{"id":"https://openalex.org/W4406521183","doi":"https://doi.org/10.1109/tcsi.2025.3527541","title":"An Efficient Window-Based Vision Transformer Accelerator via Mixed-Granularity Sparsity","display_name":"An Efficient Window-Based Vision Transformer Accelerator via Mixed-Granularity Sparsity","publication_year":2025,"publication_date":"2025-01-17","ids":{"openalex":"https://openalex.org/W4406521183","doi":"https://doi.org/10.1109/tcsi.2025.3527541"},"language":"en","primary_location":{"id":"doi:10.1109/tcsi.2025.3527541","is_oa":false,"landing_page_url":"https://doi.org/10.1109/tcsi.2025.3527541","pdf_url":null,"source":{"id":"https://openalex.org/S116977442","display_name":"IEEE Transactions on Circuits and Systems I Regular Papers","issn_l":"1549-8328","issn":["1549-8328","1558-0806"],"is_oa":false,"is_in_doaj":false,"is_core":true,"host_organization":"https://openalex.org/P4310319808","host_organization_name":"Institute of Electrical and Electronics Engineers","host_organization_lineage":["https://openalex.org/P4310319808"],"host_organization_lineage_names":["Institute of Electrical and Electronics Engineers"],"type":"journal"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"IEEE Transactions on Circuits and Systems I: Regular Papers","raw_type":"journal-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":false,"oa_status":"closed","oa_url":null,"any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5102515657","display_name":"Qiwei Dong","orcid":null},"institutions":[{"id":"https://openalex.org/I881766915","display_name":"Nanjing University","ror":"https://ror.org/01rxvg760","country_code":"CN","type":"education","lineage":["https://openalex.org/I881766915"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Qiwei Dong","raw_affiliation_strings":["School of Electronic Science and Engineering, Nanjing University, Nanjing, China"],"raw_orcid":"https://orcid.org/0009-0005-6848-0375","affiliations":[{"raw_affiliation_string":"School of Electronic Science and Engineering, Nanjing University, Nanjing, China","institution_ids":["https://openalex.org/I881766915"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100347369","display_name":"Siyu Zhang","orcid":"https://orcid.org/0000-0001-5531-8655"},"institutions":[{"id":"https://openalex.org/I881766915","display_name":"Nanjing University","ror":"https://ror.org/01rxvg760","country_code":"CN","type":"education","lineage":["https://openalex.org/I881766915"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Siyu Zhang","raw_affiliation_strings":["School of Electronic Science and Engineering, Nanjing University, Nanjing, China"],"raw_orcid":"https://orcid.org/0000-0001-5531-8655","affiliations":[{"raw_affiliation_string":"School of Electronic Science and Engineering, Nanjing University, Nanjing, China","institution_ids":["https://openalex.org/I881766915"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5100696999","display_name":"Zhongfeng Wang","orcid":"https://orcid.org/0000-0002-7227-4786"},"institutions":[{"id":"https://openalex.org/I881766915","display_name":"Nanjing University","ror":"https://ror.org/01rxvg760","country_code":"CN","type":"education","lineage":["https://openalex.org/I881766915"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Zhongfeng Wang","raw_affiliation_strings":["School of Electronic Science and Engineering, Nanjing University, Nanjing, China"],"raw_orcid":"https://orcid.org/0000-0002-7227-4786","affiliations":[{"raw_affiliation_string":"School of Electronic Science and Engineering, Nanjing University, Nanjing, China","institution_ids":["https://openalex.org/I881766915"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I881766915"],"apc_list":null,"apc_paid":null,"fwci":3.0751,"has_fulltext":false,"cited_by_count":10,"citation_normalized_percentile":{"value":0.9147318,"is_in_top_1_percent":false,"is_in_top_10_percent":true},"cited_by_percentile_year":{"min":97,"max":99},"biblio":{"volume":"72","issue":"9","first_page":"4751","last_page":"4764"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11992","display_name":"CCD and CMOS Imaging Sensors","score":0.9998999834060669,"subfield":{"id":"https://openalex.org/subfields/2208","display_name":"Electrical and Electronic Engineering"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T11992","display_name":"CCD and CMOS Imaging Sensors","score":0.9998999834060669,"subfield":{"id":"https://openalex.org/subfields/2208","display_name":"Electrical and Electronic