New computing paradigms are highly demanded in the “Big Data” era to efficiently process, store and extract useful information from overwhelmingly rich amount of data. New computing systems based on large scale memristor circuits emerges as a very promising candidate due to its capability to both store and process information, thus eliminating the von Neumann bottleneck in the conventional complementary metal oxide semiconductor (CMOS) based computers. As the lateral scaling of the device geometry approaching its physical limit, three-dimensional stacking of multiple device layers becomes necessary to further increase the packing density. Moreover, innovations in the 3D circuits design can also provide the memristor system with low power consumptions, low latencies and high throughput. In this work, systematic study based on device engineering, circuit design and fabrication process development was carried out. Highly parallel kernel operations were experimentally demonstrated in the 3D memristor circuits for deep neural network and video processing applications. Finally, a transfer printing method for 2D materials was invented as an enabling technology for future 3D hybrid circuits.