SK Hynix Unveils 3D FeNAND Technology for AI Computing
SK Hynix has published research detailing its 3D FeNAND technology applied to Analog Compute-in-Memory (A-CiM), as presented in an IEEE paper. The technology leverages ferroelectric materials within a 3D NAND-like architecture to execute direct analog computations. This development addresses critical efficiency bottlenecks in AI processing by significantly reducing data movement between memory and processors. By utilizing Analog Compute-in-Memory, it enhances both processing speed and energy efficiency for hyperscale AI models. Unlike traditional 3D NAND that traps electrons using charge-trap nitride (CTN), 3D FeNAND utilizes the stable polarization states of ferroelectric materials to control current. This enables a single storage cell to generate multiple conductance states, making it well-suited for storing AI weight data.
## BACKGROUND
Analog Compute-in-Memory (A-CiM) merges data storage and computation within the same physical substrate, allowing matrix multiplication operations to happen directly inside the memory using analog signals. Ferroelectric flash memory (FeNAND) incorporates ferroelectric materials like doped hafnium oxide into NAND flash architectures to overcome the endurance, power consumption, and scaling limitations of conventional flash memory.