As semiconductor processes grow ever more complex and miniaturized, yield-improvement technologies using artificial intelligence (AI) are drawing attention. Typical examples include using AI image analysis to identify microscopic defects on wafer surfaces in real time, or training AI on the many variables arising in complex processes and then optimizing through real-time monitoring. Adoption of AI is expected to become more active among foundry (contract chip manufacturing) companies that are struggling to raise yields. - Analyzing semiconductor data with machine learning and deep learning According to the industry on the 15th, Korean AI solution company Crosssight is developing AI-based software (SW) that improves yield in semiconductor processes. Semiconductors, which integrate countless microscopic circuits, operate stably only when the chips are uniformly aligned. For this, overlay, which indicates the degree of misalignment between the upper and lower circuit layers, and critical dimension, which refers to the error in circuit width, must be examined. However, because of cost, conventional metrology examined only about 1% of the total. Crosssight has developed a solution that predicts errors by analyzing data from semiconductor equipment based on machine learning algorithms. The company explained that with it, wafer overlay and critical dimension can be predicted in bulk down to 0.2 (nm; one nanometer is one billionth of a meter). The company plans to apply the solution first to the exposure (lithography) process, and intends to introduce AI to the etching, thin-film deposition, cleaning, and polishing stages by 2026. Another Korean AI solution company, AIBIZ, developed the platform 'DutchBoy', which raises yield by analyzing data generated in semiconductor manufacturing processes with deep learning algorithms to detect anomalies and identify their causes. With this technology, AIBIZ received the Minister of Science and ICT Award at the '2024 Software Industry Day' ceremony. - Good semiconductor equipment maintenance also improves yield AI-based yield-improvement technologies are also being applied on-site overseas. Lam Research of the U.S., one of the world's top four semiconductor equipment makers, recently launched 'Dextro', an AI robot designed to optimize maintenance of wafer fabrication equipment. Because nanometer-level precision is required, semiconductor manufacturing equipment draws on advanced physics, robotics, and chemistry. For this reason, wafer fabs have hundreds of process tools installed that require regular, sophisticated maintenance. Dextro is designed to improve equipment cost efficiency by repeatedly performing critical maintenance tasks that require submicron (under 1 ㎛) precision. Dextro has already been deployed in fabs around the world, helping improve yield by minimizing production variability. Chris Carter, Vice President of Lam Research's Customer Support Business Group, stressed, "Dextro was built to work alongside engineers in cutting-edge fabs," adding, "With precision and repeatability beyond human limits, it performs complex maintenance tasks and has improved manufacturing yield." Kim Young-joo, Executive Vice President and Head of the Memory Etch Technology Team at Samsung Electronics, assessed, "When semiconductor manufacturing equipment needs maintenance, work must be done efficiently to reduce equipment downtime and prevent wasted cost," adding, "Dextro's maintenance contributes to yield improvement." Germany's chemical materials company Merck teamed up with U.S. AI software company Palantir to unveil 'Athinia', a platform that can stabilize semiconductor production, aiming to improve yield. Athinia uses AI algorithms to analyze data gathered by all participants in the semiconductor manufacturing ecosystem. The analyzed data is shared through the cloud with various partner companies in the supply chain to raise productivity. - AI increasingly used in process control and defect analysis AI predicts defects that may occur in fabs through large-scale data analysis. For example, it uses image analysis technology to identify microscopic defects on wafer surfaces in real time. This raises the accuracy of yield prediction and reduces the defect rate. AI also plays an important role in process control. In semiconductor manufacturing, various variables influence one another across complex process steps. AI learns the relationships among the variables and, through real-time monitoring, finds the optimal conditions at each step to improve the process. AI is also increasingly used in defect analysis. It learns in advance the many types of defects that occur in fabs, recognizes their patterns, and responds to problems beforehand. In particular, it uses deep learning-based image analysis technology to automatically classify wafer defects and provides data for process improvement on that basis. In the past, processes were so mature that of every 100 units produced, more than 99 were good, but recently, as semiconductor miniaturization has advanced to extremes, yields have been falling. For the 4-nanometer node, Samsung's foundry yield has reportedly only recently reached 70%, while TSMC's exceeded 80% last year. Considering that a wafer costs roughly as much as a mid-size sedan per sheet, calculations show that improving yield by just 1% generates trillions of won in additional sales. This is why attempts to improve yield using AI are expected to become more active. The related market is also expected to grow. Market Research Future, a market research firm, projected that the industrial AI market will grow from about $2 billion in 2022 to $90 billion in 2032. Reporter Kang Kyung-joo qurasoha@hankyung.com