A major theme at SEMICON West 2016 is Smart Manufacturing, a.k.a. Industry 4.0 and Industrial IoT (IIoT). One definition of smart manufacturing, said Tom Salmon, the SEMI vice president of collaborative technology platforms, is the use of production and sensor data with manufacturing technologies to enable adaptability in processing. It encompasses automation, data exchange, and the transfer of product design data and manufacturing state data.

SEMI estimates that by 2020 there will be about a billion IoT devices at work in manufacturing facilities. By 2020, global manufacturers will invest $70 billion in IoT solutions that year, compared with $29 billion in 2015.

Figure 1. What the future may look like for smart manufacturing in the semiconductor industry.

Currently, these devices are used largely to track factory assets, to consolidate control rooms, and to increase analytics functionality through predictive maintenance. The goal is that product design data and manufacturing state data will travel through the manufacturing process with the product. This requires that data is communicated to product lifecycle systems at the product companies and to service providers simultaneously.

A number of SEMI standards are facilitating this shift, including Equipment Data Acquisition (EDA), to improve and facilitate communication between manufacturer’s data gathering software applications and factory equipment.

SEMI kicked off an advisory council around smart manufacturing, and will coordinate a Smart Manufacturing symposium at SEMICON West on Wednesday, July 14, and again at SEMICON Europa on Oct. 25 in Grenoble, France.

Thomas Sonderman, vice president/GM of Rudolph Technologies’ software business, said the advisory council links the fabless and the equipment OEM supplier communities. One goal, Sonderman said, is “to help understand what’s required to really take on these concepts, and turn them into something that people can use to improve their overall fab efficiency.”

At the Smart Manufacturing Symposium, Sonderman will discuss what he calls traceability: optimizing the supply chain by blending IoT technologies. How information is acquired and used for Big Data predictive analytics and machine learning is one key aspect. “How do you turn data into some kind of actionable intelligence? I think the idea is to get some consensus around what it actually is, and then what’s required to make it successful,” Sonderman said.

Data security is also important. Data that comes out of fabs is of interest to suppliers, the fabless community and IP companies, among others who create a virtual IDM. “How does a Qualcomm get access to their relevant information, and on the other side, how does a company like Tokyo Electron Ltd. (TEL) or Applied Materials or Lam Research get access to that same information so that everybody can make the right decisions and shift the paradigm from reactive to a predictive/proactive approach.

“We need to go from ‘Hey, I have this problem. What caused it? How can I go fix it,’ to ‘What kind of analysis do I need to do to run my business? What kind of business intelligence is required to run the business, and how can I create analytical scenarios so that I can make sure that I have the information relevant to me to make decisions I need to minimize my time to market, and maximize my profitability?’”

In order for smart manufacturing to succeed, companies must be able to build confidence that they can share data securely. (At Wednesday’s symposium, NextNine, an Israeli IT security company, will present its work with TEL, several U.S. security agencies, and others concerned with moving information around securely).

One opportunity, Sonderman said, is to provide information-linking capabilities to 200mm and smaller wafer manufacturers, making RF filters, sensors, and other products.

“They don’t have a lot of the traditional capabilities that you come to expect. The idea is to link their information together but do it in a way where you can adapt it into those older facilities,” he said.

Rather than use a standard SECS/GEM interface, some tool data can be acquired wirelessly.

“There are all types of information that are relevant to the products, and if you think about what goes on a lot in the fabs it is linking what goes on in the product to what’s going on inside the tools. At legacy or non-leading-edge technology fabs, some of this in itself is a challenge,” he said.

Manufacturers also seek to link metrology data, with two different threads of information coming in: one from wafer-level metrology, and another stream of information from the equipment, which collects data each time the wafer crosses that piece of equipment. Also relevant is product information, including processes that can run multiple products. Figure 1 shows how this kind of data may be collected and shared in the future.

“The concept here is that you link these together in threads and then you create what we call the thread synchronization engine, which allows taking all of this relevant information and create a tapestry of data, which is a very pure data set that’s very representative of the combination of all these different factors,” Sonderman said.

The same types of information threads are woven together in the back-end (packaging) operations, where advance analytics are becoming as essential as in front-end processes.

Analytics are multifaceted, involving everything from visualization, data mining, spatial pattern recognition, and virtual metrology information. “Ultimately what I’m doing is trying to create a wafer-level signature and a tool-level signature and combine those together to create some kind of information I can take action on. That’s the actionable Data Now concept,” he said.

The goal is to combine information, separating the signal from the noise, and then analyze the data to ascertain whether or not a given process step or combination of process steps has contributed to yield loss. By drilling down into the shared data, engineers can discover whether a tool or set of tools is causing the problems.

“This is where things get really interesting. First, you have got to link everything together across the supply chain. Then you have to start looking at how do I drill down inside the equipment?” he said.

Large fabs with literally thousands of tools in operation are collecting huge amounts of information, essentially time series-based data. Linking tool information into an analytical combination with wafer-level information (what was going on inside the tool when those wafers were processed) is a powerful way to improve efficiencies. “That’s where this combination of big data analytics and traditional real time FDC is coming together,” Sonderman said.

To make this work, companies need a Big Data architectural environment, which combines structured data (in many cases in an Oracle database) with unstructured data (often text data, such as maintenance logs). Finally, there is a third space, a combination of time series-based data, such as images and spatial patterns.

The challenge, Sonderman said, is to link all the data together, standardizing the data so that it can be matched with various machine-learning algorithms. “From that I can analyze the data and start spitting out useful information that people can take action on,” he said.

To do that, the industry must deal with the security challenge. “There are ways to solve that challenge, but if we don’t solve that as an industry — and it really is an industry challenge — then we’re going to be handcuffed in terms of being able to take this technology to its ultimate realization. I think that’s now become the priority, versus preparing for the next wafer size and all that,” Sonderman said.