Đang chuẩn bị liên kết để tải về tài liệu:
Adaptive Techniques for Dynamic Processor Optimization_Theory and Practice Episode 2 Part 4

Tham khảo tài liệu 'adaptive techniques for dynamic processor optimization_theory and practice episode 2 part 4', kỹ thuật - công nghệ, cơ khí - chế tạo máy phục vụ nhu cầu học tập, nghiên cứu và làm việc hiệu quả | Chapter 9 Variability-Aware Frequency Scaling in Multi-Clock Processors 211 Figure 9.2 Delay distributions for Ncp 1 2 10 . Unfortunately determining the number of independent critical paths in a given circuit in order to quantify this effect is not trivial. Correlations between critical path delays occur due to inherent spatial correlations in parameter variations and the overlap of critical paths that pass through one or more of the same gates. To overcome this problem Ncp is redefined to be the effective number of independent critical paths that when inserted into Equation 9.2 will yield a worst-case delay distribution that matches the statistics of the actual worst-case delay distribution of the circuit. The proposed methodology estimates the effective number of independent critical paths for the two kinds of circuits that occur most frequently in processor microarchitectures combinational logic and array structures. This corresponds roughly to the categorization of functional blocks as being either logic or SRAM dominated by Humenay et al. 9 . This methodology improves on the assumptions about the distribution of critical paths that have been made in previous studies. For example Marculescu and Talpes assumed 100 total independent critical paths in a microprocessor and distributed them among blocks proportionally to device count 12 while Humenay et al. assumed that logic stages have only a single critical path and that an array structure has a number of critical paths equal to the product of the number of wordlines and number of bitlines 9 . Liang and Brooks make a similar assumption for register file SRAMs 11 . The proposed model also has the advantage of capturing the effects of almost-critical paths which would not be critical under nominal conditions but are sufficiently close that they could become a 212 Sebastian Herbert Diana Marculescu block s slowest path in the face of variations. The model results presented here assume a 3o of 20 for channel length