Matches in SemOpenAlex for { <https://semopenalex.org/work/W4386763995> ?p ?o ?g. }
Showing items 1 to 66 of
66
with 100 items per page.
- W4386763995 abstract "Through the attention mechanism, Transformer-based large-scale deep neural networks (LSDNNs) have demonstrated remarkable achievements in artificial intelligence applications such as natural language processing and computer vision. The matrix-matrix multiplication operation (MMMO) in Transformer makes data movement dominate the inference overhead over computation. A solution for efficient data movement during Transformer inference is to embed arithmetic logic units (ALUs) into the memory array, hence an accelerator-in-memory architecture (AIMA). Existing work along this direction has not considered the heterogeneity of parallelism and resource requirements among Transformer layers. This increases the inference latency and lowers the resource utilization, which is critical for the embedded systems domain. To this end, we propose HAIMA, a hybrid AIMA and the parallel dataflow for Transformer, which exploit the cooperation between SRAM and DRAM to accelerate different MMMOs. Compared to the state-of-the-art Newton and TransPIM, our proposed hardware-software co-design achieves 1.4x-1.5x speedup, and solves the problem of resource under-utilization when DRAM-based AIMA performs the light-weight MMMOs." @default.
- W4386763995 created "2023-09-16" @default.
- W4386763995 creator A5006345319 @default.
- W4386763995 creator A5039639491 @default.
- W4386763995 creator A5068477244 @default.
- W4386763995 creator A5072023674 @default.
- W4386763995 creator A5078793726 @default.
- W4386763995 creator A5090246160 @default.
- W4386763995 date "2023-07-09" @default.
- W4386763995 modified "2023-09-26" @default.
- W4386763995 title "HAIMA: A Hybrid SRAM and DRAM Accelerator-in-Memory Architecture for Transformer" @default.
- W4386763995 cites W2801000640 @default.
- W4386763995 cites W2963339397 @default.
- W4386763995 cites W2963926728 @default.
- W4386763995 cites W3034999214 @default.
- W4386763995 cites W3097655915 @default.
- W4386763995 cites W3135906938 @default.
- W4386763995 cites W3138516171 @default.
- W4386763995 cites W4225739994 @default.
- W4386763995 cites W4280496502 @default.
- W4386763995 cites W4283835962 @default.
- W4386763995 doi "https://doi.org/10.1109/dac56929.2023.10247913" @default.
- W4386763995 hasPublicationYear "2023" @default.
- W4386763995 type Work @default.
- W4386763995 citedByCount "0" @default.
- W4386763995 crossrefType "proceedings-article" @default.
- W4386763995 hasAuthorship W4386763995A5006345319 @default.
- W4386763995 hasAuthorship W4386763995A5039639491 @default.
- W4386763995 hasAuthorship W4386763995A5068477244 @default.
- W4386763995 hasAuthorship W4386763995A5072023674 @default.
- W4386763995 hasAuthorship W4386763995A5078793726 @default.
- W4386763995 hasAuthorship W4386763995A5090246160 @default.
- W4386763995 hasConcept C118524514 @default.
- W4386763995 hasConcept C149635348 @default.
- W4386763995 hasConcept C154945302 @default.
- W4386763995 hasConcept C173608175 @default.
- W4386763995 hasConcept C2776214188 @default.
- W4386763995 hasConcept C41008148 @default.
- W4386763995 hasConcept C68339613 @default.
- W4386763995 hasConcept C7366592 @default.
- W4386763995 hasConcept C9390403 @default.
- W4386763995 hasConceptScore W4386763995C118524514 @default.
- W4386763995 hasConceptScore W4386763995C149635348 @default.
- W4386763995 hasConceptScore W4386763995C154945302 @default.
- W4386763995 hasConceptScore W4386763995C173608175 @default.
- W4386763995 hasConceptScore W4386763995C2776214188 @default.
- W4386763995 hasConceptScore W4386763995C41008148 @default.
- W4386763995 hasConceptScore W4386763995C68339613 @default.
- W4386763995 hasConceptScore W4386763995C7366592 @default.
- W4386763995 hasConceptScore W4386763995C9390403 @default.
- W4386763995 hasLocation W43867639951 @default.
- W4386763995 hasOpenAccess W4386763995 @default.
- W4386763995 hasPrimaryLocation W43867639951 @default.
- W4386763995 hasRelatedWork W1509211761 @default.
- W4386763995 hasRelatedWork W1531488649 @default.
- W4386763995 hasRelatedWork W1583465708 @default.
- W4386763995 hasRelatedWork W1585350690 @default.
- W4386763995 hasRelatedWork W2133693067 @default.
- W4386763995 hasRelatedWork W2366027386 @default.
- W4386763995 hasRelatedWork W2391299576 @default.
- W4386763995 hasRelatedWork W2582456645 @default.
- W4386763995 hasRelatedWork W3037767301 @default.
- W4386763995 hasRelatedWork W2479014312 @default.
- W4386763995 isParatext "false" @default.
- W4386763995 isRetracted "false" @default.
- W4386763995 workType "article" @default.