The Processing Using Memory Paradigm:In-DRAM Bulk Copy, Initialization,\n Bitwise AND and OR

In existing systems, the off-chip memory interface allows the memory\ncontroller to perform only read or write operations. Therefore, to perform any\noperation, the processor must first read the source data and then write the\nresult back to memory after performing the operation. This approach consumes\nhigh latency, bandwidth, and energy for operations that work on a large amount\nof data.

Several works have proposed techniques to process data near memory by\nadding a small amount of compute logic closer to the main memory chips. In this\narticle, we describe two techniques proposed by recent works that take this\napproach of processing in memory further by exploiting the underlying operation\nof the main memory technology to perform more complex tasks. First, we describe\nRowClone, a mechanism that exploits DRAM technology to perform bulk copy and\ninitialization operations completely inside main memory.

We then describe a\ncomplementary work that uses DRAM to perform bulk bitwise AND and OR operations\ninside main memory. These two techniques significantly improve the performance\nand energy efficiency of the respective operations.\n