Memory represents a major bottleneck in modern embedded systems. Traditionally, memory organizationsfor programmable systems assumed a fixed cache hierarchy. Withthe wideningprocessor-memory gap, more aggressive memory technologies and organizations have appeared, allowing customization of a heterogeneous memory architecture tunedfor the application. Howevel; such a processormemory eo-exploration approach critically needs the ability to explicitly capture heterogeneous memory architectures. We present in this paper a language-based approach to explicitly capture the memory subsystem conjiguration, and pelform exploration of the memory architecture to trade-off cost versusperformance. Wepresent a set of experiments using our Memory-AwareArchitectural Description Language to drive the exploration of the memory subsystemfor the TIC621I processor architecture, demonstrating a range of cost andpelformanee attributes.
Prabhat Mishra, Peter Grun, Nikil D. Dutt, Alexand