ISSN 1000-1239 CN 11-1777/TP

### SER-Tvpack: An SER Estimation-Based Clustering Method for SRAM-Based FPGAs

Xia Jing1,2,3, Wang Tiancheng2, Lü Tao2, Li Huawei2, Kuang Jishun1

1. 1(College of Information Science and Engineering, Hunan University, Changsha 410082) ;2(State Key Laboratory of Computer Architecture (Institute of Computing Technology, Chinese Academy of Sciences), Beijing 100190)
• Online:2014-08-15

Abstract: With the widely use of SRAM-based FPGA (SFPGA) in various fields, reliability becomes increasingly an important concern in SFPGAs. We propose an SER (soft error rate) estimation-based clustering method, namely SER-Tvpack, by adding SER as a reliability factor to the cost function. Combining EPP (error propagation probability) and estimated NER (node error rate) by using the existing ISPL metric, which has been shown to predict every post-placement wirelength accurately, we can estimate the NER factor and get the estimated SER in the clustering stage. According to the fact that the SER of inter-CLBs nets is much higher than that inside CLBs, SER-Tvpack reduces the soft fault rate (SFR) by the means of absorbing high SER nets into the CLBs as much as possible and leaving the low SER nets out of the CLBs. Experimental results show that the proposed SER-Tvpack reduces SFR by 14.5% compared with the baseline T-Vpack, while the previous F-Tvpack reduces SFR by 4.11%. Furthermore, it achieves better performance for reducing the critical path delay by 2.31% in comparison with F-Tvpack, with only 0.04% area overhead increase.

CLC Number: