Samsung, Naver agree to partner in developing next-gen AI semiconductor solutions

Samsung Electronics Co. and Naver Corp. have agreed to cooperate in developing next-generation AI semiconductor solutions optimized for artificial intelligence (AI) systems.

The two companies announced Tuesday that they have agreed to jointly develop new semiconductor solutions that will solve data bottlenecks and maximize power efficiency in AI systems.

The performance improvement of hyperscale AI leads to an exponential increase in the amount of data and calculations to be processed, but the existing computing system has limitations in improving performance and efficiency, so an emerging need for new AI-dedicated semiconductor solutions.

Accordingly, Samsung Electronics plans to provide semiconductor solutions optimized for super-scale AI systems by overcoming the memory bottleneck through the convergence of next-generation memory solutions.

Samsung has been developing next-generation memory solutions based on smart SSD, which is a next-generation computational storage technology, processing-in-memory in high bandwidth memory (HBM-PIM) and processing-near-memory with built-in calculation functions in high-performance memory, and compute express link (CXL) that efficiently processes large amounts of data.

Naver, which operates a proprietary hyperscale AI HyperClova, plans to maximize the performance and efficiency of the hyperscale AI to suit the next-generation semiconductor solution.

A lightweight algorithm that removes unnecessary parameters from the hyperscale AI model that has completed learning or simply adjusts the weight between parameters is optimized for next-generation semiconductor solutions.

Starting with this collaboration, the two companies agreed to continue working together to diffuse next-generation memory solutions that would support high-performance computing such as computational storage.

Write to Ji-Eun Jeong at jeong@hankyung.com