⚡ 本页包含 AI 生成的分析内容,仅供参考
该论文提出了一种14位动态命中逻辑方案,并集成了8Kb SRAM,用于zEnterpriseTM处理器的L1缓存查找结构。通过采用“搜索命中”方案(XOR后接AND功能,预充电为未命中)优化了性能、时序和功耗,并利用可编程ABIST引擎实现了SRAM和命中逻辑的联合测试。
logic functions are making their way onto critical path SRAMs in the L1 cache look up structure. Described in this paper is a 14 bit dynamic hit logic scheme with an embedded 8K bit SRAM in IBM’s 45nm SOI [3]. The hit logic uses a “search-for-a-hit” scheme (i.e., XOR’s followed by AND functions, pre-charged to a miss) to provide optimal performance, timing, and power. A custom microcode programmable Array-BuiltIn-Self-Test (ABIST) engine tests both the SRAM and hit logic function jointly, resulting in comprehensive “at-speed” test coverage to guarantee circuit functionality and timing margins. The SRAM is organized as a 64×15b×8W (way, or set) array and uses a 6T SRAM cell (1Read/1Write, 0.462µm2) in a “domino” hierarchal dual read bitline approach [1] In the Load/Store Unit (LSU) of the zEnterprise 196 microprocessor core, this hit
Antonio R Pelella, Yuen H Chan, Bargav Balakrishnan, Pradip Patel,
Daniel Rodko, Richard E Serton IBM Systems and Technology Group, Poughkeepsie, NY With the push to ever higher core frequencies, more