Matches in SemOpenAlex for { <https://semopenalex.org/work/W4280496502> ?p ?o ?g. }
- W4280496502 abstract "Transformer-based models are state-of-the-art for many machine learning (ML) tasks. Executing Transformer usually requires a long execution time due to the large memory footprint and the low data reuse rate, stressing the memory system while under-utilizing the computing resources. Memory-based processing technologies, including processing in-memory (PIM) and near-memory computing (NMC), are promising to accelerate Transformer since they provide high memory bandwidth utilization and extensive computation parallelism. However, the previous memory-based ML accelerators mainly target at optimizing dataflow and hardware for compute-intensive ML models (e.g., CNNs), which do not fit the memory-intensive characteristics of Transformer. In this work, we propose TransPIM, a memory-based acceleration for Transformer using software and hardware co-design. In the software-level, TransPIM adopts a token-based dataflow to avoid the expensive inter-layer data movements introduced by previous layer-based dataflow. In the hardware-level, TransPIM introduces lightweight modifications in the conventional high bandwidth memory (HBM) architecture to support PIM-NMC hybrid processing and efficient data communication for accelerating Transformer-based models. Our experiments show that TransPIM is 3.7× to 9.1× faster than existing memory-based acceleration. As compared to conventional accelerators, TransPIM is 22.1× to 114.9× faster than GPUs and provides 2.0× more throughput than existing ASIC-based accelerators." @default.
- W4280496502 created "2022-05-22" @default.
- W4280496502 creator A5025573294 @default.
- W4280496502 creator A5036778557 @default.
- W4280496502 creator A5039571679 @default.
- W4280496502 creator A5072579616 @default.
- W4280496502 date "2022-04-01" @default.
- W4280496502 modified "2023-10-04" @default.
- W4280496502 title "TransPIM: A Memory-based Acceleration via Software-Hardware Co-Design for Transformer" @default.
- W4280496502 cites W1981943579 @default.
- W4280496502 cites W2034861439 @default.
- W4280496502 cites W2514838290 @default.
- W4280496502 cites W2606722458 @default.
- W4280496502 cites W2761132374 @default.
- W4280496502 cites W2765234579 @default.
- W4280496502 cites W2766489088 @default.
- W4280496502 cites W2791186466 @default.
- W4280496502 cites W2801000640 @default.
- W4280496502 cites W2896090304 @default.
- W4280496502 cites W2909331201 @default.
- W4280496502 cites W2940862705 @default.
- W4280496502 cites W2949591530 @default.
- W4280496502 cites W2949989598 @default.
- W4280496502 cites W2963339397 @default.
- W4280496502 cites W2963926728 @default.
- W4280496502 cites W2979874885 @default.
- W4280496502 cites W2980200167 @default.
- W4280496502 cites W2980688670 @default.
- W4280496502 cites W3016166938 @default.
- W4280496502 cites W3093557412 @default.
- W4280496502 cites W3096609285 @default.
- W4280496502 cites W3100710793 @default.
- W4280496502 cites W3100985894 @default.
- W4280496502 cites W3103415979 @default.
- W4280496502 cites W3103837983 @default.
- W4280496502 cites W3130716829 @default.
- W4280496502 cites W3133347161 @default.
- W4280496502 cites W3134274954 @default.
- W4280496502 cites W3146763006 @default.
- W4280496502 cites W3205088407 @default.
- W4280496502 cites W3207087741 @default.
- W4280496502 cites W3213412675 @default.
- W4280496502 cites W4288083528 @default.
- W4280496502 doi "https://doi.org/10.1109/hpca53966.2022.00082" @default.
- W4280496502 hasPublicationYear "2022" @default.
- W4280496502 type Work @default.
- W4280496502 citedByCount "6" @default.
- W4280496502 countsByYear W42804965022023 @default.
- W4280496502 crossrefType "proceedings-article" @default.
- W4280496502 hasAuthorship W4280496502A5025573294 @default.
- W4280496502 hasAuthorship W4280496502A5036778557 @default.
- W4280496502 hasAuthorship W4280496502A5039571679 @default.
- W4280496502 hasAuthorship W4280496502A5072579616 @default.
- W4280496502 hasConcept C111919701 @default.
- W4280496502 hasConcept C118524514 @default.
- W4280496502 hasConcept C149635348 @default.
- W4280496502 hasConcept C171675096 @default.
- W4280496502 hasConcept C173608175 @default.
- W4280496502 hasConcept C176649486 @default.
- W4280496502 hasConcept C188045654 @default.
- W4280496502 hasConcept C2777904410 @default.
- W4280496502 hasConcept C41008148 @default.
- W4280496502 hasConcept C63511323 @default.
- W4280496502 hasConcept C74912251 @default.
- W4280496502 hasConcept C82687282 @default.
- W4280496502 hasConcept C93446704 @default.
- W4280496502 hasConcept C9390403 @default.
- W4280496502 hasConcept C96324660 @default.
- W4280496502 hasConcept C98986596 @default.
- W4280496502 hasConceptScore W4280496502C111919701 @default.
- W4280496502 hasConceptScore W4280496502C118524514 @default.
- W4280496502 hasConceptScore W4280496502C149635348 @default.
- W4280496502 hasConceptScore W4280496502C171675096 @default.
- W4280496502 hasConceptScore W4280496502C173608175 @default.
- W4280496502 hasConceptScore W4280496502C176649486 @default.
- W4280496502 hasConceptScore W4280496502C188045654 @default.
- W4280496502 hasConceptScore W4280496502C2777904410 @default.
- W4280496502 hasConceptScore W4280496502C41008148 @default.
- W4280496502 hasConceptScore W4280496502C63511323 @default.
- W4280496502 hasConceptScore W4280496502C74912251 @default.
- W4280496502 hasConceptScore W4280496502C82687282 @default.
- W4280496502 hasConceptScore W4280496502C93446704 @default.
- W4280496502 hasConceptScore W4280496502C9390403 @default.
- W4280496502 hasConceptScore W4280496502C96324660 @default.
- W4280496502 hasConceptScore W4280496502C98986596 @default.
- W4280496502 hasLocation W42804965021 @default.
- W4280496502 hasOpenAccess W4280496502 @default.
- W4280496502 hasPrimaryLocation W42804965021 @default.
- W4280496502 hasRelatedWork W1608814317 @default.
- W4280496502 hasRelatedWork W2138825797 @default.
- W4280496502 hasRelatedWork W2159716314 @default.
- W4280496502 hasRelatedWork W2171298529 @default.
- W4280496502 hasRelatedWork W2345611555 @default.
- W4280496502 hasRelatedWork W2491097902 @default.
- W4280496502 hasRelatedWork W2920825666 @default.
- W4280496502 hasRelatedWork W4236777984 @default.
- W4280496502 hasRelatedWork W4280496502 @default.
- W4280496502 hasRelatedWork W4288419222 @default.
- W4280496502 isParatext "false" @default.
- W4280496502 isRetracted "false" @default.