As global tech giants develop various architectures to maximize AI data processing efficiency, demand is growing for NAND that can meet diverse requirements. According to industry sources cited by South Korean media, as part of its strategy to meet diversified demand, Samsung Electronics is internally developing three types of z-NAND flash memory. In addition to zNAND-O, for which a model has already been publicly demonstrated, the portfolio also includes zNAND-T and zNAND-I, which are optimized for throughput and random performance, respectively.
zNAND-O: Built for On-Device AI
The "O" in zNAND-O stands for "on-device", and it is a product designed specifically for on-device AI. Unlike the traditional method of transmitting data to the edge of the chip through wire bonding, it uses through-silicon via (TSV) technology to stack NAND layers into 4 or 8 layers. It shortens the data transmission path, thereby improving speed and energy efficiency.
zNAND-T: Aimed at High Throughput
The "T" in zNAND-T stands for "throughput." Its concept is similar to HBF proposed by SK hynix and Sandisk, namely, targeting data center AI semiconductor demand by optimizing read/write bandwidth performance and using TSV to stack NAND dies. Its hallmark is maximizing bandwidth to meet the needs of high-capacity, high-performance AI data centers.
zNAND-I: Focused on High IOPS
The "I" in zNAND-I is short for IOPS (input/output operations per second), which represents random performance. It emphasizes performance in frequent read/write operations involving small amounts of data and maximizes IOPS. Although wider bandwidth allows large amounts of data to be processed at once, higher IOPS means more data I/O operations can be completed per second. This is precisely why high-IOPS NAND is becoming increasingly popular in the field of AI inference — during AI inference, such as KV Cache workloads, low-capacity temporary data needs to be processed frequently.
All Rights Reserved © 2026 MemoryMarket.com
