09 Aug 2022 - Joe Zott
What do the 2020s signify for the advanced process semiconductor industry, given recent technological, fab, and government funding announcements?
A transition is undoubtedly taking place in this industry, which currently has three fab players: TSMC, Samsung, and Intel, alongside a vast infrastructure of innovators and suppliers.
The availability of semiconductor process nodes at the most advanced level has plateaued over the last decade. However, the combined forces of technological innovations, product demand (and the capacity to pay for them), and government-supported capex programs ensure that growth continues.
Innovations in Semiconductor Process Nodes
In its Q4 2021 results, TSMC revealed that semiconductor wafers based on 5-nanometer (nm) and 7nm processes accounted for half of its revenue. The 7nm wafers generated 27% of total wafer revenue, while the 5nm chips contributed 23%. TSMC also stated that wafers manufactured using advanced technologies (including process nodes sized at 7nm and below) made up over 50% of the company’s total wafer revenue.
TSMC is expected to allocate a significant portion of its capex to these advanced processes this year. Tom’s Hardware estimates that 70% to 80% of TSMC’s planned 2022 capex will be directed toward the development of advanced processes, particularly on 2nm and 3nm nodes, following the company’s substantial investment in the 5nm process last year.
Samsung and Intel have made similar announcements, though many challenges lie ahead.
Despite these challenges, optimism prevails.
With volume production of 3nm starting later this year, TSMC is focusing on the 2nm generation, driven by demand for high-performance computing chips. Advanced technologies below 7nm account for half of TSMC’s Q1 revenue of US $7.6 billion, exceeding expectations. All six product areas increased in the first quarter.
Semiconductor Process Node Roadmap
After examining various advanced technology roadmaps, the high-level outlook appears to be:
Smartphones
I also researched what semiconductor process has been used for mobile phones over the last decade and what is likely to be used in next decade.
The smartphone trendline will largely follow the historical trendline of the past few decades, but by 2026, this trend will decelerate as process node innovation slows.
Smartphones will likely remain on the cutting edge, but processors may increasingly take advantage of leading-edge process technology (possibly in response to Apple). Apple has already started to use multidie/chiplet technology, and by the mid-to-late 2020s, chiplets will likely be increasingly necessary to enhance performance.
On January 4, 2023, MacRumors reported that Qualcomm and MediaTek had not yet decided whether to join the 3nm camp this year, despite both hoping to keep up with Apple’s process upgrade for their flagship mobile SoCs.
Other technology factors
A couple of things are not apparent in a simple semiconductor process description: memory and integration (chiplet or PCBA).
PCBA
While I have worked extensively with semiconductors over the years, I consider myself more of a product-focused individual. I have been examining existing products to understand what advanced semiconductors look like from a product perspective.
Smartphones are at the cutting edge of process technology, but there is not much to observe in this area. Most advanced smartphones, like the Samsung S22, are just a few advanced chips on a small PCB with minimal surrounding support.
GPUs, however, provide much more interesting examples.
Chiplet
Chiplet technology is becoming increasingly important as a means to maintain performance growth in the face of slowing process node advancements. Chiplets allow for the combination of multiple smaller dies to create a single, more powerful and efficient chip. This modular approach can help companies optimize performance and power consumption while reducing manufacturing costs and risks.
Integrated memory
Integrated memory is another critical aspect of advanced semiconductor technology. As chip designs become more complex and require faster access to memory, integrating memory directly into the chip can offer significant performance improvements. This can be achieved through various techniques, including high-bandwidth memory (HBM), which stacks memory dies vertically, providing a more efficient and faster connection to the processor.
Financial Implications
The advanced technology comes with financial costs, reflected in both capex and opex.
Capex
TSMC’s CEO, CC Wei, stated that the company expects its HPC platform to be its strongest growing platform in 2022 and the largest contributor to its growth. The 2022 capital budget remains steady at between US$40 million and $44 billion, although TSMC expects capacity to remain tight throughout the year.
Similar sentiments have been echoed by Intel, Samsung, and others in line with their growth plans.
The current factory capex level is astonishing when considering the proportion of semiconductor factories in the worldwide capex investment for EVERYTHING. The investment levels appear to be continuously increasing.
It remains to be seen how much capital the world can invest in advanced semiconductor fabs. The most advanced process node semiconductor fabs have represented about half of the total semiconductor fab capex in the past decade, and the trend has been rising in recent years. With TSMC, Samsung, and Intel’s 2023 semiconductor capex exceeding $100 billion, the rising capital cost of semiconductor fabrication is evident.
Opex
The price per wafer continues to increase, adding to the operational expenses of manufacturing advanced semiconductor devices. It is important to analyze the various factors that contribute to the increasing operational expenses in the semiconductor industry, especially for advanced process nodes.
Increasing Complexity
As process nodes shrink, the complexity of manufacturing increases, requiring more sophisticated tools and techniques. This results in higher costs for equipment, maintenance, and the necessary research and development to stay competitive. The increased complexity also leads to longer manufacturing cycles, impacting throughput and operational efficiency.
Higher Yield Losses
Advanced semiconductor manufacturing is highly sensitive to defects, with smaller feature sizes making it more challenging to achieve high yields. Yield losses occur when a portion of the manufactured wafers contains defects, rendering them unusable. This can significantly impact operational expenses, as the cost of wasted materials and lower throughput can be substantial.
Skilled Workforce
The demand for a highly skilled workforce in the semiconductor industry is growing as the technology becomes more complex. Companies require engineers and technicians with specialized knowledge and expertise to maintain and operate the advanced tools and equipment needed for smaller process nodes. This drives up labor costs, as the demand for skilled workers often outpaces supply, leading to increased competition for talent and higher wages.
Energy Consumption and Environmental Concerns
Advanced semiconductor manufacturing requires a significant amount of energy to power the complex equipment and maintain the stringent environmental conditions necessary for cleanroom facilities. This results in higher utility costs, contributing to the operational expenses.
Moreover, there is increasing pressure on the semiconductor industry to address its environmental impact and adopt sustainable practices. This may lead to additional expenses related to implementing and maintaining eco-friendly technologies, waste management, and regulatory compliance.
Cost of Quality and Reliability
As semiconductor technology advances, there is an increasing demand for high-quality and reliable products, especially in applications such as automotive, aerospace, and medical devices. This requires additional investment in quality control
Conclusion
The price per wafer continues to increase, adding to the operational expenses of manufacturing advanced semiconductor devices.
