Fingerprinting Workloads for Reconfigurable Shared Accelerators
TimeTuesday, July 12th6pm - 7pm PDT
LocationLevel 2 Lobby
DescriptionEmbedded systems include many special-purpose heterogeneous accelerators each designed to execute a single-software-kernel. One way to simplify and improve design efficiency is to increase the coverage of accelerators. Previous works introduce complex and time-consuming approaches based on graph-isomorphism to find similarities between many workloads from different domains and design hardware modules that run multiple workloads. We introduce a fingerprint of workload that encapsulates static and dynamic behavior of kernels and uses machine learning methods to find acceleration candidates among different domains that can share the same application-specific-hardware with virtually zero false-negatives and false-positives that can be reduced to 1%.