University of Essen
Resource optimization in energy constrained real-time adaptive embedded systems highly depends on accurate energy and throughput estimates of processor peripherals. Such applications require lightweight, accurate mathematical models to profile energy and timing requirements on the go. This paper presents enhanced mathematical models for data cache energy and throughput estimation. The energy and throughput models were found to be within 95% accuracy of per instruction energy model of a processor, and a full system simulator's timing model respectively. Furthermore, the possible application of these models in various scenarios is discussed in this paper.