UniC-Vision: A 14.4Gb/s 7.3pJ/b Universal Vision Transformer OFDM Channel Estimation Accelerator for B5G/6G AI-RAN

📋 论文概要

该论文提出了一款用于B5G/6G AI-RAN的通用视觉变压器OFDM信道估计加速器UniC-Vision，通过算法-硬件协同优化和架构优化，在28nm CMOS工艺下实现了14.4Gb/s的吞吐量和7.3pJ/b的能效，比现有系统在可靠性上提升7.7至18.9dB，吞吐量提升16至39倍，满足超可靠低延迟的6G需求。

💡 主要创新点

• 首次将视觉变压器（Vision Transformer）应用于OFDM信道估计，利用其全局注意力机制提升估计精度和鲁棒性。
• 采用算法-硬件协同优化方法，针对信道估计任务定制神经网络结构，并设计专用加速器架构，实现高吞吐量和低能耗。
• 通过架构级优化（如流水线、数据复用、并行计算）在28nm工艺下达到14.4Gb/s的业界领先吞吐量，同时保持7.3pJ/b的高能效。

🏷 关键词

📄 原文摘要

Abstract This work presents an AI-RAN channel estimation accelerator for next-generation communications, offering hyper reliability, low latency, and universal frequency range coverage, thereby meeting B5G/6G requirements. Utilizing algorithm-hardware co- optimization and architecture optimizations, the prototype achieves a throughput of 14.4Gb/s at 7.3pJ/b in 28nm CMOS, surpassing the state-of-the-art channel estimation systems by 7.7 to 18.9dB in reliability and 16-to-39× in throughput. Future wireless systems beyond 5G and 6G (B5G/6G) aim to meet the stringent requirements of hyper-reliable low-latency communication (HRLLC), imposing unprecedented constraints on wireless physical layer (PHY) design. 6G involves a broad frequency range (FR), including sub-6GHz (FR1), mm-Wave (FR2), and cm-Wave (FR3), along with ultra-massive MIMO and wide bandwidth, which introduces new physical phenomena and severe workload [1]. Under such reliability and latency-hungry conditions,

👥 作者与机构

Sangbu Yun1,2, Chanhee Lee1,2, Soonhyun Kwon1, Jeongtaek Chang3, Zhengya Zhang3, Youngjoo Lee1

Korea Advanced Institute of Science and Technology, Daejeon, Korea, 2Pohang University of Science and Technology, Pohang, Korea, 3University of Michigan, Ann Arbor, MI