Engineering"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T13114","display_name":"Image Processing Techniques and Applications","score":0.9997000098228455,"subfield":{"id":"https://openalex.org/subfields/2214","display_name":"Media Technology"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10531","display_name":"Advanced Vision and Imaging","score":0.9979000091552734,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/granularity","display_name":"Granularity","score":0.722456693649292},{"id":"https://openalex.org/keywords/transformer","display_name":"Transformer","score":0.6167778372764587},{"id":"https://openalex.org/keywords/window","display_name":"Window (computing)","score":0.6079427599906921},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.5549797415733337},{"id":"https://openalex.org/keywords/electronic-engineering","display_name":"Electronic engineering","score":0.47444307804107666},{"id":"https://openalex.org/keywords/electrical-engineering","display_name":"Electrical engineering","score":0.3603687882423401},{"id":"https://openalex.org/keywords/engineering","display_name":"Engineering","score":0.2125941812992096},{"id":"https://openalex.org/keywords/voltage","display_name":"Voltage","score":0.19884270429611206}],"concepts":[{"id":"https://openalex.org/C177774035","wikidata":"https://www.wikidata.org/wiki/Q1246948","display_name":"Granularity","level":2,"score":0.722456693649292},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.6167778372764587},{"id":"https://openalex.org/C2778751112","wikidata":"https://www.wikidata.org/wiki/Q835016","display_name":"Window (computing)","level":2,"score":0.6079427599906921},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5549797415733337},{"id":"https://openalex.org/C24326235","wikidata":"https://www.wikidata.org/wiki/Q126095","display_name":"Electronic engineering","level":1,"score":0.47444307804107666},{"id":"https://openalex.org/C119599485","wikidata":"https://www.wikidata.org/wiki/Q43035","display_name":"Electrical engineering","level":1,"score":0.3603687882423401},{"id":"https://openalex.org/C127413603","wikidata":"https://www.wikidata.org/wiki/Q11023","display_name":"Engineering","level":0,"score":0.2125941812992096},{"id":"https://openalex.org/C165801399","wikidata":"https://www.wikidata.org/wiki/Q25428","display_name":"Voltage","level":2,"score":0.19884270429611206},{"id":"https://openalex.org/C111919701","wikidata":"https://www.wikidata.org/wiki/Q9135","display_name":"Operating system","level":1,"score":0.0}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1109/tcsi.2025.3527541","is_oa":false,"landing_page_url":"https://doi.org/10.1109/tcsi.2025.3527541","pdf_url":null,"source":{"id":"https://openalex.org/S116977442","display_name":"IEEE Transactions on Circuits and Systems I Regular Papers","issn_l":"1549-8328","issn":["1549-8328","1558-0806"],"is_oa":false,"is_in_doaj":false,"is_core":true,"host_organization":"https://openalex.org/P4310319808","host_organization_name":"Institute of Electrical and Electronics Engineers","host_organization_lineage":["https://openalex.org/P4310319808"],"host_organization_lineage_names":["Institute of Electrical and Electronics Engineers"],"type":"journal"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"IEEE Transactions on Circuits and Systems I: Regular Papers","raw_type":"journal-article"}],"best_oa_location":null,"sustainable_development_goals":[{"display_name":"Affordable and clean energy","id":"https://metadata.un.org/sdg/7","score":0.47999998927116394}],"awards":[{"id":"https://openalex.org/G6422954184","display_name":null,"funder_award_id":"2022YFB4400600","funder_id":"https://openalex.org/F4320335777","funder_display_name":"National Key Research and Development Program of China"}],"funders":[{"id":"https://openalex.org/F4320335777","display_name":"National Key Research and Development Program of