Samsung Elec strives to widen gap with rivals with advanced memory technology

Samsung Electronics Co., the world’s largest memory chip maker, is accelerating its efforts to cement its memory leadership with Processing in Memory (PIM) and Processing near Memory (PNM), touted as next-generation memory technologies that are expected to ensure the seamless operation of Hyperscale AI.

According to Samsung Electronics on Thursday, the company has recently completed standardization of software required to run it latest memory solution High Bandwidth Memory-Processing in Memory (HBM-PIM).

HBM-PIM is the company’s most advanced memory solution that uses the next-generation memory technology PIM. It was considered the industry’s first programmable PIM solution tailored for diverse AI-driven workloads such as HPC, training and inference, when it was first unveiled last year, according to the company.

Its HBM-PIM is a perfect memory solution for Hyperscale AI because it enables a computer to handle large volumes of data. With the fast migration to digitation, the memory industry needs more advanced architecture that can boost AI computing capabilities.

Using PIM technology allows internal arithmetic process, significantly reducing data movement between CPU and memory. It enhances energy efficiency and artificial intelligence accelerator system performance.

When the HBM-PIM was applied to a GPU accelerator card during a recent Samsung-AMD experiment, the accelerator’s performance was boosted by two times compared with the existing GPU accelerator without HBM-PIM. It also used 50 percent less energy.

Minimizing data movement is possible because the HBM-PIM brings processing power directly to a data storage by placing a DRAM-optimized AI engine inside each memory bank – a storage sub unit. The HBM-PIM can be integrated into existing systems, making its application easier.

Samsung Electronics is also preparing software that supports the technology and plans to unveil the solution next month.

The tech giant’s another big bet is its Compute Express Link (CXL)-based Processing near Memory (PNM).

CXL is an interface used in a computing system, which makes it easy to expand memory capacity and helps efficient use of an accelerator used with processor and memory. It integrates various interfaces into one which allows direct communication between each equipment and memory share.

PNM, like PIM, is a memory technology that reduces data movement between CPU and memory by utilizing memory data arithmetic function. It improves system performance and reduces bottleneck issue between CPU and memory by locating arithmetic function adjacent to memory.

With its relentless efforts to improve its memory technologies, Samsung Electronics is seeking to help accelerate the development of Hyperscale AI, an AI that can replicate human thinking and decision making with its own learning capabilities, and widen the gap with its global peers.

Samsung Electronics has retained position as the world’s largest memory chipmaker since 1998. According to global market research firm Omdia, Samsung Electronics accounted for the world’s largest 43.4 percent of DRAM and 33.3 percent of NAND flash memory market in the second quarter ended June.

Shares of Samsung Electronics were trading 1.43 percent higher at 56,700 won ($40) on Monday morning.

[ⓒ Maeil Business Newspaper & mk.co.kr, All rights reserved]