Micron abandons 4F ² DRAM, institution calls ASML or inbound hybrid bonding
After Micron abandoned the 4F ² DRAM architecture, some institutions predict that ASML is developing hybrid bonding devices.
Recently, Micron announced its plans for future generations of DRAM, HBM, and NAND at a technology roadmap webinar. Scott DeBoer, the President and Chief Technology and Product Officer of Micron, stated that Micron has been continuously researching 4F ² DRAM for approximately 20 years and has also manufactured prototype products for comparative evaluation; But the route was limited in complexity, cost, and further miniaturization, and ultimately chose not to proceed. DeBoer also pointed out that the density advantage of 4F ² compared to traditional 6F ² may be overestimated. The bit density of DRAM largely depends on the spacing and size of storage capacitors, rather than solely on whether the transistor unit adopts a 4F ² or 6F ² layout.
The key to this judgment is that DRAM continues to rely on the path of reducing two-dimensional cells in exchange for density, approaching the region of diminishing returns. 4F ² may seem to compress the cell area on the layout, but the true determinant of bit density is whether the capacitor can continue to be made deep, narrow, and stable; When the aspect ratio etching, deposition uniformity, leakage, reliability, and yield costs are synchronously increased, the book value of simply switching from 6F ² to 4F ² may be offset by process complexity and manufacturing risks. Micron did not deny miniaturization, but instead prioritized vertical integration, packaging interconnection, and system energy efficiency as more important sources of density after the planar limit slows down.
That's why SemiAnalysis's recent review of ASML patents has attracted attention. After analyzing 27 patent applications submitted by ASML, the organization believes that ASML is developing hybrid bonding equipment and is likely to be the first to launch chip to wafer D2W hybrid bonding tools.
ASML has neither confirmed nor denied, only stating that they will not comment.
But if the prediction comes true, the significance is not just about having one more equipment supplier, but about forming the same industry logic as Micron's abandonment of 4F ²: as unit scaling slows down, the storage industry needs new "equivalent scaling" methods, and hybrid bonding is the key process to continue pushing logic, storage, and HBM stacking towards high density and high bandwidth.
From a technical perspective, hybrid bonding involves direct bonding between dielectric and copper, replacing the traditional micro bump and bottom filling connection method. This can shorten vertical interconnection, reduce RC delay and power consumption per bit, and support finer I/O spacing, thinner chips, and higher stacking.
For HBM, the D2W mode is more flexible than wafer to wafer bonding, allowing for the selection of known good quality storage dies to be paired with logic substrates, reducing the risk of large-scale scrap, and better suited for AI accelerators to rapidly iterate HBM3E, HBM4, and subsequent high bandwidth, large capacity, and low-power products.
For the DRAM body, hybrid bonding can also support the migration of peripheral circuits to the bottom or top of the array, freeing up the array area and shortening the bit line connections without forcing 4F ², which is equivalent to transforming the problem of "drawing cells smaller" into a system engineering of "stacking the array and logic closer".
The market has already traded this expectation in advance. On the day of the news, BESI's stock price plummeted by 4.8%, cutting more than half from its June high; ASML Amsterdam listed stocks closed slightly higher. BESI has previously been downgraded by UBS and Bank of America, with Bank of America specifically citing "potential risks of ASML entering competition" as one of the reasons for downgrading to "neutral".
The concern behind this is that if ASML extends its experience in photolithography, alignment, measurement, and process control to the bonding process, the existing bonding equipment layout, customer bargaining power, and standard setting pace may be rearranged; For storage plants, this is both an opportunity for supply chain diversification and a new round of equipment capital expenditure and process synergy competition.
However, the combination of hybrid keys and non capacitive miniaturization alternatives cannot automatically solve the density pressure left by Micron's abandonment of 4F ². Mass production still faces challenges such as sub micron alignment, surface cleanliness, cavity control, low heat budget, warpage management, post bonding testing, and redundancy repair; D2W emphasizes flexibility, but needs to balance rhythm and cost. W2W has better alignment and interface quality, but is constrained by overall yield. Especially in the high-capacity and thin stacking scenarios of HBM, thermal management and interface defects will be further amplified, and any single point breakthrough requires the coordination of design rules, DRAM processes, packaging testing, and terminal loads.
