What is it about?
Bringing computing capabilities to DRAM memories reduces the energy requirements of applications by avoiding data movement throughout the system. This work improves the area and energy efficiency of these solutions by designing together the hardware and software. On one side, it increases the communication bandwidth between computing and memory elements. On the other, it supports the execution of applications optimized for size to enable the computation of many simultaneous operations. Overall, these techniques achieve energy reductions of up to 67% when compared to state-of-the-art designs.
Featured Image
Photo by Bill Fairs on Unsplash
Perspectives
This publication highlights the opportunities of designing hardware and software together to achieve the best behavior of novel solutions computing at the DRAM memory.
Rafael Medina Morillas
Ecole Polytechnique Federale de Lausanne
Read the Original
This page is a summary of: SideDRAM: Integrating SoftSIMD Datapaths near DRAM Banks for Energy-Efficient Variable Precision Computation, ACM Transactions on Embedded Computing Systems, August 2025, ACM (Association for Computing Machinery),
DOI: 10.1145/3762641.
You can read the full text:
Resources
Contributors
The following have contributed to this page
