Smart Manufacturing
Lights Out at USI
The EMS behemoth is on the cusp of an all-automated future.
by MIKE BUETOW
By almost any measure Universal Scientific Industrial is an EMS behemoth. Yet most of the press surrounding USI over the past few years has been tied to its recent acquisition of AsteelFlash. The deal, completed last month, added 17 manufacturing sites and about $1 billion in topline revenue. For the first time, USI will have sites in the US, Africa and Western Europe.

Today, USI has 27 manufacturing locations in 10 countries, over 24,000 employees and revenue of more than $7 billion. That’s good for the 11th spot in the current CIRCUITS ASSEMBLY Top 50 rankings. There’s no missing the company now.

Yet for all its size, USI could just as easily be recognized for its technical prowess. The company is on the cutting edge of so-called lights-out manufacturing, where few if any staff are found on the factory floor where hundreds of SMT machines run seamlessly, connected by sophisticated software and AGVs feeding the cells on a just-in-time basis.

Jim Cao, general manager of the Greater Shanghai and Smart Manufacturing/SiM business unit, USI, is charged with developing and overseeing the company’s worldwide Smart manufacturing strategy. The company is entering year six of the plan, and the progress is notable. He spoke about how Industry 4.0 has changed the way the EMS firm operates with CIRCUITS ASSEMBLY in late November.

MB: In a press release on Nov. 10, you said, “USI is adopting Smart manufacturing to enhance our competitiveness through the digitalization of our production processes, including the automation of infrastructure and logistics using robotics and automated guided vehicles.” What does Smart manufacturing mean to USI?

JC: At USI, we have a very detailed Industry 4.0 Smart manufacturing plan and roadmap. We call it Worldwide 5 Star I4.0 Smart Manufacturing Roadmap and Management System. It lays out our stage-by-stage plan over the next five to seven years to bring up the Smart manufacturing level of our worldwide factories. Every year we measure the progress we made. We have month-by-month reviews of the system, and at the end of the year we do a more careful rating of all the factories of the progress made toward the goal.

Factory inspection machine awaiting use
Figure 1. All inspection machines from SPI to AOI are linked and constantly driving process improvements.

Our Smart manufacturing definition is based on the Industry 4.0 contents, and it has four pillars:

  1. Automate all machines. The focus is to reduce manual labor and minimize any manual handling costs and issues.
  2. Industry 4.0 data automation. Data automation is the essence of I4.0. It digitizes and connects all data to the central server, so we have real-time data from all production equipment and process SPC data and RMS (recipe management system). Every production lot is scan-in and scan-out at every process station using 2-D barcode. One hundred percent of the units, through the entire production line, also have a 2-D barcode system, and are scanned and loaded in a central server. We know exactly, for every unit we ship out, if there is any issue coming back; it is easy to trace back to which unit, which shift and which hour.
  3. Our Automatic Material Handling System. That fully automates our entire components incoming to the warehouse and dispatched by AGV to each SMT pick-and-place in JIT mode, right before a pick-and-place machine runs out of components (Figure 1). We have eliminated the kitting room and associated labor. Every machine we have is linked to a production floor management system we call Shop Floor. Shop Floor is connected to our ERP system. Our data automation is integrated with Shop Floor, so we know exactly to the minute when every machine is running out of material. The system would automatically dispense the right components from Smart Storage driven by robots by retrieving right corresponding component reels and put the reel on a cart, which carries them to the corresponding machine.
  4. AI-based learning system. We are just getting started. It’s a major undertaking. We see huge potential there as well, to have the machine at a certain level, automatically adjust the inspection criteria, to reduce or reject, and also eliminate under-rejects, to inspect and fine-tune the parameters, because there are so many variables (FIGURE 2). For instance, the substrates we use have thickness variations within a certain spec range, and components do as well, so if we keep fine-tuning the process parameters within our specifications, driven by the SPI, we can improve our quality.
A visual breakdown of robotic cells speed and quality
Figure 2. Robotic cells replaced hundreds of technicians and markedly improved speed and quality.

