GreenArmy：綠色微雲伺服系統晶片平台技術-總計畫( I )

Abstract

此計畫基於螞蟻雄兵的概念，以大量嵌入式處理器取代x86伺服器，提供大量計算能力，形成晶片微雲伺服器。本計畫規劃建構一個整合超越超多核心系統設計的綠色微雲伺服系統晶片架構、系統軟體與節能電路的前瞻SoC整合技術，亦即是在著重於大規模的晶片系統伺服器架構。並結合醫療電子、綠能電子、或車用電子等應用領域。此『GreenArmy : 綠色微雲伺服系統晶片平台技術』研究目標包括：I. 超越多核心架構與先進記憶體架構技術：發展下世代高度擴放性架構以及可調整分割的NOC，並建立多核心ESL平台。此外與旺宏電子合作利用變相化記憶體全力發展支援省電之晶片系統伺服器。II. 發展電源與散熱管理系統：關鍵在於發展適用於晶片系統伺服器之處理器運算虛擬化與記憶體虛擬化技術、平行化系統與工具以及軟體層電源與散熱最佳化管理。III. 開發前瞻電路技術：提供前瞻處理器溫感電路與電源管理設計技術，並開發前瞻NOC路由器技術整合記憶體與NOC。最後利用CIC Morpack SiP技術做到多顆晶片之3D堆疊整合。IV. 跨領域系統應用軟體：結合具有醫學資訊之軟體應用背景老師，與生醫資訊專長教授醫生合作，嘗試跨領域整合，將最複雜計算生理訊號分析(應用於及時診斷)逐步移植本計畫所設定之晶片系統。

This project replaces the prevalent x86 server with a cloud computing inspired server-on-chip design that incorporates a large quantity of on-chip embedded processors to provide a boost to computing power. We propose the GreenArmy server-on-chip architecture that integrates a many-many-core embedded processor design, and an SoC technology to integrate system software and power management circuits. We focus on the larger architectural aspects of server-on-chip, instead of the designs of each individual CPU cores. We have also taken in to account that the server-on-chip design will be adopted by medical electronics, green electronics and automobile electronics.The researching goals of ’GreenArmy: Server-On-Chip ’project includes:I. Memory architecture technology beyond current multi-core design: Develop a cloud-computing inspired server-on-chip with highly scalable next-gen architecture and autonomous NoC. Furthermore, we adopt PCM to achieve high energy efficiency.II. Power and heat management system: Develop a processor and memory virtualization technologies for Server-on-Chip power management to optimize parallel system and software level power management.III. Advanced circuit design technology: We aim to provide thermal sensor circuits and power management design technology, and integrate memory and NoC through NoC router technology. We will utilize CIC Morpack SiP technology to achieve 3D chip stacking.IV. Cross-field system application software: We will bring in professors in different fields, including those with medical background to implement the server-on-chip design to the complex computation of biological signal analysis.