Will the U.S. CHIPS Act Succeed?
Intel, Micron, Samsung, TSMC and others have obtained U.S. government funding. Polar, a new U.S. foundry vendor, is emerging.
By Mark LaPedus
Over the years, the United States has seen a substantial and alarming decline in overall semiconductor manufacturing capacity—a trend that raises some national security and supply chain concerns in the nation.
But with the help of several semiconductor companies, the U.S. is taking steps to boost its chip production. Intel, Micron, Samsung, TSMC and others are all building new semiconductor manufacturing facilities, called fabs, in the U.S. Plus, a new U.S. foundry vendor, Polar Semiconductor, is emerging.
These and other companies have obtained U.S. government funding. This funding program, called the CHIPS and Science Act, is designed to accelerate the development of fabs and chip production in the U.S. Still, it will take several years before the U.S. reverses a decline in overall chip manufacturing capacity, which is an important part of the semiconductor industry.
Semiconductors, or chips, are small and critical devices used to power a multitude of products, such as appliances, cars, industrial equipment, medical products, PCs and smartphones. Basically, companies use specialized software tools to design a chip. Then, chips are put into production in large, expensive fabs. Each advanced fab can cost from $10 billion to $20 billion today.
Chips and components mounted on a printed circuit board
Over the years, several nations or regions, namely China, Europe, Japan, Korea, Taiwan, U.S. and others, have developed their own and substantial domestic semiconductor industries. They all have a similar philosophy: Having a robust semiconductor industry is important for one’s competitiveness and security. It also creates thousands of new jobs.
The U.S., for one, has a robust semiconductor industry, but there are some gaps. For example, the U.S. is the leader in designing new and innovative chips. But over the years, the U.S.’ share of worldwide semiconductor manufacturing fab capacity has declined from 37% in 1990, to 19% in 2000, and to 10% in 2022, according to the U.S. Semiconductor Industry Association (SIA) and Boston Consulting Group (BCG).
Companies in the U.S. have built new fabs over the years, but the nation hasn’t kept pace with other regions, namely Asia. In 2022, 82% or more of the world’s chip manufacturing fab capacity was (and still is) located in Asia, according to the SIA and BCG. Moreover, Taiwan, namely TSMC, produces a large percentage of the world’s most advanced logic chips. The Asia region also produces the majority of memory chips, IC packages and other products.
In other words, the U.S. is dependent on Asia and other regions for the production of a large percentage of its chips. That’s problematic for U.S. national security as well as the nation’s competitiveness and the supply chain.
In response, the U.S. government in 2022 launched a new program called the CHIPS and Science Act, which set aside $39 billion in grants for manufacturing and $11 billion in funding for R&D and workforce development. As part of this plan, the U.S. government will award select companies with grants and/or tax credits for use in building new fabs and/or expanding existing chip facilities in the U.S.
Initially, the CHIPS program was mired in red tape. It took the U.S. government a year to award any funding. Recently, though, it has announced a series of funding awards, including:
*Intel, Samsung and TSMC have all been awarded CHIPS funding. This will help bring up leading-edge logic fab capacity in the U.S. These fabs produce advanced logic chips, such as AI devices and processors.
*Micron has obtained CHIPS funding, which promises to boost memory chip production in the U.S.
*BAE, GlobalFoundries, Microchip and Polar Semiconductor have been awarded CHIPS funding. This will be used to expand the production of critical chips at more mature processes.
*Polar will benefit in other respects. The funding, along with equity investments from others, will help transform Polar from an obscure semiconductor manufacturing company into a U.S.-owned foundry vendor.
*Absolics, an affiliate of Korea’s SKC, obtained funding. The company will produce substrates for packaging in the U.S.
“The CHIPS and Science Act has put America on course to significantly strengthen domestic semiconductor production and R&D, but more work is needed to finish the job,” said John Neuffer, SIA’s president and chief executive.
Government funding won’t solve everything. For example, the U.S. suffers from shortages of fab technicians and engineers. It’s unclear how to solve that problem. Then, the initial CHIPS funding awards only covers a small part of the cost to build a new fab. Chipmakers will need to find ways to foot the remaining bill.
Even with CHIPS funding, chipmakers face some risks when it comes to building new fabs. Fabs aren’t built on a whim. Companies won’t add any new or meaningful fab capacity unless business conditions are favorable and there is enough demand for their chips in the market. An under-utilized fab translates into losses. All told, even with the CHIPS Act, there is no guarantee that the U.S. will ever make any meaningful gains in chip production.
