Lights out manufacturing is gaining steam across the semiconductor industry, accelerating productivity, improving quality, and reducing costs and environment impact.
These benefits are the result of years of strategic investments in technologies like machine-to-machine communication, data analytics, and robotics to achieve higher levels of autonomy.
Semiconductor factories have long depended upon automation of operations to keep production on time and to meet yield and quality objectives, but since the introduction of Industry 4.0 (aka Smart Manufacturing), major semiconductor producers have increased their investment in IT infrastructure, data analytics, IoT sensors, digitization, and robotics. 
By limiting human intervention, a lights-out factory (also called a fully automatic factory) can operate more smoothly and respond to issues in a non-disruptive manner. The goals are maximum tool uptime, no unscheduled downtime for maintenance, and real-time lot dispatching. Fabs and packaging houses have been continuously improving their operations for years, but with the availability of new technologies (e.g., IoT sensors, cloud computing), they can implement higher levels of automation and achieve a bigger payback for their efforts.
“From my perspective, it’s a journey,” said Bob Reback, general manager and vice president of the Cimetrix Connectivity Group at PDF Solutions. “In the 1980s we used robots for automated material handling and computers to connect and control equipment. We remotely started and stopped the equipment, collected data, and did very basic process control. We took humans out of the equation and ran factories with software instead of humans. That’s smart manufacturing. But if you look at the cost and the reliability of the systems, it’s been getting better and better. Today, everyone is doing some level of smart manufacturing. The question is, ‘How do you get to smarter manufacturing?’ The answer is, ‘Continuous improvement.’”
The definition of smart manufacturing has evolved, as well. “Factory automation in the past automated all the repetitive manufacturing processes, including material transportation and equipment automation. Today, smart manufacturing in fabs refers to the synergizing of IoT, data-driven, and AI-driven technologies to optimize operational performance,” said James Lin, deputy division director of smart manufacturing at UMC. “Fabs are facing increasingly difficult and complex manufacturing challenges. By introducing smart manufacturing projects they can achieve the goals of process optimization, flexible production, shortened delivery time, manpower quality, and enhanced efficiency.”
Smart manufacturing projects increasingly will be implemented, and the duties of factory workers will shift alongside of those implementations.
“The benefits of a fully automated factory include increased efficiency, productivity, and accuracy, as well as lower rates of error. However, implementing a lights-out factory requires significant investment in automation technology, and may not be feasible for all manufacturing processes,” said Won Lee, vice president of smart factory engineering at Amkor Technology. “Furthermore, it is important to note that a fully automated factory does not necessarily mean that no human workers are involved in the manufacturing process. While the production floor may be devoid of workers, there may still be human operators overseeing the control systems and maintenance of the equipment, as well as performing other functions such as quality control and inventory management.”
Others agree. In fact, Industry 5.0 emphasizes a human-centric implementation.
“A ‘lights-out factory’ traditionally implied a factory where human activity is at a minimum, that it can operate in the dark. However, this is not a realistic goal, and a more achievable definition of a ‘fully automated factory’ is a facility that can run without intervention for a period of time,” said Mark da Silva, senior director of smart manufacturing initiative at SEMI. “Having a human in the loop is increasingly seen as necessary. Industry 5.0, which lies beyond Industry 4.0, places the well-being of the worker at the center of the production process for a sustainable, human-centric industry.”
Strategies and technologies for automation
As several industry experts noted, a fully autonomous factory operation requires an investment to enable the IT infrastructure, data analytics, and automation. It also requires a framework to evaluate objectives and needs like SEMI’s Industry 4.0 Readiness Assessment Model. 
Amkor’s Lee summarized his company’s strategy in three steps:
- Data acquisition: To enable the curation and acquisition of disparate sets of useful data across the business, supply chain, and the world. The IoT allows connected machines to gather data into the system.
- Data analysis: To use advanced analytics and modern data management solutions to make sense of all the disparate data gathered by machine learning and intelligent business systems.
- Intelligent factory automation: Once the data acquisition and analysis have taken place, workflows are established, and instructions are sent to the machines and devices within the system.
As these technologies are implemented to support smart manufacturing initiatives, progress can be measured in terms of level of achievement in automation.
