Advances in Memory Optimization and In-Memory Computing

The field of memory optimization and in-memory computing is moving towards innovative solutions that improve system performance and scalability. Researchers are exploring new techniques to decompose DRAM address mappings, optimize metadata storage management, and leverage profile-guided methods to enhance temporal prefetching. In-memory computing models are being proposed to reduce CPU workload and improve data-intensive computation. Additionally, adaptive hybrid sorting paradigms are being developed to automatically select the most effective sorting algorithm based on real-time monitoring of input data patterns. Notable papers include:

  • Sudoku, which introduces a software-based tool to automatically decompose full DRAM address mappings into component-level functions.
  • Prophet, a hardware-software co-designed framework that optimizes metadata storage management using profile-guided methods.
  • In-Memory Sorting-Searching with Cayley Tree, which proposes a computing model to reduce CPU workload by leveraging in-memory computing and achieving efficient solutions for in-memory searching, computing, and sorting.

Sources

Sudoku: Decomposing DRAM Address Mapping into Component Functions

Profile-Guided Temporal Prefetching

PUL: Pre-load in Software for Caches Wouldn't Always Play Along

In-Memory Sorting-Searching with Cayley Tree

Adaptive Hybrid Sort: Dynamic Strategy Selection for Optimal Sorting Across Diverse Data Distributions

Review of Three Variants of the k-d Tree

Built with on top of