Unleashing the Power of Text Mining: Identifying AI Technologies for the Korean Semiconductor Industry

By Samudrapom DamReviewed by Susha Cheriyedath, M.Sc.Jul 6 2023

In an article in the press with the journal World Patent Information, researchers from Korea demonstrated the feasibility of using the text mining method to identify promising artificial intelligence (AI) technologies for the Korean semiconductor industry.

Background

The importance of the semiconductor industry is increasing rapidly with the recent development of critical technologies, such as robots, AI, autonomous vehicles, and the Internet of Things (IoT), that are driving the fourth industrial revolution. Countries such as China, Taiwan, and the United States are investing significantly in semiconductor research and development (R&D).

South Korea is a leading semiconductor supplier of memory devices in the world. Thus, semiconductor exports significantly influence Korean national competitiveness, which necessitates the development of a robust semiconductor value chain in the country with foreign and domestic partners.

However, small- and medium-sized enterprises (SMEs) in South Korea lack sufficient competitiveness as large companies in the Korean semiconductor memory market owing to limited R&D activities and resources. The establishment of an R&D roadmap for SMEs is crucial for Korea to achieve global competitiveness in the semiconductor industry.

Semiconductor companies maintain high levels of quality and precision by stabilizing the manufacturing process. AI technologies stabilize the memory semiconductor production that uses ultrafine processes.

AI technologies can also improve the yield and efficiency of the semiconductor manufacturing processes and facilitate innovation in semiconductor facilities. Thus, SMEs can gain a competitive edge over time and improve productivity using machine learning (ML)/AI technologies.

Several semiconductor device manufacturers are currently not using ML/AI technologies to create business value. However, the integration of ML/AI in the semiconductor value chains, including production, sales, R&D, and design stages, is becoming necessary due to the rising competition in the semiconductor market.

The integration of ML/AI technologies requires significant investments, which can present a risk to SMEs as a failed investment can adversely impact their survival and competitiveness. The investment risk can be reduced by precisely identifying the emerging ML/AI technologies that offer sustainable growth opportunities.

Text mining to identify AI/ML tech for the semiconductor industry

In this study, researchers utilized the text mining method to identify the critical emerging ML/AI technologies based on patents that SMEs can adopt and provide insights that will allow the SMEs to establish a roadmap for innovation. They analyzed the registered patents to identify the semiconductor production process characteristics and R&D trends in the semiconductor industry. The research process was classified into data analysis and collection and emerging technology recommendation-generation phases. Patent titles and IPC codes were considered in the analyses.

Initially, patent specifications only from Korea were collected from the Korea Intellectual Property Rights Information Service database. No overseas patents were considered, as Korea is a leader in the semiconductor memory industry in the world.

Patents concerning the application of ML/AI technologies in the semiconductor industry that do not hold exclusive ownership were considered in this study. More than 3560 patent specifications on the semiconductor industry and AI technology were identified using the database.

Subsequently, a network analysis of the ML/AI technologies was performed using the obtained data. International Patent Classification (IPC)-based network analysis was selected for patent network analysis (PNA) as it provides a standard code that summarizes the technical classification of patents.

At least one IPC code is required in a patent specification. Researchers analyzed the co-occurrence network of technology topics based on the IPC codes described in the collected patent specifications.

Emerging technologies were identified based on patent titles as they can reflect richer technology concepts devised by the applicant /inventor. A latent Dirichlet allocation (LDA)-based topic analysis was performed for patent title analysis to analyze existing AI technologies utilized by semiconductor manufacturers.

LDA is a topic modeling technique that allows document-wise topic distribution and topic-wise word distribution estimation. LDA was performed to overcome the limitations of standardized IPC codes.

Significance of the study

The results from the text mining and network analysis indicated that the application of deep neural networks (DNNs) was crucial in the semiconductor industry. DNNs affected different aspects of semiconductor R&D, including artificial neural network (ANN) parallelization for improved reasoning and learning abilities.

AI technology was actively investigated for monitoring semiconductor manufacturing and plasma etching processes. In the plasma etching process that controls extremely small molecular and atomic units of etched gas particles, AI displayed “nonlinearity” between process parameters and delivered results that exceeded the easily predictable range.

ML modeling and effective input data refinement were studied to address this issue. The inclusion of physics models into current statistical-based ML models was proposed as one of the alternatives. This approach suggested the need to collect smart data based on understanding physical phenomena instead of only using big data. Moreover, technology convergence was identified among etching technology, smart factories, visualization, and virtual reality.

To summarize, the findings of this study can effectively assist the Korean semiconductor industry, specifically SMEs, to identify promising ML/AI technologies for the semiconductor value chains and enable the SMEs to establish an R&D roadmap.