MB: Do all the facilities use the same equipment sets?

JC: We have factories worldwide. If you look at our US reporting, our business is pretty much two portions. The first part, which is the major part, is the so-called system-in-package (SiP) miniaturized modules. The second part is the EMS business.

For EMS, the equipment is less complicated than the SiP factories. The SiP factories’ process flow is very similar to typical IC assembly and test factories. The only major difference is we use SMT instead of wirebond. The rest of the backend and test processes are very much the same. It’s much more difficult to make modules than single-chip IC products.

Not all the factories are at the same level yet. My philosophy is we use the Zhangjian factory in Shanghai, where our Smart manufacturing development team is located, to develop all the common platforms. We use a common platform strategy to standardize the data automation protocols and equipment, which produces our development lead times and costs as well. Our strategy is to develop the common platforms, then fan out to the worldwide factories, as each factory needs it.

MB: Was the software, which is extensive, developed in-house, or do you start with machine software and modify it to meet your needs?

JC: We do both. For all the automation equipment, we migrated from the original third-party design and programming. Now a majority is in-house design and programming. Obviously, we have to integrate all the standard, off-the-shelf, third-party production software, but we work with them and integrate together, mainly through two methodologies. The first is the SECS/GEM protocol. Not all the equipment has SECS/GEM, especially the older equipment (five to seven years old). We also developed our own protocols to use with LORA IoT through machine signal tower, so we know machine status: for example, green light, yellow light, red light. SECS/GEM is by far the most flexible, comprehensive machine tool central server data communication platform. It offers two-way data transfer. The central server can control the machine, and the machine provides real-time data to the server. For older machines without SECS/GEM, it’s pretty much a one-way communication from the machine to the server. It was limited data transmission from the server to the machine, unless we can work with the supplier to have access to their PLCs and machine-level PCs.

MB: Are the AGVs you use purchased from traditional electronics factory suppliers or from outside our industry?

JC: We use a third-party supplier based in China. That technology is pretty mature now. It’s been in the market for the last 25 years. In my career, I first saw the AGV in the Intel factory in Chandler, Arizona, about 22 years ago. It’s now popular in China.

MB: We in the US are way behind you.

JC: I’m not so sure. I worked in the States for close to 30 years. When you say the US is behind, I think the main reason is the loss of most of the manufacturing jobs, which moved to China, initially for the cheap labor. But the China labor rate is inflating every year, and every China company is under pressure to speed up their own automation to stay competitive.

MB: You talked about some of your recipes. How deep does that go? Are you able to go back all the way to the component distributors or even the component OEMs and tie your inventory needs on the line, as they are starting to wear down, all the way back to the component supplier level so the just-in-time is as tight as possible?

JC: The answer is not yet. Let me explain a little bit. Our material planning system is very much SAP-based. We do not typically store months and months of inventory. Especially for the products we make, the material supply is quite tight. We use SAP to manage that material demand. In our Five Star I4.0 Roadmap, the last stage would be exactly what you said. We plan to integrate with our customer’s ERP system, so that when the customer dumps their forecast for the next six months, by connecting their ERP with our ERP, our system automatically checks the inventory, calculates yield loss, and cuts it down to the particular component-level demand for the next six months, and automatically transmits to the suppliers’ ERP system. That’s a fully integrated supply chain. We are not there yet. Obviously, the reason is very much dependent on our customers’ readiness and, more importantly, our suppliers’ readiness.

MB: Have you found that component manufacturers are eager to get to that point, or are they concerned that would create less clarity for the inventory levels because of concerns of double or triple ordering during tight inventory times? How trusting are they of software-driven inventory pulls?

JC: It’s a good question. Our suppliers are mostly the IC manufacturers. Our customers mostly are fabless. They typically do assembly and test in one of the major OSATs. They are not pushing that heavily yet.

