Faced with labor shortages, rapid production ramp-ups, and data overload, the process of designing and building greenfield factories requires a combination of engineering tools, failure of previous approaches, and superior planning from day one.
The complexity and size of semiconductor factories are increasing rapidly, as are capital costs. Chipmakers are trying to accelerate demand for multibillion-dollar factories despite hurdles such as labor shortages, material delays and intense competition.
A panel of experts at SEMICON West (see photo below) described ways in which augmented reality, artificial intelligence, synchronized data access and digital twins can help make smarter design and factory construction a reality.
Fig. 1: (LR) Chen of NVIDIA; To of Exyte; Smith of Bechtel; Fullam of Jacobs; and Mitra of Deloitte. Source: Laura Peters/Semiconductor Engineering
Part of that reality is the realization that design and engineering changes happen simultaneously. “It’s like flying a plane while you’re building it,” said Evann Smith, digital solutions manager at Bechtel. “When we think about building large-scale projects, we think of a locomotive. It’s linear, it’s going full steam down the tracks, and we know where our destination is. That’s not the semiconductor fab kind of thing. We’re talking about dynamic and fluid projects that present both a challenge and an incredible opportunity.”
This opportunity became clear to fab-less semiconductor company NVIDIA two years ago when it began developing a digital twin platform for fab infrastructure applications, a prototype of which Samsung Electronics plans to complete in 2025. “Samsung in Korea is starting to build this digital twin for its (fab) infrastructure, which was actually publicly unveiled at our GPU Technology Conference in March of this year,” said Jerry Chen, director of business development for Manufacturing & Industrials at NVIDIA. “Part of our mission is to create AIs that are able to understand physical spaces and physical cyber-physical systems. Now our goal is to build platforms that other people can build end-user solutions on.” (Watch the video from Samsung Electronics’ Seokjin Youn here).
To better meet chipmakers’ priorities and needs, Ricky To, senior manager of data & digital delivery at Exyte, called for a change in mindset when building factories. “Compared to the software industry, which is very iterative, the fail-fast approach is really taking off,” he said. “But in the building or construction industry, we usually iterate when a project is completed.”
By shifting the approach from failing slowly to failing fast, factory builders can shift the majority of change risk from late to early stages while reducing material waste and costs associated with delays.
What belongs in a factory?
According to Deloitte, a state-of-the-art fab capable of producing 20,000 to 25,000 wafers per month costs between $10 billion and $12 billion, with construction costs ranging from $4 billion to $5 billion. It takes two to three years from groundbreaking to the first silicon production, but a relatively small part of the manufacturing facility is the cleanroom facility of the fab itself. Deloitte also estimates that 6,000 to 8,000 workers are needed to build such a fab.
The tight labor market also means that the semiconductor industry is competing with many other industries for skilled workers, especially when other sectors of the economy are booming at the same time. “Demand is increasing,” said Exyte’s To. “Everything from communications systems to automotive to computers and storage systems is in demand. That means we need more facilities and more resources. And of course that also leads to higher costs and higher energy consumption. That impacts the construction industry because basically we’re going to need more people with the right skills, more equipment and software.”
SEMI estimates that more than 60 new factories will begin construction by 2030, but capital spending is typically closely tied to global economic conditions, which explains delays in tooling deliveries after groundbreaking announcements. In addition, since operating costs (tooling installation) make up the majority of a factory’s costs (50%), shell and core construction can remain unused during downturns in the semiconductor industry. SEMI further estimates that $36 billion will be invested in construction projects in 2024.
From a design perspective, the simplest way to understand factory construction is that it is generally done from the bottom up and from the outside in. “The first release packages usually focus on piles/foundations and sub-base, then concrete and external skin including steel framing, then interior fit-out and finally the MEP scope, infrastructure and utilities that serve the toolkit,” said Paul Fullam, Technology Director EMEA, Advanced Facilities at Jacobs Engineering.
Implementation of technical tools
Ideally, large construction sites will use smart technologies with augmented reality (AR) tools (see Figure 2) and autonomous robots. However, huge models, bandwidth limitations and maintaining an uninterrupted network connection can prove challenging.