“Currently, UMC is focusing on manufacturing model optimization and re-engineering in digital transformation. In 2022, SEMI GEC (Global Executive Committee) Smart Manufacturing Initiative proposed the pyramid model of the Autonomous Smart Factory,” Lin noted. “Based on this model, we are now at the Smart 2.0 stage, where we are leveraging various data across multiple point systems and digital twins, and utilizing AI/ML to automatically generate insights, such as quality improvement, asset utilization, manpower efficiency, energy saving, and integrated supply chain.”
Fig. 1: SEMI’s Smart Manufacturing Vision pyramid originally was created by Inficon and adapted by the Smart Manufacturing GEC. Source: SEMI
Wafer factory automation needs for connecting machines to data and control systems has long driven standards for semiconductor equipment communications for vendors to support. These standards, managed by SEMI, continue to provide a foundation for smart manufacturing and include the following:
- SECS/GEM defines communication between equipment and factory network.
- GEM300 enables greater automation.
- Equipment Data Acquisition improves communication between a factory’s data gathering software applications and the factory equipment.
Assembly and test facilities also need to standardize the integration between equipment software and automation systems. But historically, OSATs have not utilized these standards. Business operations are a bit more complex, due to the variety of equipment, products, and collateral (e.g., trays) in the facilities. Recently, some OSAT have made concerted efforts to convert systems and connect equipment data with SECS standards to provide the technology framework to implement Industry 4.0 approaches.
Changes also are needed at the tool level. “Several systems need to be converted. Production and inspection processes should be standardized to facilitate automation implementation. Equipment should be more intelligent and capable of recognizing information about production materials on its own, with built-in artificial intelligence capabilities,” noted Joon Ahn, vice president of IT division at Amkor Technology. “Legacy systems such as ERP/MES/CIM should be able to communicate in real-time with new on-site systems and handle more data. Additionally, simulation or prediction systems should incorporate machine learning algorithms to reflect reality more concretely.”
To address inadequate equipment connectivity, for instance, ASE Group is requesting that all new production equipment be SECS standard compliant. For existing equipment their automation teams have developed capabilities to achieve an automatic connection as well as convert data output into compatible SECS formats. After years of development, ASE’s production equipment now meets SECS standards. 
Another technology that engineering teams rely upon is data analytics. Automation and data analytics assist in the shift to from reactive predictive actions, and engineering teams can choose the level of automation in the response. In addition, multi-variant analysis of all the available data sharpens the focus on likely root causes of device failure issues.
“For customers that want to move past reactive analytics we provide two paths,” said David Park, vice president of marketing at Tignis. “We provide both predictive and prescriptive analytic solutions, and they can be done independently or together. In the reactive mode, the customer is primarily trying to determine what happened and why did it happen. Then, depending on their time-to-data, they may be doing this for hours to days to weeks after the event. The primary goal of predictive analytics is to determine if something bad will happen soon, enabling the fab to take proactive action. So at the simplest level, this provides a fully automated early warning system. No human is needed, and the customer can decide what they want to do when a warning occurs.”
This can be taken a step further with prescriptive analytics. “Just because you can monitor all the inputs and environmental variables of a manufacturing process to look for a signal that strongly implies something is out of control, it doesn’t mean that the same system can tell you the warning’s root cause,” Park said. “Predictive analytics provides a minimal list of univariate, bivariate, and trivariate root causes. This enables the customer to find the needle in the haystack much faster because it can immediately show them the most likely cause of the alert. Again, there is no human in the loop, and one isn’t needed because this provides actionable information.”
Implementation successes and challenges
As semiconductor companies, wafer fabs, and OSATs implement Industry 4.0, engineering teams reap the efficiency and cost benefits, as measured by a reduction in processing errors, a shift from reactive to predictive actions, improved traceability, yield, and quality, and accelerated identification of production issues.
“Our projects include industrial AI (IAI) projects, IT infrastructure, smart supply chain, physical and information security protection, and the integration of semiconductor upstream- and downstream-related industries to jointly create a smart manufacturing ecosystem — achieving the most efficient use of manpower, equipment, material, and methods,” said UMC’s Lin. Moving forward, he noted, “We have implemented “people-centric” smart manufacturing platforms and systems centered on three major directions — company-wide AIoT solutions, intelligent big data analysis, and the fusion of multiple AI technologies. Achieving an autonomous smart factory will not only make UMC more competitive in terms of quality and efficiency, but also enable us to reduce waste and the environmental impact of our production activities.”
