CAD & VLSI 연구실 CAD & VLSI Lab

Software that designs semiconductor chips

CAD (Computer Aided Design) refers broadly to design carried out with computers and is essential to improving the performance of semiconductor chips. This laboratory develops software that analyses and predicts circuit performance quickly and accurately, studies automation of system design, and explores yield optimisation through machine learning. It sits at the centre of the loop in which AI designs chips and chips in turn run AI.

데이터 중심 컴퓨팅 및 AI 시스템 연구실 DISCOS: Data-Intensive AI Computing & System Laboratory

System software and data platforms for AI

The performance of an AI service is governed not only by the model but by the system software that stores, moves and processes its data. This laboratory studies performance optimisation of distributed data platforms, vector databases, storage systems and operating systems for large-scale AI and data-intensive applications. Its goal is cross-layer optimisation that raises both throughput and response time for AI services in cloud and distributed environments.

시스템 모델링 및 최적화 연구실 Systems Modeling & Optimization (SMO) Lab

Translating problems into mathematics and finding optimal solutions

The Systems Modeling and Optimization Laboratory establishes method before application. In systems modelling it expresses the system behind an optimisation problem in mathematical language and derives its solution; in intelligent optimisation it studies solutions in which a computer reads its environment, learns, and optimises on its own. Reinforcement learning, simulation-based optimisation and stochastic control are the fields it works in, all of them directly connected to the theoretical foundations of AI today.

지능형 컴퓨터 아키텍처 및 임베디드 컴퓨팅 연구실 Embedded Computing Lab (ECL)

Redesigning computer architecture at the front line of change

Embedded computing applies computer hardware and software to mobile devices, multimedia, networks, automobiles, biomedical instruments and many other areas. Through computer architecture, operating systems and compiler technology this laboratory builds systems tuned to the requirements of each application, and studies embedded systems optimised for deep learning. Every time a new application such as deep learning, big data or the smartphone appears, existing computer architectures are asked to be redesigned, and this laboratory stands at the front of that change.

데이터 중심 컴퓨팅 및 AI 시스템 연구실 DISCOS: Data-Intensive AI Computing & System Laboratory

Redesigning LLM serving from GPU to storage

For large-scale LLM inference, this laboratory studies inference execution and scheduling techniques that maximise GPU utilisation and throughput using token-length prediction and continuous batching. It optimises the placement, movement and eviction of the KV cache and designs cost-efficient LLM serving architectures that meet SLOs. Ultimately it aims to improve both the throughput and the tail latency of cloud-based RAG-LLM services through cross-layer design that links the LLM runtime, KV cache, vector database, file system, storage and GPU scheduler.

머신러닝 시스템 연구실 Machine Learning Systems Lab

Accelerating training where machine learning meets HPC

The Machine Learning Systems Laboratory works at the intersection of machine learning and high-performance computing. It develops parallel processing techniques that accelerate the training of large deep learning models on distributed-memory systems and shared-memory multiprocessor platforms. For modern GPUs it designs matrix and tensor algorithms with strengthened data locality (GEMM, SpMM, SpGEMM, convolution, tensor contraction and others), and studies model compression alongside them.

고성능 인공지능 시스템 연구실 High-Performance AI Systems Lab

On-device LLMs and AI accelerator design

As the compute and memory demands of AI models keep growing, research into AI systems capable of efficient training and inference matters more and more. This laboratory designs high-performance AI systems that handle AI models efficiently on heterogeneous systems mixing CPUs, GPUs and NPUs. On-device LLM inference, neural architecture search, AI accelerator design and multi-dimensional deep learning parallelism for LLMs are its core subjects, and designing algorithm and hardware together is its method.