长鑫存储宣布在芯片技术上取得突破，这一技术是三星电子等竞争对手多年来一直在追求的，这一未经证实的说法震动了股市，并反映出中国在关键领域的积极推进。
Corp. declared it achieved a breakthrough in chip technology that rivals like Samsung Electronics Co. had pursued for years, an unsubstantiated claim that jolted stocks and reflects China’s aggressive push into a critical sphere.
这家国内存储龙头企业计划推出一种新型芯片，可将更多算力集成在单个组件中，长鑫存储总裁曹康宇周末在上海举行的一个行业会议上表示。曹康宇称，这款面向数据中心服务器的新产品采用所谓的 4F² 设计，预计将于年底前发布，公司发言人证实了这一表态。
The domestic memory leader plans to roll out a new type of chip that packs more computing power into a single component, CXMT President Cao Kanyu told an industry conference in Shanghai over the weekend. The upcoming product for data center servers sports a so-called 4F² design and should launch before the end of the year, Cao said in comments confirmed by a company spokesperson.
如果长鑫存储已成功达到 4F² 门槛，意味着它在三星电子和 SK 海力士之前实现了这一目标，尽管美国制裁禁止这家中国企业获得 ASML Holding NV 等公司的顶尖设备。曹康宇未提供更多细节，但广泛传播的言论助推了全球芯片股的抛售潮。
If CXMT has succeeded in achieving the 4F² threshold, that means it got there before Samsung and SK Hynix Inc., despite US sanctions that bar the Chinese firm’s access to tools such as ASML Holding NV’s top-end machines. Cao didn’t provide more details, but the widely reported remarks helped spark a selloff in global chip stocks.
周二，三星电子和 SK 海力士股价在韩国下跌逾 7%，拖累更广泛的韩国综合股价指数（Kospi）下跌，此前美光科技股价隔夜收跌 4%。
Samsung and SK Hynix shares fell more than 7% in South Korea on Tuesday, weighing down the broader Kospi index, after Micron Technology Inc. closed down 4% overnight.
“若能在 DRAM 生产中成功且更早地实现 4F² 架构，将大幅提升长鑫存储相对于同行的竞争地位，”晨星分析师余静洁表示。“话虽如此，存储龙头企业和设备制造商均同样强调了这一技术转型的难度，我们仍预计长鑫存储 4F² 产品发布与以良率实现商业化之间仍存在显著的时间差距。”
“A successful and earlier implementation of 4F² architecture in DRAM production would vastly improve CXMT’s competitive position versus peers,” said Jing Jie Yu, an analyst with Morningstar. “That said, memory leaders as well as equipment manufacturers have both similarly echoed the difficulty in this technological transition, and we still expect a meaningful time gap between the unveiling of CXMT’s 4F2 product and commercialization at viable yields.”
曹康宇的表态——此前几个月，长鑫存储在一次盛大的夏季 IPO 后曾短暂成为中国市值最高的公司——凸显了其在芯片制造领域缩小与外国竞争对手差距的努力，该领域对 AI 发展至关重要。
Cao’s declaration — months after CXMT briefly became China’s most valuable company following a blockbuster summer IPO — underscores its effort to close the gap with foreign rivals in chipmaking, an area essential to AI development.
行业现行的6F²标准指的是芯片表面积等于“特征尺寸”（即晶体管之间的基本线宽）的平方的6倍。三星和SK海力士正在探索将其缩小至4F²的方法，理论上可以在更小的面积内容纳相同的单元，从而提高效率。
The industry’s existing 6F² standard refers to a surface area of 6-times the square of the “feature size” of a chip: the basic line-width between transistors. Samsung and SK Hynix are exploring ways to shrink that to 4F², in theory packing the same cells into a smaller area for greater efficiency.
长鑫存储的曹表示，他们正通过堆叠改变晶体管的布局方式。所谓的垂直环绕栅（Vertical Surrounding Gate）方法应能减少漏电及其他副作用——这是更大竞争对手在尝试突破6F²标准以降低尺寸时长期面临的问题。
CXMT’s Cao said it’s changing the way transistors are laid out via stacking. The so-called Vertical Surrounding Gate approach should reduce leakage and other side effects — a problem long associated with efforts by bigger rivals to reduce sizes beyond the 6F² standard.
如果成功，这种新架构将成为华为技术有限公司到DeepSeek等中国企业为规避美国技术限制而采取的创新变通方案的最新例证。
If it works, the new architecture marks the latest example of innovative workarounds that Chinese companies from Huawei Technologies Co. to DeepSeek devise to get around US technology curbs.
“投资者神经紧张，正在寻找借口以获利了结今年表现优异的AI相关股票，尤其是内存股票，”瑞士私人银行联合银行（Union Bancaire Privee）董事总经理Vey-Sern Ling表示。
“Investors are jittery and looking for excuses to take profit on AI-related stocks that have outperformed this year, especially memory stocks,” said Vey-Sern Ling, a managing director at Union Bancaire Privee.
华为在5月推出了Tau缩放定律，作为几十年来以缩小晶体管来提升芯片性能的摩尔定律的替代方案。这家中国芯片领军企业则通过LogicFolding技术，寻求在芯片内加速信号传输并提高晶体管密度。该公司已推出搭载此技术的芯片，用于今年发布的智能手机。
Huawei in May unveiled its Tau Scaling Law, an alternative to the decades-old Moore’s Law of shrinking transistors to boost chip performance. The Chinese chip champion instead seeks to speed up transmission and increase transistor density within a chip through LogicFolding. It’s already rolled out chips using this technology that run on smartphones launched this year.
长鑫存储正加速提升先进内存的产能，以应对AI需求的激增。具备制造复杂产品的能力也将使公司在未来几年全球供应紧张局势开始缓解时处于更有利的位置。
CXMT is racing to ramp up production of advanced memory to meet surging AI demand. The ability to manufacture sophisticated products would also position the company better as a global supply crunch begins to ease in coming years.
目前，长鑫存储在高带宽内存领域仍属落后——这种半导体是英伟达公司AI加速器所使用的类型。但公司在更广泛的DRAM市场正取得稳步进展。据Counterpoint Research的数据显示，其在该领域的市场份额已从去年的4%扩大至今年第二季度的10%。
It remains a laggard in high-bandwidth memory, the type of semiconductors used in Nvidia Corp. AI accelerators. Still, it’s making steady progress in the broader DRAM market. The company’s share of that field had expanded to 10% by the second quarter of this year, compared to just 4% a year ago, data compiled by Counterpoint Research shows.