Processing in Memory for Big Data Application and Heterogeneous Memory Systems
The primary focus of this research is to explore if simple processing cores embedded within 3D DRAM memories (or Processing-In-Memory) benefit Big Data Analysis applications. Current 3D DRAM implementations use one layer of the 3D stack for logic. This layer will contain memory controllers and other logic needed to manage the memory. However, there is sufficient space to include simple processing cores to perform memory intensive computations. In this research we will evaluate how Big Data applications that rely on MapReduce programming model can benefit from Processing-In-Memory (PIM) technologies. More specifically we will evaluate the number and the types of cores to embed in 3D DRAM to achieve optimal balance between performance and energy consumed while executing Big Data and other emerging applications. We hope to gain a good understanding of what functions can benefit from PIM technologies.