Efficient Hardware Architectures for Accelerating Deep Neural Networks: Survey