自研CXL会砸自己DRAM饭碗？三星、SK海力士、美光同时放弃CXL控制器自研
- 核心观点：全球三大内存巨头（三星、SK海力士、美光）已放弃自研CXL控制器，转向外购专业芯片设计公司产品，这标志着CXL生态从一体化方案转向专业化分工，内存厂商回归制造核心，避免冲击其最重要的通用DRAM业务。
- 关键要素：
- 美光已关闭CXL控制器研发部门并采用PrimeMass方案；SK海力士终止项目并转向PIM研发；三星移除自研路线图，保留前沿研究并外购控制器。
- 数据中心客户偏好模块化架构（控制器独立部署+标准化低价DIMM），而非内存厂商推行的集成式“自研控制器+DRAM”一体化方案。
- 内存厂商面临两难：若推广集成CXL产品，将直接与核心收入来源“通用DIMM模块”竞争，存在蚕食DRAM业务的风险，故放弃自研以保盈利模式。
- 专业芯片设计公司（Astera Labs、Montage Technology、PrimeMass）正填补CXL控制器市场空缺，受益于这一产业分工转变。
- 三大厂商的转向不意味着放弃CXL，而是将研发资源集中于先进DRAM工艺、HBM及PIM等更具长期竞争力的领域，推动CXL生态迈向成熟商业化。
Original author: Li Jia
Original source: Wall Street CN
The world's three largest memory chip manufacturers have fully suspended their in-house development plans for CXL (Compute Express Link) controllers, opting instead to purchase products from specialized chip design companies. This collective shift reflects the core contradiction of the memory industry: Aggressively promoting self-developed chips could cannibalize their most important revenue source—the general-purpose DRAM module market.
On July 20, according to a report from South Korean tech media ZDNet Korea, Samsung Electronics, SK Hynix, and Micron Technology have all scaled back or abandoned the commercialization plans for their own CXL expansion device controllers. Specialized chip design companies such as Montage Technology, Astera Labs, and PrimeMass are filling this void.
This change signifies a reshaping of the division of labor within the CXL ecosystem. Memory manufacturers will focus on the manufacturing process, while design leadership shifts to independent chip design companies.
For the capital markets, this not only benefits the relevant chip design firms but also means that, in the short term, the three major memory manufacturers will not engage in new competition around complete CXL solutions. Their core profit model will continue to revolve around traditional DIMM memory products.
Strategic Adjustments by the Three Giants
The pace at which the three companies exited self-developed controllers varied.
Micron was the most decisive. According to reports, the company has closed its CXL controller R&D department and adopted PrimeMass's controller solution. Related products have already entered Micron’s product catalog.
SK Hynix has formally notified partners that it has terminated its in-house controller project and redeployed the R&D team to the PIM (Processing In Memory) business, indicating its desire to concentrate resources on the next-generation memory architecture with greater long-term competitiveness.
Samsung's adjustment was relatively more cautious. Citing a semiconductor industry insider, the report stated that Samsung has removed its self-developed CXL controller from the official commercialization roadmap, retaining only advanced R&D projects. The team is currently exploring areas such as combining LPDDR with CXL.
Meanwhile, Samsung has turned to external procurement for controller products. This means that the company's previously planned "self-developed controller + CXL Memory Module (CMM)" integrated product has, in effect, been put on hold.
Why Did Self-Developed Controllers Lose Their Appeal?
The initial vision of the three major memory manufacturers was remarkably consistent: integrate the controller and DRAM on the same circuit board and sell it as a complete solution to capture higher product added value.
However, the market ultimately chose a different path. Data center customers prefer a modular architecture, where the controller is deployed independently on the motherboard and paired with standardized, low-cost DIMM memory modules. This approach not only reduces procurement costs but also retains flexibility for future upgrades and supply chain choices.
The shift in customer demand has put memory manufacturers in a dilemma. Persisting in promoting integrated products not only requires bearing the R&D costs of the controller but may also hinder market acceptance due to higher prices. More importantly, if customers reduce their purchases of standard DIMMs, it would directly impact the memory manufacturers' core revenue streams.
Citing a semiconductor industry source, the report notes: If the complete CXL product has to directly compete with the company's DIMM business, the incentive to commercialize the controller diminishes significantly. Facing the risk of cannibalizing core business, management finds it difficult to continue investing resources.
In other words, rather than cultivating a new market that could weaken the DRAM business, it is better to let specialized firms handle the controller, allowing the companies to continue focusing on memory manufacturing where they hold the greatest economies of scale.
CXL Ecosystem Enters a Stage of Specialized Division of Labor
Industry insiders believe that the withdrawal of the three major manufacturers from self-developed controllers does not signify abandoning CXL.
On the contrary, it is more like a typical specialization of labor during the maturation of an industry. Controller design is handed over to Fabless chip design companies, while memory manufacturers continue to leverage their advantages in manufacturing and DRAM technology, each creating value in their most competitive domains.
With this trend established, Astera Labs, Montage Technology, and PrimeMass are poised to further solidify their positions in the CXL controller market. Meanwhile, Samsung, SK Hynix, and Micron can concentrate their R&D resources on areas with greater long-term value, such as advanced DRAM processes, HBM, and PIM.
From an industry chain perspective, the developmental logic of CXL has not changed; what has changed are the roles of the participants. Memory manufacturers are no longer attempting to build a complete platform but are reverting to their roles within the supply chain division of labor. This also indicates that the CXL ecosystem is transitioning from an early exploratory phase to a more mature stage of commercialization.
