마이크로칩 8월
This page was machine-translated and may differ from the original. View original

Samsung Electronics and Naver Join Hands for Next-Generation AI Semiconductor

Google 우선 소스Published2022.12.06 15:32


Samsung Electronics Develops Memory Based on Smart SSD, HBM-PIM, and CXL
Naver Optimizes 'AI Model Lightweighting Algorithm' Solution

Samsung Electronics and Naver have joined forces to develop next-generation AI semiconductor solutions.

Samsung Electronics and Naver announced on the 6th that they will cooperate on the development of AI semiconductor solutions to develop and verify AI algorithms and to combine their experience and technology in AI services.

'Hyperscale Artificial Intelligence' is an AI that has evolved one step further than conventional AI, capable of situational learning, thinking, and making judgments like the human brain. As performance improves, data and computational loads are increasing exponentially, but existing computing systems face limitations, leading to the emergence of a need to build new computing infrastructure.

The two companies announced that they would further strengthen their competitiveness in AI technology by developing new semiconductor solutions that can resolve data bottlenecks in AI systems and maximize power efficiency.

Samsung Electronics has developed computational storage technologies such as Smart SSDs, HBM-PIM (Processing-in-memory) and PNM (Processing-near-memory) which embed computational functions into high-performance memory, and CXL (Compute Express Link), based on a next-generation interface capable of efficiently processing large amounts of data.

Processing-in-Memory (PIM) is a technology that implements data computation functions performed by a processor within memory, improving the performance and energy efficiency of AI accelerator systems by reducing data movement between the CPU and memory. A solution utilizing this concept is HBM-PIM.

Samsung Electronics predicted that by confirming that utilizing HBM-PIM increases average performance by approximately twofold and reduces energy consumption by approximately 50% compared to existing GPU accelerators, it will be able to efficiently improve bottlenecks caused by current memory capacity and bandwidth limitations and contribute to power savings.

Naver announced that it is maximizing the performance and efficiency of large-scale AI by optimizing lightweight algorithms—which remove unnecessary parameters from fully trained large-scale AI models or simply adjust weights between parameters—for next-generation semiconductor solutions while operating 'HyperCLOVA'.

Han Jin-man, Executive Vice President and Head of Strategy Marketing at Samsung Electronics’ Memory Business Division, stated, "Through collaboration with Naver, we will develop optimal semiconductor solutions capable of resolving memory bottlenecks in massive-scale AI systems," adding, "We will expand our market-leading next-generation memory lineup, including PIM and computational storage, through semiconductor solutions that reflect the needs of AI service companies and users."

Jeong Seok-geun, Head of Naver Clova CIC, stated, "By combining the knowledge and know-how Naver has acquired while servicing HyperClova with Samsung Electronics' advanced semiconductor manufacturing technology, we will be able to create unprecedented new solutions capable of solving the challenges facing the latest AI technologies." He added, "Starting with this collaboration, Naver will further expand the scope of its technology and continuously strive to strengthen the competitiveness of domestic AI technology."r />
본 기사에 대한 정정·반론·추후보도 청구는 보도 청구 안내를, 그간 게재된 보도문은 정정·반론보도 모아보기를 참고해 주세요.
김예지 기자
김예지 기자