Fig. 2: Exyte’s structural engineer uses augmented reality to overlay design components and actual piping on site. Source: Laura Peters/Semiconductor Engineering
“Augmented reality is already being used in the field and is a brilliant tool,” said To. “But there are limitations. We have to work offline all the time because our plant models are huge. This means we have to slice the model and then load it onto an iPad. Unfortunately, the hardware itself is the limitation. So when we need to move to another room, we remove the old model and then load the new model. This becomes a very long, tedious process if the internet connection is slow or intermittent.”
He added that robotic scanning, such as with a Boston Dynamics robot, can scan while avoiding obstacles once it is trained and knows the environment. “Robotic scanning is a great technology that has been used for many years, but it is extremely, extremely slow. So we are trying to automate an entire process so that we can bring information into our design model more quickly. Then that scan data is streamed into a platform where it can remove the noise and automatically register and merge the different scans that have been performed. Finally, it moves into a platform where it overlays the scans on the 3D model for comparison to detect any deviations,” To said.
Faster production
The speed at which chipmakers want to scale up production is not a new dynamic, but with the multitude of sensors, connections, data streams and stakeholders involved, the complexity and scale have increased.
“Semiconductor fab construction projects have traditionally been fast-tracked, with overlap between engineering and design activities,” says Jacobs Engineering’s Fullam. “Successful implementation in this environment requires tight integration of design and engineering.” He cites two levers that chipmakers are using to further shorten project timelines: the use of reference designs that can be used from fab to fab, and offsite manufacturing, where assemblies or systems are built off-site and then shipped to the site for integration.
Fig. 3: Reference designs and greater use of offsite manufacturing (OSM) are two strategies aimed at shortening the design and construction timeline. Source: Jacobs Engineering
Semiconductor commercial companies have been using off-site manufacturing (OSM) for at least a decade because it is in their best interest to assemble key parts of their work products in a controlled off-site factory environment.
“The benefits include higher quality, greater safety and fewer restrictions on stacking goods on site,” says Fullam. “But the dynamics are different now: the OSM level our clients are looking for is not 20 to 30 percent of the total contract in terms of labor hours, but 50 percent and more. This creates a new dynamic in design and in the way design and construction are coordinated. When designing for OSM, you need to bring in specialists who understand shipping splits, understand vessel loads when modules are shipped from overseas, and understand rigging, sequencing, etc. as part of the design process. These are things you wouldn’t normally have dealt with during the design phase.”
Results you can see
On semiconductor sites, there is a disjointed digital landscape in the design and engineering toolset that inevitably includes solutions from multiple vendors, including Hexagon, Assemble, Oracle, SQL Server and many others. “Bechtel uses multiple solutions that are used throughout the design, procurement and engineering process. But critically, they are all connected in a data mesh architecture that enables secure, real-time data synchronization – not data sharing and not end-to-end,” Smith said. “This is synchronized across our systems so we understand what is happening where, when and how.”
To avoid confusion, shared visibility is at the heart of intelligent design and construction portals. “One key point is making data visible. There are countless software tools that can make live data on work in progress widely available, and we’ve seen benefits on projects that leverage such real-time data sharing,” said Jacobs’ Fullam. “In terms of BIM, many of our clients and stakeholders, particularly operations staff, process concepts visually. So when you develop a concept model spatially – even at a relatively basic level to explore several different configuration options – stakeholders can make key decisions and options earlier in the design cycle. That reduces the potential for changes later,” he added. “We’ve found the practice of ‘rapid prototyping’ that BIM platforms enable to be really powerful.”
Other panelists agreed. Exyte’s To emphasized that synchronization between the physical twin models as they are created and the large design models can be an extremely slow process. Synchronization requires processing multiple data formats from video recordings, image capture, laser scans, etc.
Diploma
Semiconductor factories are perhaps the most complex and expensive manufacturing facilities in the world. It takes two to three years from groundbreaking to commissioning, known as first silicon. By leveraging some of the smart technologies that enable the chips themselves, the construction industry is working to shorten timelines beyond historical norms. There are growing trends toward synchronous data sharing, AR, and implementation of autonomous robots. Alongside these trends, large-scale hardware/software platforms to process massive data streams could become indispensable.
Related reading
Digital twins target the efficiency of IC tools and manufacturing
Virtual representations improve performance and productivity throughout the design and manufacturing workflow, but deployment can vary in effectiveness and timeframe.