Fig. 2: Technologies supporting a ‘people-centric’ smart manufacturing solution. Source: UMC
Since implementing Industry 4.0 initiatives, Amkor has observed a 60% increase in process engineers’ productivity, which is attributed to using more data analytics and implementing fault detection and classification (FDC). These technologies enable real-time quality control, which supports achieving zero defects on automotive products. 
Converting to fully automated facilities requires planning and commitment. Once started, it quickly becomes a conversion of multiple facilities.
In 2015, ASE Group began investing in technology for converting to lights-out operations. To define the needed technology and employee training, it created a committee comprised of its information management center and automation teams for each of its business units (e.g., lead frame, ball-grid array, flip chip, etc.) By 2021, the company had 25 lights-out factories in operation, with more than 600 automation engineers trained. 
Still, the path to success isn’t entirely smooth. Data usage and integrity, implementing a digital thread, communication compatibility, and the uniqueness of some factory operations all have been cited.
“The biggest challenge is clean, accurate, and consistent data,” said Tignis’ Park. “Without it, you have a garbage in/garbage out situation. The customer needs to have a solid data infrastructure in place to take full advantage of AI/ML.”
In addition, there’s challenges in compatibility between machine communication protocols to access the data and the noise in sensor data measurements.
“Since there are various communication protocols used between machines, ensuring compatibility is not easy,” said Amkor’s Lee. Also, IoT data typically requires efforts in collecting and analyzing as big data, and there is a significant amount of noise in sensor data that requires effective filtering. To overcome these challenges, we are creating standard communication protocols or creating interface programs between different models, introducing cloud databases to ensure smooth scalability, and using artificial intelligence technology to automatically filter out noise data.”
For engineering teams to effectively continue increasing smart manufacturing approaches, they need to design their systems with intent.
“Smart manufacturing is focused on increasing productivity using data, networks, machine learning, and digital systems, which goes far beyond “simply” automating processes or using in-line manufacturing data for process control,” said Russell Dover, general manager, Service Product Line, Lam Research. “It requires a holistic digital vision to create a “digital thread” throughout your operations. A digital thread is the construct of using a consistent digital approach from design, through to manufacturing, sales, support, and even end-of-life of a product.”
The assembly process presents challenges due to the diversity of production collateral used. “In the assembly process, the carrier’s form containing the material changes continually as the process progresses,” said Amkor’s Lee. “Therefore, more variables need to be considered and customized automation than is necessary in a fab.”
Fab and OSAT operators have invested significantly in IoT devices and networks, robotics, ML/AI software, cloud computing, and big data analytics. By doing so, they have achieved fully automated, lights-out factories, and they are reaping the benefits.
“The demand for increasingly faster cycles of learning, and at lower cost, makes Industry 4.0 promising for semiconductor manufacturing,” said Lam’s Dover. “Powered by big data, artificial intelligence and sensor technologies, smart manufacturing techniques can convert data generated by process tools into tangible value for chipmakers, including improved yield, reduced downtime, and lower wafer costs.”
Multiple semiconductor industry sectors are stepping up their investment in smart manufacturing approaches.
“We’re helping equipment makers and factories pursue smart manufacturing in semiconductors, advanced packaging facilities, photovoltaic solar panel, and electronics factories. But it’s basically the same playbook,” said PDF Solutions’ Reback. “We need to connect to things on the factory floor, automatically collect data, and once we have good data — then we can look at applying analytics.”
Reback noted that customers are more and more interested in placing some of this data in the cloud and using new analytical approaches. “The big difference is that today, with all of the investments in AI and ML, the technologies and tools that are available for us to collect and analyze data are far superior to what we’ve had in the past. We’re seeing people across the ecosystem interested in doing projects to expand that envelope of what we can achieve.”
Smarter Ways To Manufacture Chips
Early successes are spurring further investments, with a concentration on high ROI projects.