The "Yield Dilemma" in the Semiconductor Industry

In the post-Moore era, the semiconductor industry is facing a tough "yield" battle. As chips become more complex and transistor density doubles, the difficulty of chip manufacturing yield management is also growing exponentially, especially for wafer fabs pursuing advanced processes, which can be a fatal blow.

Yield: The Lifeline of Semiconductor Factories

Yield is the core competitiveness of semiconductor factories, also known as the "lifeline" of semiconductor factories.

The semiconductor yield is the percentage of the total number of actual chips produced to the maximum number of chips (integrated circuits) on a wafer. In other words, yield is the ratio of the actual production quantity to the total input quantity. The higher the yield, the higher the productivity, and yield directly affects costs and production capacity utilization. Therefore, improving yield is crucial in the semiconductor industry. At the same time, yield is also an important indicator for evaluating the competitiveness of enterprises, directly reflecting the stability of the manufacturing process and the reliability of product quality.

In the logic chip field, guided by Moore's Law, the most advanced process that can be mass-produced currently is 3 nanometers. In this field, TSMC and Samsung are the two major players, engaging in fierce competition in this 3nm battle.

Samsung announced mass production of 3nm in June 2022, a few months ahead of TSMC. Samsung is the first in the industry to introduce Gate-All-Around (GAA) technology in 3nm, while TSMC still uses FinFET transistor architecture. However, Samsung has encountered setbacks in yield. Although Samsung recently denied reports from Korean media ZDNet Korea that Samsung's 3nm yield is below 20%, due to yield issues, Samsung has indeed lost a large number of customers.

Google's Tensor processor was outsourced to Samsung Electronics' foundry division before the fourth generation, but starting from the fifth generation with the introduction of 3nm technology, it switched to TSMC. Samsung's own Exynos 2500 chip is also plagued by yield issues. On June 25, well-known analyst Ming-Chi Kuo posted on the X (formerly Twitter) platform, stating that Samsung's self-developed Exynos 2500 processor 3nm chip has lower yield than expected, leading to shipment delays.

It is expected that this year, Fabless companies and major IT factories, including smartphones, servers, artificial intelligence, etc., will start using 3nm as the main process. In the face of this important evaluation metric of yield, due to TSMC's leading advantage, it is expected to receive the majority of 3nm orders from major factories, which may further widen the market share gap between TSMC and Samsung Electronics However, even TSMC, currently, the yield of its 3nm process is not high compared to the previous generation. TSMC is actively improving the yield of 3nm as well.

The low yield issue is not limited to the logic chip field, storage chips also face the same challenge. Generative AI is consuming a large amount of HBM storage chips, but yield has always been one of the obstacles to large-scale production of Nvidia GPU chips. According to Reuters, Samsung's high-bandwidth memory HBM3 chip production yield is about 10% to 20%, while SK Hynix's HBM3 yield can reach 60% to 70%. In the traditional DRAM storage sector, Samsung's fifth-generation 10-nanometer (1b) process DRAM yield has not reached the general industry target of 80% to 90%, forcing Samsung to establish a dedicated working group last month to address this issue.

Next, generative AI is still advancing rapidly, with an insatiable demand for chip computing power. In order to obtain higher performance and computing power AI chips:

On one hand, logic manufacturers are actively breaking through to the 2nm process. TSMC and Samsung are currently planning to start mass production of 2nm in 2025, while also focusing on advanced packaging technologies such as FOPLP (Fan-Out Panel Level Packaging).

On the other hand, storage manufacturers are also heavily investing in the research and development of the next generation of HBM (HBM4), with various advanced technologies under consideration, such as hybrid bonding.

It is foreseeable that yield will once again become a major challenge. According to the "Korea Economic Daily" report, SK Hynix is recruiting dozens of positions related to HBM. In its recruitment notice, it is hoped to find chip experts who can improve foundry processes and test logic chips to increase the yield of HBM chips.

Low yield is a common problem faced by semiconductor manufacturers and a bottleneck restricting the development of the chip industry. From an economic perspective, improving chip yield can also be seen as another continuation of Moore's Law. How to improve yield in the context of constantly emerging new technologies has become a key issue that every semiconductor manufacturer must address.

Why is it so difficult to improve yield?

To improve yield, it is necessary to first clarify the reasons for the low yield. There are many factors that affect the yield of semiconductor manufacturing, but they are nothing more than "people, machines, materials, methods, environment, and measurement."

Specifically:

People: The technical level and operational norms of operators directly affect the stability of the production process.

Machines: The precision and stability of semiconductor manufacturing equipment directly affect the microstructure and size of the chips.

Materials: The purity and consistency of raw materials affect the quality of the chips.

Methods: Process parameter control and operational accuracy are crucial to yield

Environment: The environmental factors in the production workshop can affect the chip manufacturing process.

