Abstract: As the power consumption is becoming a more dominant factor for IoT and high-performance computing applications, innovations in the architecture are getting attention. Originated from an approach to rebalance the digital and analog domain, in-memory computing（IMC） has a possibility to enable a more efficient computation. On the way to this architecture change, it is vital to build a new infrastructure for analog computation. Its fundamental elements include memory technologies with analog properties optimized and a circuit system for analog computing. In addition, non-volatile memory-based technology can fully exploit the energy efficiency of IMC with its low static power. In this paper, recent researches in these key enabling technologies are reviewed, after an introduction to the principles of IMC. Applications in scenarios such as deep learning will also be discussed, followed by a prospect for future developments.