From
Send to

HBM leadership is coming to us: Samsung CEO

Tech giant sets up HBM-only research team in AI push

April 1, 2024 - 15:12 By Jo He-rim
Samsung Electronics CEO Kyung Kye-hyun in charge of the tech giant’s Device Solutions Division. (Samsung Electronics)

Samsung Electronics has established a special team dedicated to High Bandwidth Memory, going all-out to get an edge in the advanced memory chip, which is currently in the spotlight as a key component in raising the processing speed for artificial intelligence applications.

Highlighting the importance of HBM in the wake of the AI boom, Samsung Electronics Vice Chairman Kyung Kye-hyun vowed to keep up efforts to secure leadership in the burgeoning market and revealed the company has formed a team dedicated to the advanced chip.

"In AI applications, having high-capacity HBM means a competitive edge. This is the reason clients want HBM3 and HBM3E 12H," Kyung wrote in a Korean-language Instagram post mentioning his business trip in the US, on Friday.

"We have formed a team dedicated to HBM, and the team is doing its best to improve quality and productivity. We are gaining leadership in the HBM market with this effort."

HBM chips stack DRAM chips vertically to increase the data processing speed. The HBM3E 12H is the 12-stack HBM3E chip that Samsung has developed faster than crosstown rival SK hynix or runner-up Micron Technology, based in the US. The other firms have announced plans to start mass production of 8-stack HBM chips.

For HBM4, memory bandwidth would double, but the traffic bottleneck occurring between the memory and computing would remain, Kyung explained.

"Many clients want to resolve this problem and develop customized HBM4 in their own ways, and we will do the work with them," he said.

In his post, the CEO also highlighted the company's cutting-edge gate-all-around technology for the manufacturing of highly-advanced 2-nanometer chips, and the company's AI accelerator, the Mach-1, which is currently under development.

"We should reduce the logic power and enhance performance to train AI’s intelligence in diverse applications. This is the reason clients want the 2-nanometer gate-all-around (processing technology)," Kyung wrote.

"For this, many clients are already providing, or have promised to provide, their chips for testing. And we will successfully develop the technology to make sure these attempts lead to the actual development of 2 nm products."

Memory chip aside, Samsung also runs foundry and logic chip design businesses and aims to become the world's top memory chipmaker by 2027.

"Clients’ interest is also growing in the Mach-1, which is used for AI inference. Some clients want to use Mach for applications with over 1 trillion parameters. We need to prepare the Mach-2 faster than we expected."

Naver, South Korea's IT giant, is expected to be the first client of Samsung's Mach-1 AI accelerator.

According to market tracker Omdia, Samsung's chip division placed No. 3 in the global chip market last year, after Intel and Nvidia, which each witnessed their sales boost with central processing units and graphic processing units. Samsung logged a revenue of $4.47 billion last year.

In the post, Kyung also revealed he has been meeting clients while traveling to five cities in four days in the US, and that he was able to test drive Tesla's Cybertruck at the electric vehicle firm's headquarters.

“It was more comfortable than I expected, and it also had a great accelerating force. The truck appeared to be great at tracking the surroundings with its 10 cameras and the short steering radius and big wipers were impressive,” Kyung said, recalling his experience with Tesla's Cybertruck.

SK hynix and Samsung, the two top memory chipmakers in the world, currently dominate the market, respectively taking 53 percent and 38 percent of the market volume, according to TrendForce, a market tracker. Micron Technology accounts for the remaining 9 percent share.