Smart Manufacturing Gains Momentum

Problems remain for legacy infrastructure, but adoption will continue to grow as gaps are identified and plugged.


Smart manufacturing is gaining traction as a way of addressing increased market fragmentation while still leveraging economies of scale. The goal is to add a level of flexibility into manufacturing processes that until recently was considered impossible. Although the approach makes sense in theory, real-world implementation is proving far from consistent.

Horizontal color image of businesswoman - unrecognizable person - working with digital tablet in large futuristic factory. Woman standing on top of a balcony, holding touchpad and checking inventory of a manufacturing company on touchscreen tablet. Focus on futuristic machines in background, businesswoman's hands holding black tablet defocused.

Sometimes referred to as Industry 4.0, particularly in Europe, or the Industrial Internet of Things, smart manufacturing represents a fundamental rethinking of automation within industry. In effect, it reverses Henry Ford’s famous comment, “Any customer can have a car painted any color that he wants so long as it is black.” With smart manufacturing, a car could be made in any color, with any options, but with the same kind of quality control and at a comparable cost.

The semiconductor industry is one of the best examples of this scheme in action today, in large part because fab equipment and processes are updated so frequently. Foundries already are producing multi-project wafers on a mass scale, and they have built ecosystems that can communicate at least some details about inventory shortages or surpluses. These imbalances used to cause market gyrations in the early days of semiconductor production. That doesn’t happen very often anymore. And on the manufacturing side, at least the more recently developed tools are equipped with enough sensors to be able to spot irregularities before they cause yield or quality problems, along with communication capabilities to relay that data almost instantaneously.

“The semiconductor industry is farther along than other industries with the ability to inform, control and automate,” said Tom Salmon, vice president of collaborative technology platforms at SEMI, and the person charged with spearheading the group’s Smart Manufacturing Initiative. “That extends from the front end to the back end within the fab. There is also smart manufacturing going on between the fabs and the tools manufacturers. If you look at the tools from KLA, Lam and Applied Materials, they can tap into a tool with remote diagnostics.”

A key missing piece is what SEMI calls a “digital thread,” which is the ability to look at the supply chain end-to-end. “If there is a natural disaster, that has an impact on supply chains,” said Salmon. “The goal is to develop ways to avoid downtime and supply issues.”

But what happens in advanced chip manufacturing doesn’t translate so easily to other industries. Gaps that exist in the semiconductor world are much wider elsewhere. Inserting sensors inside an aluminum stamping operation, for example, or a food processing plant, can generate useful information, but tying it all together into a cohesive system where meaningful data can be collected and acted upon in real time is a monumental challenge. Some of these other industries are run the same way they were being run before semiconductors were even invented. Adding these capabilities to other parts of the electronics supply chain is hard enough.

“The semiconductor world had a method of communication between machines where they understand all the processes,” said Michael Ford, senior marketing development manager in Mentor Graphics’ Valor Division. “They’re quite advanced. But if you look at PCB assembly, it’s quite the opposite. There are hundreds of machine providers and they have never agreed on any format for sharing information. The top ones sell a smart line solution, but that only works if you buy from one supplier.”

The tech industry began addressing this issue as early as 1996, when the Electronic Data Interchange specification was introduced by the National Institute of Standards and Technology (NIST). The standard was an outgrowth of military logistics, which created a uniform way of classifying military supplies. Tech companies used it as a way of managing their supply chain, particularly in the PC era.

Communication has progressed significantly since then. Software now can be used to connect disparate protocols to provide much more than just supply chain information. “If you look at line-level manufacturing, in the past you had to have people there to find defects,” said Ford. “With smart manufacturing, you can now apply six-sigma capabilities to predict when something will fail. At that point, you can direct engineers to make changes. It’s a closed-loop feedback. You also can provide lean, just-in-time materials management because you know what’s on a machine.”

Branching out
But even within semiconductor manufacturing, communication isn’t quite so seamless. While the most advanced 300mm fab equipment has an array of sensors, alarms and communication capabilities, there is far less sophisticated technology even in the best-equipped 200mm fabs. Some of the older equipment has been retrofitted, but results vary greatly. While a big foundry may employ smart manufacturing, not all of the manufacturing may be classified as smart.

There also are degrees of what is considered smart. Machine learning and artificial intelligence are likely to show up in industry first, in part because a business case can be made for adding those capabilities, and in part because companies can afford them.

As this approach is adopted, it is expected to drive demand for new capabilities within the underlying technology, notably in rapid communication within chips and between chips.

“This is more adaptive, real-time decision-making,” said Mike Gianfagna, vice president of marketing at eSilicon. “The real question is whether it will have the algorithms to learn, improve and adapt. That all comes back to how you store, maintain and move information in a rapid way, which is where high-bandwidth memory comes in. You’re already seeing this with chips from companies like Nvidia, which has a massive compute platform with HBM. As this gets adopted, it makes the manufacturing process itself more robust. And it opens up the opportunity for big packages and systems that are no longer limited by the stepper field.”

Both the technology to make it happen, and the concepts to tie it all together, are improving. What’s less obvious is that they are winning adherents across a broad swath of markets that have shown little interest in the past in this kind of technology.

