Management of Cloud Data in Heterogeneously Integrated Main Memory

High-density nonvolatile memories (NVMs) are promising in the big data and deep-learning applications. To overcome the disadvantages of NVMs, they are often configured with volatile memories. This is especially important for cloud servers because of memory demands of virtual machines. We thereby propose a memory management method using 3D heterogeneous integration. We achieve an efficient memory management by utilizing the high-bandwidth enabled by the 3D integration. Average latency is improved by 30% over non-migration scheme.