China","ror":null}],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":44,"referenced_works":["https://openalex.org/W1861492603","https://openalex.org/W2108598243","https://openalex.org/W2899663614","https://openalex.org/W2963150697","https://openalex.org/W3096609285","https://openalex.org/W3138516171","https://openalex.org/W3159727696","https://openalex.org/W3170841864","https://openalex.org/W3175199633","https://openalex.org/W3188427387","https://openalex.org/W3206453033","https://openalex.org/W3207918547","https://openalex.org/W4214493665","https://openalex.org/W4214588794","https://openalex.org/W4214686755","https://openalex.org/W4280595844","https://openalex.org/W4283811336","https://openalex.org/W4285601701","https://openalex.org/W4285821851","https://openalex.org/W4291653336","https://openalex.org/W4293025165","https://openalex.org/W4308479898","https://openalex.org/W4312290555","https://openalex.org/W4312340826","https://openalex.org/W4313134080","https://openalex.org/W4360831786","https://openalex.org/W4360832001","https://openalex.org/W4362500802","https://openalex.org/W4378805624","https://openalex.org/W4379116023","https://openalex.org/W4379116138","https://openalex.org/W4382317631","https://openalex.org/W4385187240","https://openalex.org/W4385245566","https://openalex.org/W4385768233","https://openalex.org/W4386065441","https://openalex.org/W4386065588","https://openalex.org/W4386072014","https://openalex.org/W4387789586","https://openalex.org/W4389337739","https://openalex.org/W4390190613","https://openalex.org/W4392341509","https://openalex.org/W4393141090","https://openalex.org/W4401357692"],"related_works":["https://openalex.org/W2748952813","https://openalex.org/W2931688134","https://openalex.org/W2377919138","https://openalex.org/W2378857091","https://openalex.org/W103652678","https://openalex.org/W4226090359","https://openalex.org/W2059697060","https://openalex.org/W936373746","https://openalex.org/W2975817033","https://openalex.org/W2864363823"],"abstract_inverted_index":{"Vision":[0],"Transformers":[1],"(ViTs)":[2],"have":[3],"achieved":[4],"excellent":[5],"performance":[6],"on":[7,69],"various":[8],"computer":[9],"vision":[10],"tasks,":[11],"while":[12],"their":[13,58],"high":[14],"computation":[15],"and":[16,109,118,137,168,216],"memory":[17],"costs":[18],"pose":[19],"challenges":[20],"for":[21,202],"practical":[22],"deployment.":[23],"To":[24],"address":[25],"this":[26,73],"issue,":[27],"token-level":[28],"pruning":[29,53,113],"is":[30,99],"used":[31],"as":[32],"an":[33,77],"effective":[34],"method":[35],"to":[36,46,54,82,114,143,158,198],"compress":[37],"ViTs,":[38],"discarding":[39],"unimportant":[40],"image":[41],"tokens":[42],"that":[43],"contribute":[44],"little":[45],"predictions.":[47],"However,":[48],"directly":[49],"applying":[50],"unstructured":[51],"token":[52,112],"window-based":[55,84,145],"ViTs":[56,85],"damages":[57],"regular":[59],"feature":[60],"map":[61],"structure,":[62],"resulting":[63],"in":[64,212],"load":[65],"imbalance":[66],"when":[67],"deployed":[68],"mobile":[70],"devices.":[71],"In":[72],"work,":[74],"we":[75,126,151],"propose":[76],"efficient":[78],"algorithm-hardware":[79],"co-optimized":[80],"framework":[81],"accelerate":[83],"via":[86],"adaptive":[87],"Mixed-Granularity":[88],"Sparsity":[89],"(MGS).":[90],"At":[91,122],"the":[92,103,123,165],"algorithm":[93,98,184],"level,":[94,125],"a":[95,128,133,153],"hardware-friendly":[96],"MGS":[97,183],"developed":[100],"by":[101],"integrating":[102],"inherent":[104],"sparsity,":[105],"global":[106],"window":[107],"pruning,":[108],"local":[110],"N:M":[111],"balance":[115],"model":[116],"accuracy":[117,191],"its":[119],"computational":[120,178],"complexity.":[121],"hardware":[124],"present":[127],"dedicated":[129],"accelerator":[130],"equipped":[131],"with":[132,176],"sparse":[134],"computing":[135],"core":[136],"two":[138],"lightweight":[139],"auxiliary":[140],"processing":[141,166],"units":[142],"execute":[144],"calculations":[146],"efficiently":[147],"using":[148],"MGS.":[149],"Additionally,":[150],"devise":[152],"dynamic":[154],"pipeline":[155],"interleaving":[156],"dataflow":[157],"achieve":[159,186],"on-chip":[160],"layer":[161],"fusion,":[162],"which":[163],"reduces":[164],"latency":[167],"maximizes":[169],"data":[170],"reuse.":[171],"Experimental":[172],"results":[173],"demonstrate":[174],"that,":[175],"similar":[177],"complexity,":[179],"our":[180,204],"highly":[181],"structured":[182],"can":[185,206],"comparable":[187],"or":[188],"even":[189],"better":[190],"than":[192],"previous":[193],"compression":[194],"methods.":[195],"Moreover,":[196],"compared":[197],"existing":[199],"FPGA-based":[200],"accelerators":[201],"Transformers,":[203],"design":[205],"achieve$1.80~\\sim":[207],"~6.52\\times":[208],"$and$1.16~\\sim":[209],"~12.05\\times":[210],"$improvements":[211],"terms":[213],"of":[214],"throughput":[215],"energy":[217],"efficiency,":[218],"respectively.":[219]},"counts_by_year":[{"year":2026,"cited_by_count":6},{"year":2025,"cited_by_count":4}],"updated_date":"2026-07-29T09:40:50.615796","created_date":"2025-10-10T00:00:00"}