Transistors, Moore’s Law
In 1947, Bell Labs invented the transistor. One of the key building blocks in a semiconductor device, a transistor amplifies or switches electrical signals within a chip. By the 1950s, a growing number of new semiconductor companies emerged, hoping to create new business opportunities around transistor technology.
As the industry moved into 1960s, semiconductor companies generally did everything in-house. Companies designed their own semiconductor products, and then manufactured them in their own internal fabs. Nearly every company had fabs back then. Fabs were located and built in a company’s nation of origin.
The early fabs were rudimentary. In 1965, for example, companies produced chips using manual equipment in fabs. Back then, each fab would cost $1 million to build and equip, according to VLSI Research, now owned by TechInsights.
In 1965, chips were processed using 1.25-inch wafers in a fab (1.25-inch is the diameter size of the wafer.) Basically, to make a chip, a semiconductor company obtains a round silicon-based wafer. Then, a company takes the wafer and subjects it to various process steps in a fab. Using this methodology, multiple chips or dies are produced on each wafer.
1965 was notable for other reasons. At the time, Intel co-founder Gordon Moore published his famous paper, predicting that the number of transistors on an integrated circuit would double every year. In 1975, Moore revised this to doubling every two years. (Moore passed away in 2023.)
This observation, called Moore’s Law, became a guiding principle for the semiconductor industry. Following this concept, Intel and others in the 1970s began to introduce a new and more advanced manufacturing processes every 18 to 24 months. A manufacturing process involves the steps and recipes to make chips in a fab.
Then, at the same interval, companies introduced new chips based on the latest process. The goal was (and still is) to double the transistor density for a chip at each new process generation. The other goal is to shrink select feature sizes of a transistor by 0.7x at each generation, making it smaller and faster at each turn. Some refer to this concept as chip scaling.
This in turn enabled new and faster chips at each turn. It also paved the way towards a new era of integrated chips. In 1971, Intel introduced its first processor--the 4004. The processor, which consisted of 2,300 transistors, was manufactured using Intel’s 10μm process. This process had an effective channel length of 10µm between the source and drain in each transistor, according to WikiChip, a technology site.
Around that time, the semiconductor industry began to assign a number for each new manufacturing process. That number is called a node or technology node. The node numbers progress in descending order. (For example, chip companies introduced the 180nm process in 1998, followed by the 130nm process in 2001, 90nm in 2003, and so on, according to WikiChip.)
“The technology node refers to a specific semiconductor manufacturing process and its design rules,” said David Schor of WikiChip on its website. “Generally, the smaller the technology node means the smaller the feature size, producing smaller transistors which are both faster and more power-efficient.”
Moore’s Law helped fuel a new industrial revolution. New and faster chips helped drive emerging markets like PCs in the 1980s and cellphones in the 2000s. In 1982, Intel introduced the 80286 microprocessor, which helped fuel the PC market. Based on a 1.5µm process, the 80286 integrated 134,000 transistors on the same device. Meanwhile, in 1990, Motorola announced the 68040, a 32-bit microprocessor used in Apple’s Mac line. Based on a 650nm process, the processor consisted of 1,200,000 transistors.
Soaring fab costs
On the downside, fab costs began to soar. Generally, fab costs are determined by several factors, including labor cost (10%-15%), material cost (35%-40%) and capital cost (40%-50%), according to a recent paper from the University of California at Berkeley.
In the 1980s, companies saw a spike in all areas. At each node, it became more expensive and difficult to develop a new manufacturing process. It required new and more costly equipment.
On top of that, semiconductor companies migrated to larger wafer sizes, which also contributed to the cost. The industry migrated from 1.25-inch wafers to 4-inch wafers in 1975, and to 5-inch wafers in 1985, according to VLSI Research.
In theory, a company can process twice as many chips by moving to a larger wafer size, thereby lowering its manufacturing costs. But by moving to larger wafers, companies required new and more expensive fab equipment.
The cost to build and equip a 4-inch fab was $9 million in 1975. That cost jumped to $100 million for a 5-inch fab in 1985, according to VLSI Technology.
2-inch (51mm), 4-inch (100mm), 6-inch (150mm), and 8-inch (200mm) wafers. Credit: Wikipedia)
In the 1990s, the industry moved to 200mm wafers. In 1997, the starting price for a 200mm fab was around $800 million, according to UC Berkeley. Then, in the 2000s, fab costs continued to escalate when the industry moved to 300mm wafers and fabs. In 2001, a 300mm fab could cost about $2 billion, it added.
