On the Impact of Device-Level Techniques on Energy-Efficiency of Neural Network Accelerators | Publicación