Konan Technology to Demonstrate LLM on Heterogeneous AI Semiconductor Platform

Konan Technology, an artificial intelligence (AI) software company, announced on April 29 that it has been selected as an implementing company for the "Development of Distributed Inference and Model Optimization Technology Based on Heterogeneous AI Semiconductors" project, which is part of the K-Cloud Technology Development Project promoted by the Ministry of Science and ICT.

The total project budget is 10.4 billion KRW, and the project will be carried out for approximately 4 years and 9 months, until December 2029. The lead research and development institution is the Electronics and Telecommunications Research Institute (ETRI), and the joint research consortium includes Rebellions, Seoul National University, and the Korea Advanced Institute of Science and Technology (KAIST).

Recently, demand for large language model (LLM)-based services has surged, but high inference costs have become a limiting factor. To address this, the importance of optimization technologies and distributed software development that utilize various AI semiconductors such as NPUs and PIMs is increasing.

The goal of the project is to secure efficient distributed inference and model optimization technologies in environments that integrate various types of AI semiconductors. The project will develop a service framework that enables flexible execution of AI models in heterogeneous AI semiconductor environments, as well as an integrated demonstration service based on LLM-RAG that can operate on actual user devices.

Konan Technology will be responsible for the integrated demonstration service of LLM-RAG distributed inference in heterogeneous AI semiconductor environments. Based on its technological expertise in AI software and experience in LLM and RAG development, the company aims to secure a technological foundation that enables stable AI service operation across diverse semiconductor environments. Through demonstration cases of high-performance distributed inference architectures, the company is expected to contribute to enhancing the self-reliance and competitiveness of domestic AI technology.

Oh Changmin, Executive Director of the Language and Speech Research Center at Konan Technology and the project’s research and development lead, said, "It is highly meaningful to participate in a core project for the technological self-reliance of domestic AI infrastructure," adding, "We will realize advanced demonstration services in heterogeneous AI semiconductor-based inference environments and contribute to the commercialization of next-generation AI infrastructure."

Konan Technology will introduce new generative AI products, including Konan LLM and Konan RAG-X, at the "2025 Konan Technology AI Showcase - Media Briefing" to be held on May 13.