Using Data To Improve Yield

Semiconductor manufacturers are always looking for an edge to improve operating efficiency and to increase yields on chip lots. For some, the answers include big data analytics, as well as technology to move that data around more quickly.

Chipmakers, board assemblers, and related businesses are turning to the Internet of Things, especially Industrial IoT technology, to reduce defects and grow their bottom lines. They also are utilizing artificial intelligence and predictive maintenance on factory floors.

This may seem like an obvious next step, but the impact is farther reaching than collecting and analyzing data would seem to indicate. By figuring out what goes wrong and being able to actually trace that back to the source of the problem, efficiency can improve by double digits in operations that are already considered highly efficient. Moreover, it can be used to help reshape manufacturing processes and flows, breaking down barriers and silos that made sense in the past but have since become a hindrance.

“This is what the IIoT or Industry 4.0 is supposed to do — improve things and factor in the manufacturing,” said David Park, vice president of worldwide marketing for Optimal+. “Now it’s not just theoretical. It’s not just a promise. We’re actually able to show that we can do it. And customers are benefiting from it, especially for areas like autonomous vehicles, where quality has to got to get a lot better.”

Park noted that data for chip failures needs to be correlated at the OEM level. “When you share the data, this is the kind of improvement you can see. And this only happens when you share data. Otherwise, you’ve reached a plateau.”

Many companies are reaching the same conclusions. Data is at the center of a push toward smart manufacturing (also known as Industry 4.0 in Germany).

“A big focus for us right now is how IoT and IIoT are really changing the world,” said Roberta Gamble, a partner and vice president at Frost & Sullivan. “From my own experience looking at everything from discrete manufacturing to electric power to oil and gas and other industries, manufacturing really is among the more cutting-edge industries. It’s really amazing what is being done right now because of sensors and data and analytics, and being able to get into AI and prediction across these areas. Our own research has shown that if we really took advantage of IIoT, we would be about a $9 trillion market globally, and almost half of that, about $4 trillion, would be just in the manufacturing space. There’s a huge amount of opportunity that we’re seeing.”

Fig. 1: Smart manufacturing. Source: Cisco

Sharing data across these operations is easier said than done, however. Wayne Allan, Micron’s senior vice president of global manufacturing, noted his company has 13 plants around the world, with about 20,000 employees.

“That distributed network really drives a lot of challenges when it comes to learning, and making sure that we’re sharing those lessons across sites, and using a common platform and common data,” Allan said. “That has really driven us to embrace big data and Industry 4.0. In addition to that, we’re in a very challenging industry, as you know, as we’re constantly driving more and more bits per wafer, or shrinking our feature sizes on all of our devices. It’s becoming more and more challenging to do so in a cost-effective way. We’re adding more and more processes, so cycle times will typically get longer, capital equipment must constantly be up. There’s a tremendous amount of pressure on cost and efficiency. There are a lot of higher expectations, too, with some of the new applications that we’re going into where we’re filling system solutions, and working with automotive customers who expect these are life systems, if you will. And we’re talking about memory that goes into automobiles with driver assistance, so we’ve got to make sure that the quality is at exceptionally high levels.”

This puts a focus on getting things right during manufacturing. “Process technology is always getting a lot of focus on innovation, as well as design,” he noted. “In the last several years, the doors opened wide for us in manufacturing. We’ve always had this opportunity with the kind of data we have access to now and computing power that we have. Smart manufacturing has become a tremendous opportunity for us within our manufacturing organization. We started adding even more value when it comes to innovation in how we take some of these challenges we have and overcome them.”

In 2013, Micron started aggregating data from production tools and sensors into a central database. Acoustic technology was applied to process chambers to diagnose equipment conditions. Vibration data and visual detection are also vital.

“One of the things we’re able to do exceptionally well right now is tool management outside the fab, where we have a remote operations center now, because we can connect into the tools,” Allan said. “They can monitor the fab and change the schedule, do different things to manage the facility right there remotely. And then, of course, have a chance to interact more effectively, because they’re all together, rather than scattered across a large cleanroom.”

He noted that Micron has implemented advanced analytics and visualization in employing data from the wafer fabrication facilities. Machine learning technology can respond with predictive maintenance before a tool goes down. The company has had a data science team since 2013, helping drive a 10% improvement in throughput and a faster yield maturity by 25% over three years.

“IT plays a big role,” Allan added. “We’ve only just started.”

Tim Long, Micron’s director of enterprise data science, said the chip company has installed about 250 nodes in its manufacturing sites and has collected 13 petabytes of data in massively parallel processing data warehouses, using Structured Query Language (SQL) technology. “We like to democratize data,” he added, while always accounting for proper cybersecurity.

Equipment vendors are helping to facilitate this data gathering. Tim Archer, LAM Research‘s chief operating officer, said it was “an exciting time to be in etch and deposition,” with the big density increase in memory chips. In addition to predictive maintenance, he noted that Lam is working on self-maintaining tools, using robotics inside the system to order a new part and install it.

“We’re using hundreds of sensors to generate gigabytes of data,” said Archer. “We’re not unique in this. The challenge is how to transmit data, analyze data, and develop actionable intelligence. That has an outside impact on the memory business.”