“There are a lot of older industries where there are still a lot of things that need automating for the wellness of the workers,” said Jean-Eric Michallet, vice president of sales and marketing at Leti. “If you look at libraries, there are a lot of books that need to be moved around. That results in back problems for the people who work there, so libraries are installing automation. When a person returns a book, they give it to a machine to sort and put on a shelf. So now the librarians can advise people on which books to take out, rather than spending their time filing books.”

This is one of the benefits of the IIoT. There are overlapping principles and techniques in markets that seemingly have little in common, but from an infrastructure standpoint that may not be the case.

“In the semiconductor industry we have clean manufacturing,” said SEMI’s Salmon. “If you look at a biomedical facility, the same capabilities may be just as beneficial. In a job shop, you may have a batch processing flow, where communication is set up around the flow. In packaging and assembly with a PCB, that’s a type of flow shop, too. You can modify standards for both lines.”

SEMI also is looking at the skill sets needed for these jobs. Data scientists might have a role in software information and control, automotive engineering and process engineering. “We also will be looking at use cases in a fab. We haven’t gotten to that level yet, but we are working on specific use cases with the Fab Owners Association (FOA). This year we will open that up to a larger special interest group so we can provide case studies and use cases, such as how to retrofit a legacy fab with minimal impact and secure solutions.”

Making those solutions secure is a big issue. The whole premise of smart manufacturing is better communication within a manufacturing facility, and between that facility and trusted suppliers or partners.

While most of the new technology comes with built-in security, factories typically are a combination of new and old. In some cases, factories might have been built more than 50 years ago. Even if the manufacturing equipment is updated, the communications infrastructure may be dated. That means it may not be secure, because security works far better when it is developed at the architectural stage rather than as an add-on, and it might not be able to handle the volume of data required to fully take advantage of smart manufacturing.

“Security is a bigger issue than ever, because this technology makes it possible to go deeper and deeper into the data,” Scot Morrison, general manager for embedded platform software at Mentor Graphics. “Initially, there was a lot of thought that a secure perimeter was enough and that it does not matter what goes on internally. That has been largely debunked.”

Adding security into older systems isn’t easy, and it’s not always effective. “Security can be enabled through legacy protocols,” Morrison said. “But even where it has been set up for appropriate levels of security, the fragility of it and the difficulty in setting it up erode its value over time. Security is diminished or it is taken down to debug something and never turned back on. What you really need here is security that is intrinsic so you can’t turn it off.”

Morrison noted that communication becomes the backbone of this system. Robots can be replaced piecemeal, but legacy protocols are much harder to reconfigure, which is essential for just-in-time manufacturing.

Leti has a security lab where it has been looking at adding obfuscation techniques into chips, such as crypto IP with security loops. “There is a cost, of course, and you need to determine what kind of security you really need and make sure that solution is compatible with existing semiconductor processes. You also need to make sure you can track IP in the supply chain.”

Political hot potato
Another issue is how many people smart manufacturing will replace, and at what level. There is strong political pressure in Europe and Japan, and more recently in the United States, to add or maintain manufacturing jobs. Smart manufacturing will create new jobs, but not at the same level as before. Many of those new jobs will involve data management, software engineering, and advanced networking skills, rather than the lower-level skills they replace.

“Robots will replace humans in some places, but if you look at washing machines, no one misses those jobs,” said Michallet. “There will be more added value jobs. It’s not just engineers. At the end, you still need humans to take actions. A plane is already autonomous. It can land automatically.”

What skill sets those people will require remains to be seen. Leti predicts that over time there might be fewer software people in manufacturing because the emphasis will shift to hardware solutions, which require more mechanics and physics than software. Others believe the skills will be heavily based upon algorithms and software.

“In the conversations we have with companies, a lot of the discussion is about data,” said SEMI’s Salmon. “But the objective is not to ramp up the data. The focus is on how to ask the right questions and how to use that data most effectively. That could mean parsing out only the data that you need.”

Who will design those systems? And can they be retrained from existing work forces?

Smart manufacturing has been talked about for the past several years, and sensors are being added into industrial settings for measuring a variety of functions, such as whether liquid is moving through a pipe or whether there is excessive vibration in a machine. But true utilization of that data in a cohesive and well-thought-out manner is going to be much more difficult than it sounds.

In new industries, this kind of approach may be a simple decision. It can be architected into an operation from the outset. In established industrial operations, mixing and matching equipment and bringing everything up to speed will be much harder, much riskier from a security and data flow standpoint, and potentially much more expensive. But the direction is mapped out, even if the individual steps to get there are not. And as companies begin implementing this technological approach, remaining competitive may require much larger investments, driving changes in how manufacturing is done, and what kind of technology is required to make it all work.

Change is coming, but how it gets implemented and by whom is not entirely clear.

Related Stories
Plugging Holes In Machine Learning
Part 2: Short- and long-term solutions to make sure machines behave as expected.
Agricultural IoT: Outstanding In The Field
The Internet of Things for agriculture. Yes, it’s a thing—and a huge market.
Medical IoT Showing Signs Of Life
Market growth is expected to be strong, but development cycle is slow.
Ready For Social Robots?
Companies rev up new technologies, but it’s still unclear how successful they will be.
What’s Next For IoT Security?
The recent cyberattacks highlighted the security lacking in many IoT devices. Solutions are on the way.

Leave a Reply

(Note: This name will be displayed publicly)