1 paper
Lennart Bamberg, Ardalan Najafi, Alberto Garcia-Ortiz
Specialized compute blocks have been developed for efficient DNN execution. However, due to the vast amount of data and parameter movements, the interconnects and on-chip memories…