Neural Network Quantization for Low-Power