LOCAL: Low-Complex Mapping Algorithm for Spatial DNN Accelerators

Deep neural networks are a promising solution for applications that solve problems based on learning data sets. DNN accelerators solved the processing bottleneck as a domain-specific processor, and like other hardware solutions, there must be exact compatibility between the accelerator and other software components, especially the compiler. This paper presents a LOCAL (Low-Complexity mapping Algorithm) that is favorable to use at the compiler level to perform mapping operations in one pass with low computation time and energy consumption. To clarify the problem's scope, we first introduce a formal definition of the design space then we present the LOCAL algorithm's idea. The simulation results show 2x to 38x improvements in runtime with lower energy consumption compared to the known proposed dataflow mechanisms.