⚡ 本页包含 AI 生成的分析内容，仅供参考
该论文提出了一种针对生成式AI大模型的高容量高带宽HBM4 DRAM，通过2048个I/O实现3.3TB/s带宽和36GB容量，采用第6代DRAM核心与4nm FinFET基底芯片，并通过每通道TSV RDQS自动校准技术改善了数据窗口和时序裕量。
ChangHyun Bae, Joohwan Kim, Je-Min Ryu, Shin-haeng Kang, Jaehoon Lee, Young-Uk Chang, JaeKyung Lee, JongTae Hwang, Daehwan Seo, Ki-Heon Na, Young Guen Song, Daihyun Lim, Kyung-Soo Ha, Young-Soo Sohn, Sang-Joon Hwang Samsung Electronics, Hwaseong, Korea Abstract LLM-based generative AI requires high-capacity and high-bandwidth memory. We propose an HBM4 with 2048 I/Os, 3.3TB/s bandwidth, and 36GB cube density. Core die uses the 6thgen. DRAM and the base die uses a 4nm Fin-FET process. TSV count and data window are doubled for high-frequency margin. tCCDR is improved via per-channel TSV-RDQS auto calibration, enhancing rank-to-rank access. PMBIST enables at-speed testing in CoW and SiP. The on-die WDQS skew cancellation ensures robust timing, enabling reliable highspeed operation and improved yield. Large language models (LLM)-based generative AIs are employed across diverse applications: including image and video generation, text composition, and information
Sunghwan Joo, Jinyeon Kim, Yongsun Lee, Ji-Young Kim, Youngsik Lee, Yong-Min Kim, ChiSung Oh, Kyu-Ha Shim, Haesuk Lee, Young-Yong Byun,