Samsung Electronics, in collaboration with Samsung Advanced Institute of Technology, has unveiled a groundbreaking achievement in memory technology: the successful development of a manufacturing process for DRAM below 10 nanometers. This breakthrough, announced at the IEEE International Electron Devices Meeting, provides critical technological support for the demanding storage needs of AI and high-performance data centers, pushing the boundaries of the industry.
The core innovation is a novel "Cell-over-Periphery" (CoP) architecture. Unlike traditional designs where peripheral transistors are placed below the memory cells, Samsung's approach stacks the memory cells directly on top of the peripheral circuitry. This fundamental redesign tackles the persistent industry challenge where transistor performance degrades due to the high temperatures involved in conventional stacking processes.
Making this architecture possible is a new type of heat-resistant amorphous oxide semiconductor transistor. Built using amorphous indium gallium oxide (InGaO), this transistor can withstand extreme temperatures up to 550°C. Paired with a 100nm vertical channel, it is ideally suited for monolithic integration in the CoP DRAM structure. Testing shows this transistor exhibits minimal drain current degradation and stable performance in long-term reliability tests, ensuring the high performance and durability required for next-generation DRAM.
As a technology currently in the R&D stage, Samsung plans to integrate this innovation into future sub-10nm "0a" and "0b" class DRAM products. This advancement promises a leap in storage density and data transfer speeds while lowering the cost per bit, offering powerful support for applications like AI model training, supercomputing, and premium mobile devices.
ICgoodFind's Insight
Samsung's sub-10nm DRAM breakthrough represents a fundamental architectural shift, not just a process shrink. It charts a viable path for memory scaling in the AI era, ensuring that performance keeps pace with the explosive growth in data processing demands.
