Android, the most popular open source mobile platform, attracts a lot of developers who have produced numerous widespread applications (apps). It also draws attackers who have delivered a large amount of malwares to unsuspecting users, due to its open nature. This is not only a threat to national security, but also affect our daily lives. Deep learning has become one of the most popular technologies, and has gained an appreciation to academic and industrial researchers, so it will inevitably become an essential tool to perform complex analysis in a broad application fields. It is appealing to an increasing amount of research ranging from popular topics extraction to Android malware. In this paper, we provide a comprehensive investigation of Android malware detection, and discuss the characteristics of malware and its analysis methods based on deep learning. The secure ecology of Android smartphone based on deep learning is also presented. In addition, research challenges relevant to realworld issues by applying deep learning in smartphone security are discussed, focusing on research issues such as the obtain of optimal parameters, processing of adversarial sample, collection of large scale sample dataset, defence against attack, possession of interpretability and traceability. Our goal is to provide a widespread research guideline to the existing and ongoing efforts via deep learning for smartphone malware, to help researchers better understand the existing work, and to design more and more effective mechanisms to detect smartphone malware.