A 3nm FinFET 2.2Gsearch/s 0.305fJ/b TCAM with Dynamically Gated Search Lines for Data-Center ASICs

📋 论文概要

本文提出了一种面向数据中心ASIC的3nm FinFET TCAM设计，采用动态门控搜索线技术，实现了2.2Gsearch/s的搜索速度和0.305fJ/b的能效，解决了高吞吐量网络分类和转发中功耗与速度的平衡问题。

💡 主要创新点

• 在3nm FinFET工艺上实现了TCAM，利用先进节点降低功耗和面积。
• 引入动态门控搜索线技术，根据搜索模式动态关断部分搜索线，减少无效翻转功耗，从而大幅提升能效。

🏷 关键词

📄 原文摘要

classification and forwarding are fundamental tasks for data-center network (DCN) components, such as switches and routers, which are used to efficiently manage and direct network traffic. Packet classification involves examining packet headers to identify and apply specific policies, such as access control or quality of service; while packet forwarding determines the next hop for each packet based on routing tables. Ternary contentaddressable memory (TCAM) facilitates these two tasks by performing a parallel in-memory search to compare an incoming packet’s header against the rules stored in TCAM. TCAM provides fast parallel-lookup functionality making it an indispensable foundation IP for DCN ASICs. TCAM do have one major cost, the parallel all-entry search is power intensive; this

👥 作者与机构

Sushil Kumar*1, Gajanan Jedhe*1, Chetan Deshpande*1, Agastya Gogoi1,

Phoebe Su2, Kim Soon Jway3, TzeYing Seoh3 MediaTek, San Jose, CA MediaTek, Hsinchu, Taiwan 3 MediaTek, Singapore, Singapore 1 2 *Equally Credited Authors (ECAs) Packet