The use of a reconfigurable functional cache in a digital signal processor: power and performance

Due to the computationally intensive nature of the tasks that digital signal processors (DSP) are required to perform it is desirable to decrease the time required to execute these tasks. Minimizing the execution time required for the various algorithms that are commonly and frequently executed (ex: FIR filters) will improve the overall performance. It is known that hardware is able to execute algorithms faster than software, however, due to the size limitations of embedded DSP, not all of the necessary algorithms can be implemented in hardware. A reconfigurable cache architecture in combination with a DSP is proposed as an alternative to increase algorithm performance by using reconfigurable hardware rather than dedicated hardware. Another important issue to consider for embedded processors is the power consumption of the DSP. Due to the fact that most embedded processors operate by battery power, energy efficiency is a necessity. This study looks at the power requirements of a DSP with reconfigurable cache to determine the viability of such an architecture in an embedded system. Others have shown that reconfigurable cache in conjunction with a general purpose processor improves performance for some DSP benchmarks. This study shows that a DSP/reconfigurable cache combination can achieve kernel performance gains ranging from 10-350 times that of a DSP architecture operating alone and can achieve overall benchmark speedups ranging from 1.02 to 1.91 times that of the existing DSP architecture. Further, relative power consumption results show that the power consumption of the reconfigurable architecture is approximately 85 to 95% of the current architecture (5-15% power savings) and attains energy savings ranging from approximately 14 to 50%