News

Memory breakthrough by China’s CXMT puts chip rivals on back foot

Written by Nikkei Asia Published on   4 mins read

Share
Image source: ChangXin Memory Technologies (CXMT).
Image source: ChangXin Memory Technologies (CXMT).
Samsung, SK Hynix, and Micron trail a Chinese rival in a next-generation advance.

A breakthrough in next-generation memory chip design, announced by China’s ChangXin Memory Technologies (CXMT) has sent shockwaves through markets and the semiconductor industry, although analysts and industry executives said its rivals have other ways to maintain their competitive edge.

CXMT president Cao Kanyu told an industry conference that a key advance had been made in a next-generation memory cell architecture called 4F-squared, which would allow more cells on a same-sized chip than the existing 6F-squared structure.

It was another example of a Chinese company claiming a technological leap despite being denied access to the latest Western equipment and components by US trade restrictions. In May, Huawei unveiled semiconductor and system design principles, dubbed the “Tau scaling law,” that use techniques beyond the traditional shrinking of transistor sizes to improve performance.

CXMT staged a blockbuster IPO on Shanghai’s Star Market in July, briefly becoming China’s most valuable company. It is the fourth largest DRAM chipmaker globally, growing to a 10% market share at the end of the second quarter, according to Counterpoint Research. Samsung, SK Hynix, and Micron had 38%, 25%, and 24% shares, respectively.

Some analysts said CXMT’s advance was not a threat to its South Korean rivals as there were still more ways to improve productivity per wafer in memory chips rather than changing the structure.

“Korean companies have some easy methods they can use, and once those are exhausted, they’ll have to start tackling the more difficult ones. But with CXMT, the US has imposed restrictions preventing the sale of equipment, so there are no easy solutions,” said CW Chung, a senior analyst at Nomura based in Hong Kong.

“Since there’s no easy solution, the company is now resorting to stopgap measures, either multi-patterning or trying to adopt this 4F-squared technology a bit sooner.”

A South Korean chip executive told Nikkei Asia that controlling how much heat a chip generates and how to dissipate it in a densely packed structure was one of the toughest challenges memory chipmakers need to solve for new architectures like 4F-squared and 3D DRAM.

“It is yet to be seen if CXMT can successfully ramp up production and solve the thermal issue in mass production,” said the executive.

Dynamic random access memory chips, or DRAM, are critical components used in all kinds of electronic devices, from computers and smartphones to data center servers and cars. DRAM stores data in units called cells, and the area occupied by a single cell is denoted as F-squared. Most memory chips today use a 6F-squared structure, the terminology referring to how much space each cell occupies on a chip. The smaller the number, the more tightly packed the cells, allowing more memory storage.

Moving to 4F-squared would be the first such transition since the mid-2000s when the industry shifted from 8F-squared to adopt 6F-squared.

Both SK Hynix and Samsung have been working on 4F-squared chips, with the former revealing last year plans to transition to a 4F-squared vertical gate (VG) platform for its more advanced chips. Samsung has not officially revealed its strategy, but an industry source familiar with the matter said that the company was also developing 4F-squared. Local media reported that it was working on a 4F-squared structure paired with vertical channel transistors (VCT).

“If you think about how memory technology works, where ultimately you can get more bits per wafer area or more bits per dollar, accelerating those next generations of technologies is also very valuable,” Tristan Holtam, group vice president and head of corporate strategy at Applied Materials, told Nikkei Asia.

He was speaking at the US chip equipment supplier’s unveiling on October 12 of the USD 5 billion EPIC Center for research and development in Silicon Valley, where Samsung, SK Hynix, and Micron will work with it to develop next-generation semiconductors.

“If we can pull forward 4F-squared, if we can pull forward 3D DRAM, that is also going to fulfill a lot of the market requirement,” he said, adding that the EPIC Center would help chipmakers cut in half a ten- to 15-year chip design timeline.

CXMT’s Cao told the industry conference that the core of CXMT’s 4F-squared design was a proprietary transistor called vertical surrounded gate (VSG), which stands upright and wraps the electrical gate around the channel on all sides, giving the chip far greater control over the electrical current to prevent leakage.

The company also plans to launch a new high-performance server-grade RAM module, called DDR5 RDIMM, based on the technology before the end of this year. DDR5 RDIMM is designed for data centers and AI servers.

In September, CXMT also said it had achieved mass production of its fifth-generation DRAM technology platform, putting it “close to the leading edge of global memory production.” The company also unveiled a new high-capacity LPDDR5X chip, designed primarily for mobile devices and laptops, built on the same platform.

LPDDR5 is now widely used in high-end smartphones, while DDR5 is commonly found in desktops, laptops and servers. Meanwhile, server makers and AI chip developers, including Nvidia, AMD, and Qualcomm, are increasingly adopting LPDDR5 memory to support their central processing units (CPUs) and improve computing performance and power efficiency.

Memory chip prices have rocketed, driven by artificial intelligence demand that has created shortages at every level. President Lee Pei-Ing of Nanya Technology, a smaller Taiwanese rival to CXMT specializing in DRAM, said the supply gap was still massive and that his company expected its record gross margin of more than 85% to grow further in the current quarter.

“We do see the demand for AI-related applications in particular is strong, while we also see significant spill-over effects that make memory for other electronic devices very tight, and we see constraints there as well,” Lee said.

Nikkei Asia has exclusively reported that HP, Asus, and Acer have started using some CXMT memory to minimize supply constraints and that CXMT is launching the most aggressive expansion plans in the company’s history.

This article first appeared on Nikkei Asia. It has been republished here as part of 36Kr’s ongoing partnership with Nikkei.

Share