China’s leading domestic DRAM producer has announced a new step in the country’s effort to build a more capable memory-chip supply chain. ChangXin Memory Technologies, known as CXMT, says its fifth-generation DRAM technology platform has entered mass production and is now underpinning two new mobile-memory products.
The key word is “says.” Reuters reported the September 20 announcement from the Hefei-based company and made clear that the technical and manufacturing claims originate with CXMT. The news was corroborated by SCMP and China Daily, but it is not an independently audited account of output volume, yield, or technical parity with global competitors.
The announcement matters because DRAM is the short-term working memory used in phones, computers, and many other electronic devices. For a Chinese producer seeking to reduce reliance on foreign technology, a move from development claims to mass production is more consequential than another laboratory demonstration. But the details need to be read carefully, especially the difference between the number of dies a wafer could produce and the number of usable chips that pass final testing.
A fifth-generation platform, not a declared global-equivalence milestone
CXMT calls the new system its fifth-generation DRAM technology platform. SCMP identifies it as G5, which is useful as a company label but should not be treated as an industry-standard name. Reuters reported that CXMT reduced the spacing of key features in the data-storage portion of the chip to 11.95 nanometers using quadruple patterning. SCMP described the same figure as an active-area half-pitch and used the term quadruple exposure.
Those terms offer a view into the company’s approach, but they do not establish a direct apples-to-apples ranking against a particular Samsung, SK Hynix, or Micron process. Memory-chip terminology is highly specific. An 11.95-nanometer figure in a memory array is not a general logic-chip node, and it does not by itself prove that CXMT matches another company’s overall process, product performance, yield, or cost.
That is why the strongest claims in the announcement require attribution. Reuters reported that Luo Xiaodong, CXMT’s vice president and head of its marketing center, said at the 2026 World Manufacturing Convention in Hefei that the company’s process capability was now on par with the most advanced mass-produced nodes in the industry. That is an important corporate statement, but it remains an executive assessment rather than an independent technical benchmark.
The new platform is separate from the earlier report that CXMT was exploring a NAND research-and-development line amid an AI-driven memory shortage. NAND and DRAM are different memory technologies. The previous story concerned a possible NAND R&D move; the current announcement is about a fifth-generation DRAM platform, new LPDDR5X products, and CXMT’s manufacturing claims.
Two LPDDR5X products carry the first commercial message
CXMT unveiled two 24-gigabit LPDDR5X products made on the new platform. LPDDR5X is a low-power form of DRAM typically used in smartphones and other portable electronics. Reuters and SCMP both report the 24-gigabit figure. That unit is important: it is gigabits, not gigabytes. A China Daily report used conflicting gigabyte wording, which should not be repeated because the two independent reports agree on gigabits.
CXMT says each new product stores 50% more data than its previous equivalent products and that both have entered mass production. Reuters reported that the products are offered in two package formats, though the source does not provide their product names, speed grades, package names, or customer list. It would therefore be premature to say the chips are shipping in named phones or have reached a specific commercial volume.
The mobile-memory focus illustrates the practical significance of process advances. More memory density can allow device makers to offer higher-capacity configurations or change power and space trade-offs. But the company’s announcement does not disclose the products’ pricing, customer qualifications, measured power consumption, or field performance. Those are the kinds of details that will determine whether a process development becomes a durable commercial advantage.
CXMT also said it developed the platform with computer simulations and joint work with Chinese chip-equipment makers on critical production steps. The company did not name those suppliers in the reviewed reports. The point is still notable in the context of China’s push to develop a broader domestic semiconductor ecosystem, but the sources do not provide enough detail to identify which manufacturing tools or partners were involved.
EastFrontier has previously tracked how China’s domestic chipmakers have expanded their position in the local AI market. CXMT’s announcement relates to the memory layer of that broader ecosystem. It is not an AI accelerator story, but memory supply and density increasingly matter as AI workloads demand more capacity across devices and data centers.
The 50% die claim is not a 50% yield increase
The most easily misunderstood number in the announcement is CXMT’s statement that the fifth-generation platform can produce at least 50% more gross chip dies per wafer than the fourth-generation platform, using an 8-gigabit-chip baseline. On first reading, that sounds like a 50% increase in manufacturing output. The company’s chosen term, and Reuters’ explanation, are more specific.
“Gross dies” refers to the potential number of chip units laid out on a wafer before defective units are excluded. It is not the same as the final percentage of chips that pass testing. It does not establish a 50% yield improvement, a 50% increase in good dies, a 50% increase in wafer capacity, or a 50% reduction in cost. Those may be possible business implications in the future, but they are not facts supported by the current reports.
This measurement boundary is crucial in semiconductor reporting. A smaller feature spacing can allow more individual dies to fit on a wafer. Whether that produces a proportional gain in usable output depends on many factors, including defect rates, process stability, packaging, qualification, and customer acceptance. CXMT’s announcement addresses the density side of the equation. It does not release a comprehensive yield or cost record.
China Daily also reported a 45:1 depth-to-width ratio for the memory capacitor, attributing it to CXMT’s announcement. Since that technical figure was not corroborated by Reuters or SCMP, it should not carry the narrative. The more solid core of the story is the platform’s reported mass-production status, the 11.95-nanometer measurement, the two 24-gigabit LPDDR5X products, and the carefully bounded die-per-wafer claim.
The next test will be whether CXMT discloses more about volume, yields, customers, and sustained production. For now, the company has presented a credible marker of progress in China’s DRAM effort. The announcement signals a new platform and product generation moving into production, but it does not settle the much harder question of how the technology compares over time with the world’s established memory manufacturers.