Advancements in Memory-Centric Computing and Interconnect Topologies

The field of computer architecture is witnessing a significant shift towards memory-centric computing, with a focus on improving memory tiering performance, inter-chiplet interconnect topologies, and 3D MPSoC designs. Researchers are exploring innovative approaches to address the memory problem in modern computing systems, including the use of Bayesian Optimization to optimize memory tiering parameters and the development of new interconnect topologies such as the FoldedHexaTorus. Additionally, there is a growing interest in printed and flexible electronics, which offer advantages in terms of form factor, bio-compatibility, and sustainability. Noteworthy papers in this area include:

  • A paper that improves memory tiering performance by using application behavior knowledge to set parameters, resulting in a 2x performance improvement over default configurations.
  • A paper that proposes the FoldedHexaTorus inter-chiplet interconnect topology, which achieves higher throughput and lower latency than state-of-the-art topologies.
  • A paper that discusses the benefits of moving to a memory-centric design and execution paradigm, which can solve major challenges in memory technology scaling and system performance.

Sources

From Good to Great: Improving Memory Tiering Performance Through Parameter Tuning

FoldedHexaTorus: An Inter-Chiplet Interconnect Topology for Chiplet-based Systems using Organic and Glass Substrates

3D MPSoC with On-Chip Cache Support -- Design and Exploitation

Computing with Printed and Flexible Electronics

Memory-Centric Computing: Solving Computing's Memory Problem

Built with on top of