Spiking neural network (SNN) is a promising candidate for neuromorphic system designs. However, the potentials of time-based SNN are not unleashed in realistic cognitive applications due to lack of efficient coding and practical learning schemes. In this work, we aim at a flexible Precise-Time-Based Spiking Neuromorphic Architecture to bridge the gap between hardware and bio-plausibility by developing a holistic solution set spanning time-based coding, learning to decoding. We also for the first time explore the unique advantage of time-based SNN architecture: A possible flexible spatial-temporal information conversion to satisfy various design trade-offs among power, throughput and accuracy. To validate its capability, we evaluate its synaptic connection, speed, power and accuracy w.r.t. the popular rate-coding based SNN in applications like MNIST dataset. Unlike the complicated convolutional neural network (CNN) or deep neural network (DNN) that requires costly hardware resource, we hope that this work may demonstrate the advantages of the time-based spiking neuromorphic architecture system for tasks performed in the ultra low power and resource constrained platforms like mobile and IoT devices.