Samsung (SSNLF.US) unveils its 3D memory "technical bomb": 400-layer V-NAND and zHBM concept debut, targeting the AI memory throne.

On Tuesday local time, Samsung Electronics (SSNLF.US) held the Future Memory and Storage conference in Santa Clara, California, where it showcased its next-generation memory technologies targeting the age of artificial intelligence (AI). This includes the V10 Bonding V-NAND (BV-NAND) prototype with over 400 layers, vertically stacked high-bandwidth memory (zHBM), and the zNAND-O architecture, aiming to consolidate its leading position in the nearly $1 trillion storage chip market and accelerate its efforts to catch up with early movers SK Hynix and Micron Technology, Inc. (MU.US) in the AI memory field. As the world's largest memory chip manufacturer, Samsung presented its V10 Bonding V-NAND (BV-NAND) prototype for the first time at the conference. This chip adopts a novel wafer bonding architecture, exceeding 400 stacking layers, which enhances storage density by approximately 58% compared to the previous generation V9 NAND while also improving read/write speeds and input/output performance. As AI workloads expand from training large language models to real-time generation of user query answers, the demand for high-capacity, high-energy-efficient NAND flash memory is rapidly increasing. Samsung aims to better meet the storage needs of data-intensive AI systems with this technology. In a more cutting-edge domain, Samsung unveiled concept models of what it calls the industry's first zHBM and zNAND-O architecture, outlining a future roadmap for 3D stacked memory. Unlike traditional HBM currently placed beside AI processors, zHBM vertically stacks high-bandwidth memory above AI accelerators, significantly shortening data transport distances. Samsung stated that this system could achieve approximately eight times the performance improvement compared to next-generation HBM5, with memory density improved by over ten times, energy efficiency tripled, heat resistance reduced by more than half, and support for customized designs. However, the company has not yet provided a specific production timeline for HBM5 and zHBM, planning to prioritize the expansion of HBM4 production scale in the second half of this year. The zNAND-O, also showcased, is based on V-NAND technology and optimized for real-time, data-intensive AI applications. Samsung's technological push comes amid fierce competition among storage chip giants for AI data center orders. SK Hynix has gained an early advantage in the AI memory market thanks to its early positioning, while Micron is also actively vying for orders. Samsung is emphasizing its role as an end-to-end AI infrastructure supplier, making strides not only in memory hardware but also disclosing roadmaps for next-generation in-memory computing chips and high-capacity enterprise storage. On the demand side, although some investors are concerned that a sluggish Chinese smartphone market may suppress long-term memory demand, analysts generally see strong momentum in AI demand. Samsung revealed last week that long-term agreements with customers are expected to cover 60% to 70% of the company's memory sales. Citibank analysts noted in a report last month that despite concerns about softening end-market demand, the current inventory levels across the entire DRAM and NAND supply chain remain significantly lower than historical levels, providing a buffer for the industry and strengthening expectations for AI-driven growth.