English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Sparsity is a growing trend in modern DNN models. Existing Sparse-SparseMatrix Multiplication (SpMSpM) accelerators are tailored to a particular SpMSpMdataflow (i.e., Inner Product, Outer Product or Gustavsons), that determinestheir overall efficiency. We demonstrate that this static decision inherentlyresults in a suboptimal dynamic solution. This is because different SpMSpMkernels show varying features (i.e., dimensions, sparsity pattern, sparsitydegree), which makes each dataflow better suited to different data sets. Inthis work we present Flexagon, the first SpMSpM reconfigurable accelerator thatis capable of performing SpMSpM computation by using the particular dataflowthat best matches each case. Flexagon accelerator is based on a novelMerger-Reduction Network (MRN) that unifies the concept of reducing and mergingin the same substrate, increasing efficiency. Additionally, Flexagon alsoincludes a 3-tier memory hierarchy, specifically tailored to the differentaccess characteristics of the input and output compressed matrices. Usingdetailed cycle-level simulation of contemporary DNN models from a variety ofapplication domains, we show that Flexagon achieves average performancebenefits of 4.59x, 1.71x, and 1.35x with respect to the state-of-the-artSIGMA-like, Sparch-like and GAMMA-like accelerators (265% , 67% and 18%,respectively, in terms of average performance/area efficiency).

Flexagon: A Multi-Dataflow Sparse-Sparse Matrix Multiplication Accelerator for Efficient DNN Processing