A considerable portion of the time spent during databases operation processing consists in moving data around the memory hierarchy rather than actually processing it. The emergence of smart memories, as the new Hybrid Memory Cube (HMC) allows mitigating this memory wall problem, by executing instructions directly inside the memory, reducing data movement. In this work, we discuss the processing of databases inside the HMC. We focus on the select scan operator, because the scanning of columns moves large amounts of data prior to other costly operations like joins (i.e., push-down optimization). Our preliminary results with the tuple / column / vector-at-a-time query engines show performance gains of 30% for tuple-at-a-time, 11% column-at-a-time and 6× in the vector-at-a-time execution when compared to x86.