Recent AI technologies are so advanced that human can not know how AI comes up with its decisions, which is a barrier for AI to be applied on mission critical applications such as self-driving car. This research develops an explainable AI (XAI) by combining Bayesian machine learning technique with neural networks. Further, we develop hardware accelerators for XAI by using gate level self-synchronous circuit and a bit-serial architecture, whose performance will be demonstrated by humanoid robot control.