Shenzhen Zhilian Weichuang Technology Co., Ltd.

Samsung Electronics has confirmed its next-generation DRAM three-rail pathway: HBM5 will adopt a D0A bonding architecture

Publication Date: 2026-11-18 Source:Electronic Engineering Special Issue Key words: Samsung Electronics DRAM wafer View count:

According to industry insiders, Samsung Electronics recently finalized its next-generation DRAM development strategy after D1c, dividing subsequent products into three technical routes: the D0A architecture for HBM5, the D1d process for mainstream products like LPDDR, and the D0AV process based on the 4F² unit architecture, targeting high-performance DRAM. Among them, D1d and D0A are planned for mass production in the second half of next year, while D0AV is targeted for launch in 2028.


This arrangement means Samsung is not simply choosing between sticking to traditional architectures and switching to wafer bonding, but instead simultaneously advancing multiple routes based on the performance requirements and development challenges of different products. Among them, the most notable change is the sixth-generation high-bandwidth memory HBM5, which will adopt the brand-new D0A architecture, placing memory cells and peripheral circuits on different wafers and connecting them via bonding technology.


HBM5 adopts a design that separates the memory unit from the peripheral circuit


Traditional DRAM typically integrates the storage units responsible for data storage, as well as the core and peripheral circuits controlling operations such as read/write operations, on the same wafer. The D0A architecture splits the two: the memory units are manufactured using the D1b process, while the core and peripheral circuits use the more advanced D1d process, and finally the complete DRAM structure is formed through wafer bonding.


This design primarily helps improve wafer area utilization. After the peripheral circuits are transferred to another wafer, the space previously occupied by it can be used to accommodate more storage units, allowing for greater storage capacity within the same area. At the same time, peripheral circuits are no longer fully limited by the manufacturing process of memory units, allowing independent performance and power optimization to provide greater design space for products like HBM, which require high bandwidth and energy efficiency.


The two wafers are expected to be joined using hybrid bonding technology. This technology requires precise alignment and bonding of the wafer surface insulation layer to the copper interconnect structure, enabling separately manufactured memory units and peripheral circuits to work together as if they were a single DRAM chip. However, wafer alignment accuracy, bonding quality, and subsequent mass production yield will remain key issues to address as new architectures move toward commercialization.


D1d continues to serve mainstream products and reduce transformation risks


Unlike D0A, the D1d process used in mainstream products such as LPDDR will retain the traditional single-wafer architecture, with memory cells and peripheral circuits continuing to be manufactured on the same wafer, and the unit area will still be based on the 6F² structure. According to Samsung's initial plan, there were various options regarding whether to continue developing D1d after D1c or directly switch to wafer bonding architectures. The finalized dual-track solution allows the company to drive HBM architecture innovation while maintaining the relatively steady evolution pace of mainstream memory products.


This division of labor also reflects the development logic of different products. HBM5 places greater emphasis on performance, density, and energy efficiency, and is motivated to adopt new architectures; LPDDR is widely used in mobile devices, requiring a balance between power consumption, cost, reliability, and mass production scale. Continuing to use the mature structure of D1d helps avoid the process switching risks associated with wafer bonding across all product lines.


4F² and VCT continue to advance, with D0AV targeting 2028


Samsung has not abandoned its more aggressive miniaturization approach to the D0A route; the D0AV, based on the 4F² architecture, is still under development. Traditional 6F² DRAM units occupy an area 3F wide and 2F high, while 4F² reduces both sizes to 2F, allowing more storage units to be arranged within the same area. However, arranging transistors and capacitors in a smaller space significantly increases design and manufacturing difficulty.


To address this issue, Samsung has been developing vertical channel transistor (VCT) technology, which replaces transistor channels vertically. In April this year, it was reported that Samsung had obtained the first usable chip combining the 4F² unit architecture with VCT technology, proving the technical feasibility of the solution. However, it is difficult for this technology to enter mass production directly, and internal disagreements remain over the timing and methods of application.


Against this backdrop, Samsung chose to first apply wafer bonding technology originally developed with VCT to traditional 6F² memory units, gradually forming the D0A route of HBM5. Compared to simultaneously completing major structural transformations such as 4F², VCT, and wafer bonding, this strategy is expected to reduce development complexity.


According to the plan, D0AV will be launched around 2028, roughly one year later than D1d and D0A. The specific application products have yet to be determined, and the development direction is high-performance DRAM with performance requirements higher than LPDDR. The three parallel approaches show that Samsung's next-generation DRAM strategy is not just about a single process upgrade, but rather a layered layout among mass production feasibility, unit miniaturization, and advanced architecture. In the future, whether D0A can be mass-produced on schedule, whether hybrid bonding can achieve stable yields, and whether D0AV can overcome the manufacturing challenges of 4F² and VCT will be key to observing the effectiveness of its route.