During that period, the manufacturing landscape changed. In 1987, Taiwan’s TSMC emerged and pioneered the pure-play foundry model. In this new and untested strategy, TSMC didn’t design its own chips, but rather it exclusively made chips for other companies.
Other foundry vendors, such as Powerchip, UMC and Vanguard, emerged in Taiwan. Foundry vendors also emerged in China, Europe, Korea, Malaysia and the U.S.
In the 1990s, foundry vendors obtained billions of dollars of capital, enabling them to build new fabs and develop competitive manufacturing processes. At that time, the foundry business took off, as a new and growing crop of fabless semiconductor startups emerged. These startups didn’t have their own fabs. They focused on chip design and relied on foundries for their manufacturing needs.
Foundries also benefitted from another trend starting in the 1990s. At the time, many chipmakers could afford to build new fabs. But others in the U.S. and elsewhere began to shutter most or all of their fabs, and shifted their chip production to foundry vendors in Asia and other regions. For many, it simply became too expensive to build new fabs and develop a manufacturing process every 18 to 24 months.
In 1998, more than two dozen companies could make chips in their fabs based on the 180nm process, which was the most advanced technology back then. “As shrinking becomes more complex, requiring more capital, expertise, and resources, the number of companies capable of providing leading-edge fabrication has been steadily dropping,” WikiChip’s Schor said. “As of 2020, only three companies are now capable of fabricating integrated circuits on the most cutting-edge process: Intel, Samsung, and TSMC.”
Today’s most advanced fabs cost $20 billion each. It’s four times more expensive to manufacture a 7nm chip than a 45nm device, according to a recent U.S. government report. Chip-manufacturing costs for the latest nodes, namely 3nm, 2nm and beyond, are astronomical. Plus, it takes longer to migrate to each node, roughly around 24 to 36 months. In some cases, the price/performance metrics are diminishing at each node.
It's not all doom and gloom. In terms of total sales, the semiconductor market is expected to reach $588 billion in 2024, up 13.1% over 2023, according to the World Semiconductor Trade Statistics (WSTS) organization.
Chip vendors continue to develop new and faster chips for computers, smartphones and other products. Nvidia, for one, recently introduced its new GPUs for high-end computing applications like AI. Manufactured using TSMC’s 4nm process, the GPU consists of a staggering 208 billion transistors on the same device.
At the end of 2022, there were a total of 167 fabs processing 300mm wafers worldwide, according to Knometa, a research firm. Plus, in 2022, there were 224 200mm fabs worldwide, according to SEMI.
Time for the CHIPS Act
Today, there are a multitude of production fabs in the U.S. But over the years, the U.S. share of worldwide fab capacity has dwindled.
China, Korea and Taiwan produce a large percentage of today’s chips. But this region is a geopolitical hotspot. Any disruption in this region could potentially impact the supply of chips. The recent Covid pandemic was especially disruptive. During that period, there were bottlenecks in the worldwide supply chain, causing acute shortages of chips.
Suddenly, the U.S. government realized that chip manufacturing is critical to America’s security and supply chain. China, Europe, Japan and Korea came to the same conclusion regarding their respective nations.
For its part, the U.S. government in 2022 enacted the CHIPS Act, a multi-pronged program that promises to boost semiconductor R&D and manufacturing in the U.S.
Even before the CHIPS Act, Intel, Micron, Samsung, TSMC and others were separately building new fabs in the U.S. These fab projects are still in various stages of development. When the CHIPS Act was enacted in 2022, semiconductor companies hoped to obtain funding to accelerate their U.S. fab efforts. Initially, though, the CHIPS program was mired in red tape, causing delays in the funding process.
The tide is turning. To date, the CHIPS program office has announced $29.5 billion in grant awards and up to $25.1 billion in loans to nine companies across 18 projects. All of this sounds impressive. But these funds will only cover a small part of the overall cost to build a fab.
Still, it’s a start. With the help of the projected funding, U.S. fab capacity is expected to increase by 203% from 2022 to 2032, according to the SIA and BSG. In total, the U.S. is expected to increase its share of global fab capacity for the first time in decades, growing from 10% today to 14% by 2032, according to the firms. In addition, U.S. fab capacity for advanced logic processes alone will grow from 0% in 2022 to 28% by 2032, they added.
Today, meanwhile, the U.S. government is providing CHIPS funding in several areas. This includes four key manufacturing segments--mature logic; advanced logic; memory; and packaging.
More logic fabs
In a fab, mature logic processes are used to make chips at the 22nm node and above. These chips include analog ICs, power semiconductors, RF devices and others. These chips are important.
