We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo

Computer Science > Hardware Architecture

Title: HexaMesh: Scaling to Hundreds of Chiplets with an Optimized Chiplet Arrangement

Abstract: 2.5D integration is an important technique to tackle the growing cost of manufacturing chips in advanced technology nodes. This poses the challenge of providing high-performance inter-chiplet interconnects (ICIs). As the number of chiplets grows to tens or hundreds, it becomes infeasible to hand-optimize their arrangement in a way that maximizes the ICI performance. In this paper, we propose HexaMesh, an arrangement of chiplets that outperforms a grid arrangement both in theory (network diameter reduced by 42%; bisection bandwidth improved by 130%) and in practice (latency reduced by 19%; throughput improved by 34%). MexaMesh enables large-scale chiplet designs with high-performance ICIs.
Subjects: Hardware Architecture (cs.AR); Distributed, Parallel, and Cluster Computing (cs.DC); Networking and Internet Architecture (cs.NI)
Cite as: arXiv:2211.13989 [cs.AR]
  (or arXiv:2211.13989v4 [cs.AR] for this version)

Submission history

From: Patrick Iff [view email]
[v1] Fri, 25 Nov 2022 09:54:02 GMT (631kb,D)
[v2] Wed, 14 Dec 2022 14:53:23 GMT (632kb,D)
[v3] Fri, 24 Feb 2023 12:17:43 GMT (632kb,D)
[v4] Sun, 8 Oct 2023 14:27:34 GMT (631kb,D)

Link back to: arXiv, form interface, contact.