In terms of their trust of their system, I think it would be much more trustworthy to them because, if we have the fully integrated customer forecast to the supplier demand system fully linked, it would take out a lot of the human errors or manipulations. It would be data to data. It’s a real B2B. It would be much more accurate and trustworthy to our suppliers in terms of demand numbers. Right now, everyone is moving slowly in that direction. It will happen. There’s no doubt. It will improve the productivity of our customers and suppliers and ourselves as well.

MB: You talked a bit about what the AGVs are doing in the factory. Are there any other areas where you have implemented robots, and if that’s the case, what are you having those robots do?

JC: The product we make are modules. It’s a system, not an individual IC. The test time for each can be very long … a couple minutes. Each robot can support eight to 10 test sockets for long-test-time module testing. We have hundreds and hundreds of robots (FIGURE 3). In the old days, three to four years ago, we had hundreds and hundreds of operators manually inserting parts into the socket, closing the socket, pushing the button. It’s a lot of human errors. Now those hundreds of human operators are gone, replaced with robots, with much higher efficiency and much higher productivity.

You probably see the AGV on the floor (Figure 3). That’s the project we are working on right now. My plan is to have USI’s first lights-out test floor factory next year. We’re going to have at least 91 of those test systems totally lights-out, all product loading and offloading and machine status done remotely.

AGV machines operating hands-free
Figure 3. AGV move component trays hands-free across multiple floors to feed assembly machines.

We use AGVs to carry carts similar to the one on the right in Figure 1 upstairs to an elevator and then to a system next to the machine that needs the parts. It’s a fully integrated AMHS system. At the heart of that is the so-called Smart Storage system (FIGURE 4). Inside are all the component reel slots, which are all handled by robots moving quite fast, managing all the component reels. They are put on the shelves and retrieved as needed, then moved on the automatic conveyor connected to the robots at the other end to put on the AGV carts.

We have lots of dashboards around the factory. One is the AMHS Smart Storage dashboard (FIGURE 5). It shows how many reels are inside, the size of them and how many parts they have on them. All the data are real-time, on display in the system and fully automatic.

MB: I’m assuming USI does studies every time you implement an automated system. Are there standard metrics that you use to compare how much faster and accurate and higher-yielding a particular process has become?

JC: There’s a day-and-night difference in terms of accuracy. When we have human labor to do the work, the accuracy is way off. There’s no comparison. The robot we typically use has higher repeatability, placement accuracy, +/-20 microns. Humans wouldn’t be close. They’re not as dependable as machines in terms of material handling.

MB: Regarding the Five Star Rating System, how did that name come about?

JC: The “five stars” name is commonly used for hotels, and there’s clear definitions in terms of what we use for one star, two stars, or a half star. Typically the measurements include the manufacturing process automation rate. For each production machine, we divide it into three steps: loading, process, and uploading. Each station has three steps.

Three quarter view of intelligent in-warehouse storage system
Figure 4. Parts are stored on reels inside intelligent storage systems in USI’s component warehouses.

If we automate the processing part, we still need a keyboard to load the parts, offload parts, so we only count as one of three. If we automate all, it would be three steps. So, the measuring system is doing all the manufacturing process systems. The percentage of machines connected to the data system network, we call it connection rate. It’s machine downtime reduction, or MBTF-based. With I4.0 data automation, we have very much reduced machine downtime because when you have hundreds of thousands of equipment on the production floor, a lot of things are happening. When a machine goes down, the equipment technician or engineer may not know which machine is waiting for assistance.

Now we have a system to send automatic text messages to the responsible engineers, so they know immediately which machine is down and what kind of problem, so they can quickly fix it, instead of wasting a lot of machine time waiting for a human. We also measure the DL [direct labor] and IDL [indirect labor] headcount reduction, which is a fundamental objective to improve productivity. How many headcounts did we reduce? What percentage? How many technicians do we use to collect data, to go to the production floor and collect SPC and yield data, for instance, and write it down on a piece of paper, go back to an office and enter in a computer, which is not productive? Now we have real-time data collection and transmission to a central server and displayed on the dashboard, on your office desktop computer or even your smartphone. We have a comprehensive system to measure the bottom-line results from I4.0 Smart manufacturing.