Testing: The performance of testing equipment and testing methods affect the accuracy of yield evaluation.

However, as the chip manufacturing process continues to advance towards advanced processes, the factors affecting yield are becoming more and more "elusive":

On the one hand, with the advancement of technology, the number of process steps and types of materials involved in the process increases, the types of defects on the chip become more and more diverse, and the causes of defects are complex, making analysis and troubleshooting difficult. On the other hand, the continuous introduction of new processes, new materials, and new equipment also brings new challenges to yield improvement. In addition, improving yield requires a large investment of funds and manpower, including research and development investment, equipment investment, talent cultivation, etc. Moreover, some advanced detection and analysis methods also require high costs.

This is why with each advancement in the industry, it becomes increasingly difficult to improve yield.

Industry Struggles to Improve Yield

To improve the yield of semiconductor manufacturing, industry participants are making every effort to optimize and innovate in various aspects.

Reducing pollutants: The semiconductor manufacturing process involves a complex series of multiple process steps that require a large amount of gases and solutions. As the "blood" of semiconductor manufacturing, the purity and reliability of these gases/liquids largely determine the performance quality and yield of semiconductor devices. According to industry estimates, pollution can lead to yield losses of up to 50%.

Pollution control is a crucial part of the semiconductor manufacturing process. If there are pollutants in the gases and solutions, such as particles, impurities, etc., it can lead to defects in the chips, reduce yield, and even cause chip scrap. Therefore, effectively removing these pollutants and ensuring the purity of the "blood" is a major challenge faced by semiconductor manufacturing.

To meet the industry's urgent need for advanced processes and the challenge of improving yield, Pall Corporation, a global expert in filtration, separation, and purification solutions, has been increasing its investment in recent years. Last year, Pall's Gaskleen and Profile II dual production lines at its Beijing factory were officially launched. In June of this year, Pall announced the start of production at its Singapore factory, adding two important filter product lines for lithography and wet chemical processes, which can provide strong support for the current advanced process nodes in the manufacturing of logic and memory chips.

Pall can provide a full range of semiconductor filters, including CMP filters, lithography filters, gas filters, gas purification systems, ultra-pure water filters, wet etch filters, etc. These powerful filters have advanced nano-level impurity interception capabilities, which can remove particles, impurities, and other pollutants in different key processes to ensure high-quality and high-yield chips. In addition, efficient filtration technology can reduce damage to equipment by pollutants, extend the life of equipment, and reduce maintenance costs. Process Optimization: Process optimization is the cornerstone of improving yield. By continuously improving process technology, enhancing the precision and controllability of production processes, and reducing the occurrence of defects. For example, adopting advanced processes such as EUV lithography technology, 3D packaging/Chiplet/CoWos, and various advanced packaging technologies.

Integrating AI and Big Data Analysis Technology: In the semiconductor manufacturing process, there are a large number of process parameters and production data. Leveraging emerging technologies such as AI, big data, etc., to achieve higher yield and less input has become an industry consensus. The intelligentization of data management in semiconductor factories is a trend.

Engineering Intelligence (EI), Computer Integrated Manufacturing (CIM), Manufacturing Execution Systems (MES), and other industrial software are important tools in semiconductor manufacturing. With the integration of AI and big data technology, these software are continuously upgrading to provide more intelligent functions for semiconductor factories. In the intelligent manufacturing mode, the production process becomes more transparent and controllable, significantly improving production efficiency and product quality.

Automation Upgrade: Reducing manual operation links and lowering the defect rate caused by human errors are important means to improve yield. The introduction of Automated Guided Vehicles (AGV), overhead cranes, and Autonomous Mobile Robots (AMR) is accelerating the semiconductor manufacturing industry towards a new era of intelligent manufacturing. The application of automation equipment can effectively reduce manual operation links, liberate workers from heavy physical labor and repetitive work, thereby reducing the defect rate caused by human errors, and improving production efficiency and product quality.

It can be seen that improving semiconductor manufacturing yield is a complex system engineering that requires semiconductor manufacturers to innovate and improve from various aspects such as materials, processes, equipment, and management in order to effectively reduce the defect rate and enhance product competitiveness.

Conclusion

Today, yield improvement has become one of the core driving forces for the development of the semiconductor industry. We also see filter manufacturers like Pall actively expanding production, continuously improving the level and capacity of filtration technology, escorting the application of advanced processes and the improvement of yield.

Believing that through full industry chain collaboration, continuous tackling of difficulties, the semiconductor industry will definitely be able to continuously improve the level of yield, promote high-quality industry development, and contribute more to global economic growth.

Author: Du Qin, Source: Observation of the Semiconductor Industry, Original Title: "The Tragedy of Yield in the Semiconductor Industry"