A write coalescing design for write-through cache is proposed to promote the performance of application processors using on-chip buffer. The scheme of write coalition for the local data in the same row of SDRAM(synchronous dynamic random access memory)is designed to improve accessing efficiency by employing the single write mode of SDRAM and on-chip buffer. A coherence scheme for the single or multiple cache data reading or writing is also presented. Simulation for mp3 decoding data is implemented in the RTL(register transmit language)simulation environment for Leon2 processor. The simulation results show that when the on-chip buffers are optimized at 3 rows and 8 columns
average 7.8 words are written into the SDRAM after every row pre-charge
and the accessing efficiency increases from 12% to 19% at 100 MHz
and that the area of the proposed design is 0.263 mm
2
under TSMC 0.
18 μm process.
关键词
Keywords
references
Intel Inc. Intel XScale technology overview [EB/OL].[2009-07011]. http:∥www.intel.com/design/intelxscale/.
Gaisler. Leon SPARC V8 processor [EB/OL].[2009-07-11]. http:∥www.gaisler.com.
LIN Wei, YE Xiaochun, SONG Fenglong, et al. Using write mask to support hybrid write-back and write-through cache policy on many-core architectures[J]. Chinese Journal of Computers, 2008, 31(11): 1918-1928.
KANG S Y, PARK S, JUNG H Y, et al. Performance tradeoffs in using NVRAM write buffer for flash memory-based storage devices[J]. IEEE Transactions on Computers, 2009, 58(6): 44-758.
KANG W Y, SON S H, STANKOVIC J A. Power-aware data buffer cache management in real-time embedded databases[C]∥The 14th IEEE International Conference on Embedded and Real-Time Computing Systems and Applications. Piscataway, NJ, USA: IEEE, 2008: 35-44.