information about gas flows from this tool when I’m running Applied Materials’ formulation recipes, but it’s not okay to share gas flow information when the tool is running proprietary process recipes or process recipes on production wafers versus test wafers. For that kind of data categorization or security there is nothing at the moment. That’s a layer of security that I haven’t seen discussed much.

SE: Synopsys co-CEO Aart de Geus recently compared the IoT to putting kitchen windows in the bank.

Ward: That’s appropriate. It’s not just about common data. We need a lot more than that. Security has to be everywhere. If you have my data, I have to trust you and the benefits of sharing this data need to be greater than the risks.

SE: So you have to understand how to slice up all of this data, right?

Ward: Yes. We need a way to secure data and to understand data. That’s a semantic map, and then you need a passport associated with that data to limit how far certain data will go. It can’t be allowed to go any further than what manufacturers consider ‘safe’. To my knowledge, I haven’t heard any discussion about limiting data ‘travel’ in any of the industry forums. Without that kind of safety, progress on the collaboration front is going to be slow. At Applied, the training we go through in terms of how to respect and protect data is incredibly intense. We’re trusted to a high level by customers, and that requires stringent processes to be in place.

SE: But a lot of this is company-specific data, right?

Ward: Exactly right and we’re dealing with proprietary protocols and proprietary approaches to data. As an industry, we need to develop a set of standards for protecting the shared data. Being able to integrate to certain systems and tools will speed time to problem resolution and create greater cost savings for manufacturers.

SE: Assuming you work all of this out—and smart people in this industry can work anything out—what do you get that you didn’t have before?

Ward: Greater insight into how tools perform in complex volume production and how we can help customers use the tools better. If we get that insight faster, ahead of the problem instead of after the problem, then we become more predictive. We’ve been talking about predictive technologies for a long time at Applied, but we can only get so far with limited data sharing. We need to have secure data sharing and predictive technologies as a part of our everyday business. We also need to turn the knowledge about how to build, maintain and improve tools into data, so we can have it run automatically and then move onto the next level. It’s a bit like a design abstraction. There is a lot of stuff that can be dealt with automatically. Then you are able to bring higher-level thinking to higher-level problems. What do you want to do next? How do you want to develop the next process tool?

SE: A fab will also be able to determine whose tool is better for what?

Ward: Yes, the customer will be able to do that. And we won’t be able to see the data on our competitors’ tools. That will be blocked from us. But we will be able to see how our tool is operating at customer A versus how it’s operating at customer B. And we can use that to help each customer optimize the performance of their tools.

SE: That also helps determine where you invest R&D dollars, right?

Ward: Yes. We should be able to see very clearly where our tools are adding value and where we could add even more value. It gives us the opportunity to help our customers win—and to win more business, too.