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

Customized hardware accelerators have been developed to provide improvedperformance and efficiency for DNN inference and training. However, theexisting hardware accelerators may not always be suitable for handling variousDNN models as their architecture paradigms and configuration tradeoffs arehighly application-specific. It is important to benchmark the acceleratorcandidates in the earliest stage to gather comprehensive performance metricsand locate the potential bottlenecks. Further demands also emerge afterbenchmarking, which require adequate solutions to address the bottlenecks andimprove the current designs for targeted workloads. To achieve these goals, inthis paper, we leverage an automation tool called DNNExplorer for benchmarkingcustomized DNN hardware accelerators and exploring novel accelerator designswith improved performance and efficiency. Key features include (1) directsupport to popular machine learning frameworks for DNN workload analysis andaccurate analytical models for fast accelerator benchmarking; (2) a novelaccelerator design paradigm with high-dimensional design space support andfine-grained adjustability to overcome the existing design drawbacks; and (3) adesign space exploration (DSE) engine to generate optimized accelerators byconsidering targeted AI workloads and available hardware resources. Resultsshow that accelerators adopting the proposed novel paradigm can deliver up to4.2X higher throughput (GOP/s) than the state-of-the-art pipeline design inDNNBuilder and up to 2.0X improved efficiency than the recently publishedgeneric design in HybridDNN given the same DNN model and resource budgets. WithDNNExplorer's benchmarking and exploration features, we can be ahead atbuilding and optimizing customized AI accelerators and enable more efficient AIapplications.

Being-ahead: Benchmarking and Exploring Accelerators for Hardware-Efficient AI Deployment