Maciej Kranz, vice president of strategic innovations at Cisco Systems, said his company has been focused on manufacturing and IoT for the last decade. “We see four broad sets of big use cases in manufacturing,” he noted. One is connecting all operations. Remote monitoring of operations is a cost-saving opportunity. Predictive maintenance is another valuable asset.

“Our customers want to co-develop with us,” he said. “IoT, or IIoT, or Industry 4.0, whatever you call it – it’s all about data. We are seeing increasingly distributed clouds, fog computing, we call it. In two years, every vehicle will be connected.”

That will generate enormous amounts of data. Gian Yi-Hsen, regional president of the Americas for the Singapore Economic Development Board, said the city-state has been pursuing manufacturing development for five decades. “Electronics is a big portion of that,” he noted, estimating that 10% of all chips are fabricated in Singapore. Manpower, the technical talent to run all those factories and fabs in Singapore, is a constant issue for the country.

Moving data faster
Utilizing data is only part of the challenge inside of manufacturing organizations, though. It’s also about sharing that data, and that requires an understanding of what is required to handle the growing volumes of ones and zeroes.

For most semiconductor manufacturing operations, this is very manageable with equipment that is found in enterprise data centers, rather than the giant cloud operations of a Google, Facebook or Amazon or even manufacturing-specific devices.

“They’re taking our 24- and 48-port switches, just our standard enterprise switching, and they’re embedding those within their controller systems driving their semiconductor manufacturing equipment,” said Jeff Horne, director of strategic accounts for D-Link Systems. “What’s driving that is cost. You can get a really fancy German switch that does some nice things. But in the environment they’re using, all they need is a reliable switch with a good lifecycle, good performance. They take these switches, which have similar chipsets to what those really expensive ones do for a manufacturing environment, and they put these in a manufacturing environment. But not on the clean side—on the other side of the wall where they’re not exposed to any elements other than what they would normally be exposed to in an office. For about a third the cost, they get a switch that has a lifetime warranty. We’re selling thousands of these things. We’re embedding typical enterprise-type products into these environments where they fit.”

D-Link isn’t alone in observing this trend. Marvell said that in engineering its eighth generation of Ethernet switch silicon, understanding Industrial IoT and Automotive’s evolving needs for secure networking was vital. Within those switches is a “deep packet inspection engine” block, said Andrew Klaus, director of business development and architecture. “That is a programmable, hard-wired block that can inspect the packets coming in and out of the switch. Each packet can be inspected. Deep into a packet is usually just the payload. The beginning, that first 30 to 100 packets tells the network, ‘This is where I am coming from, and this is where I want to set my data.’ We can look inside of the header. We can go into level 3 to 5 to see what kind of packet it is and make sure the packets coming in are in the right format. And that is the key. For each of the switch’s eight ports, you can do blacklisting, and whitelisting. You can say ‘I only want this kind of Ethernet traffic going through the system.’ People are just starting to use these tools in the industry.”

There is a security element here, as well, said Klaus. “Because you don’t want attacks, snooping, you don’t want your network flooded with requests. Hackers can come up with all kinds of ways to do damage. So a step further is whitelisting. At home, anyone can plug into a router, it has to be very open. In a car architecture or in an industrial setting, you have a pre-defined architecture. You know what kind of data packets you want on the work. You can create hundreds of these rules and define many, many kind of packet types you will allow.”

Fig. 2: Marvell’s Ethernet switch architecture. Source: Marvell

Quantifying improvements
The entire electronics industry is looking at how to gather, move and utilize data.

“This isn’t just happening at the leading edge of manufacturing,” said Tom Salmon, vice president of collaborative technology platforms at SEMI. “We’re seeing it through all segments. Data is one of the big focal points of the Smart Manufacturing Advisory Council.”

It’s also one of the key elements in improving efficiency further up the food chain. So while yield may be high for individual components, there are huge opportunities for utilizing data in the future for other aspects of manufacturing.

“If you say you have 95% yield for a wafer, you probably can’t do much better,” said Optimal+’s Park. “But if you look at electronics, it’s a different story. Electronics can be reworked. So one company we know was spending $100 to de-bond a device, and the margins were high enough to cover that. That isn’t a yield problem. But by gathering the right data and using it effectively, they could have identified the problem that caused them to do the de-bonding. Efficiency in that case is harder to quantify.”

But the opportunity for improving efficiency is enormous by anyone’s calculation, which is why there is so much focus on smart manufacturing.

—Gale Morrison contributed to this report.

Related Stories
Smart Manufacturing Gains Momentum
Problems remain for legacy infrastructure, but adoption will continue to grow as gaps are identified and plugged.
Tracking Down Errors With Data
Optimal Plus’ CTO looks at how data can be deployed to improve quality and yield, and to find out what went wrong.
Data Leakage And The IIoT
Connecting industrial equipment to the Internet offers big improvements in uptime and efficiency, but it adds security issues.
The Rising Value Of Data
Race begins to figure out what else can be done with data. But not all data is useful, and some of it is faulty.