Last year, for example, BAE was awarded $35 million in CHIPS funding to upgrade its mature-node fab in New Hampshire, which produces chips for the F-35 fighter jet and other defense programs.
Earlier this year, Microchip was awarded $162 million, which will be used to expand its mature-node fabs in Colorado and Oregon.
Then, in February, GlobalFoundries was awarded $1.5 billion in CHIPS funding. The funds will be used to build a new fab capacity. GlobalFoundries, a foundry vendor, produces chips at the 14nm/12nm nodes and above.
Recently, Polar Semiconductor obtained $120 million in CHIPS funding as well as a $75 million investment from the State of Minnesota. Polar has a 200mm fab, which makes devices based on mature processes. It primarily makes chips for its two shareholders--Sanken and Allegro.
In addition, Polar received a $175 equity investment from Niobrara Capital and Prysm Capital, enabling Polar to transition to a U.S.-owned merchant foundry. Sanken and Allegro will become minority shareholders in Polar, and will continue their relationship as foundry customers.
Making the most advanced chips is also important for the U.S. Even though Moore’s Law is slowing, leading-edge foundries continue to develop new, advanced processes. Advanced chips involve 10nm processes and below. Today’s most advanced chips are based on 3nm processes with 2nm in R&D.
The U.S. hopes to regain its leadership here. U.S.-based Intel was once the leader in process technology until it stumbled in the mid-2010s. It was then surpassed by TSMC and Samsung.
Today, Intel is accelerating the development of its advanced manufacturing processes, hoping to regain the lead in the arena. Intel is also building new fabs in Europe and the U.S. Intel, however, recently delayed the schedule of its new $28 billion fab project in Ohio. The fab was expected to be competed in 2025. 2026/2027 is the new target.
In a major boost, Intel in March 2024 was awarded $8.5 billion in CHIPS funding. Intel will use the funds to jumpstart its U.S. fab efforts.
Meanwhile, TSMC continues to build new and advanced fabs in Taiwan. In 2020, TSMC announced plans to build a new advanced fab in Arizona, representing its most advanced facility outside of Taiwan. That fab will produce 4nm chips.
It hasn’t been easy to get the U.S. fab off the ground. Early on, TSMC discovered that there aren’t enough skilled fab and construction workers in the U.S. Then, the company also clashed with labor unions in Arizona.
Last year, TSMC delayed the production schedule of the fab from 2024 to 2025. Then, citing some delays in obtaining CHIPS funding, TSMC pushed out the schedule for its second fab in Arizona from 2026 to 2027/2028.
Then, in April 2024, TSMC obtained $6.6 billion in CHIPS funding. Besides the two fabs on the drawing board, TSMC also plans to build a third fab in Arizona by the end of the decade.
Meanwhile, Samsung owns and operates fabs in China, Korea and the U.S. Last year, citing a delay in CHIPS funding, Samsung delayed its new $17 billion fab project in Taylor, Texas.
Then, in April of 2024, the U.S. government awarded Samsung with $6.4 billion in CHIPS funding. With the help of the funding, Samsung plans to build two new leading-edge logic fabs, an R&D fab, and an advanced packaging facility in Taylor.
Memory and packaging production
The U.S., meanwhile, also wants more domestic memory production. In April, Micron obtained $6.14 billion in CHIPS funding. With the help of the funding, Micron hopes to build new DRAM fabs in New York and Idaho.
Chip-packaging is also important. Once a chip is produced in a fab, the device is encapsulated in a package. The package protects the device.
In one advanced application, multiple and different dies are assembled in a package. Thus, the package becomes a functional electronic system. That’s just one of many applications in packaging.
The majority of the world’s chip-packaging production capacity is located in Asia. A tiny percentage of this is located in North America.
In the first step to reverse those trends, the U.S. government recently awarded $75 million to Absolics. The company plans to build a substrate manufacturing facility in Georgia. Substrates are a key component in packaging.
Conclusion
Still, the question is clear: Will the CHIPS Act succeed? Some believe it will. Others say it will fail. Right now, it’s too early to say, as it’s just getting off the ground.
There are several ways to measure success. If all goes to plan, the U.S. could capture 14% of the world’s fab capacity by 2032. That’s nothing to brag about, but at least the U.S. isn’t losing ground.
The CHIPS program could put a feather in its cap if the U.S. can achieve large-scale production of chips at the world’s most advanced logic processes, that is, 3nm and beyond. Bringing up more U.S. memory and chip-packaging production are desirable.
Just reaching one major milestone is a good start.