MB: In the Zhangjian factory, is SMT assembly laid out in a straight line?

JC: We probably have best SMT production lines in the world in terms of speed and accuracy. We have multiple lines grouped together by process in order to maximize machine efficiency (FIGURE 6). We do not link the front-end machine to the backend machine dedicated through a conveyor belt, which is very old-fashioned and not so flexible because, very simply, if one of the machines in the chain goes down, the WIP will start piling up. The other machines upstream and downstream all go down. We focus very much on efficiency and machine utilization.

view of AMHS Smart Storage Dashboard
Figure 5. Among the data the AMHS Smart Storage dashboard shows are the number of reels, their size and how many parts they contain.
MB: I count at least eight placement machines in that configuration.

JC: Yes.

MB: That would be a single product being shipped through all those machines?

JC: Correct.

MB: And the line is running from left to right. How many printers would feed that configuration?

JC: One. The products we build are miniaturized. It’s actually smaller than your fingernail.

MB: I’m surprised that one printer can keep up.

JC: It can. To maximize our productivity, we use a large-size PCB. Typically, we use a 95 by 240mm substrate. On the substrate, depending on the size of the product, we have anywhere from 100+ to 400 units. We just print at the PCB level instead of the unit level.

The smallest components we use in production are 01005 or 008004. It’s very, very high density. Each module typically has 100 to 200 different components, or even more. The smallest module we make typically has 800+ components. It’s very high density.

MB: You’ve referred in press releases to a parameter management system.

JC: The first parameter management system we have implemented is the inline SPC. The SMT line in the photo has an inline SPC system in terms of the component placement accuracy. We measure, collect the data, and feed it to the SPC control module, so we know when the process is going to deviate. As you can imagine, we have hundreds and hundreds of vacuum cups on those pick-and-place machines. If any of the vacuum cups are worn out or the component is off, then we catch that because we have the fully integrated AOI machine there, so we know exactly which vacuum cup caused the problem. It’s our first step we are implementing. Down the road, the parameter management system will allow the spec to fine-tune the machine parameters based on certain self-learning capabilities.

I’ll give you another example. We have fully automated all the outgoing cosmetic inspection processes. We are very particular about any cosmetic defects, such as a very fine scratch or a tiny black dot that we could not see with our eyeballs. We have fully automated that. On the day-by-day fine-tuning of the parameters, we still have over-reject and under-reject rates. Our next step is to implement what I called the closed-loop machine-learning-based parameter management system. Based on the outcome of the over-reject and under-reject rates that came from the system, the machine is going to fine-tune the inspection parameters to narrow the defects.

MB: How long did it take to design and implement all of this?

JC: We started five years ago. One of my responsibilities is to lead the worldwide manufacturing effort. I do have a dedicated team to develop all the machine and data automation and help to implement those. It’s been about five years.

USI's printer machines
Figure 6. A single printer can feed eight SMT placement machines. USI says up to 400 highly dense SiP modules may be processed on a single substrate.
MB: How close to the original schedule are you?

JC: We are very much on schedule. On average last year, the worldwide factories improved about a quarter star. Some factories made more progress, some less. It very much depends on the business situation. Each factory is different. If the factory is booming, the manager is more willing to invest capex in automation. We are making progress very steadily toward our goal.

We take 4.0 Smart manufacturing very seriously. We invest a lot of resources and capex. We want to upgrade our manufacturing capabilities and improve our competitiveness. We are marching toward our final goal. We are on the way.

Mike Buetow is editor in chief of PCD&F/CIRCUITS ASSEMBLY; mbuetow@upmediagroup.com.