{"id":"https://openalex.org/W4386736489","doi":"https://doi.org/10.1109/tvlsi.2023.3305569","title":"An Efficient Training Accelerator for Transformers With Hardware-Algorithm Co-Optimization","display_name":"An Efficient Training Accelerator for Transformers With Hardware-Algorithm Co-Optimization","publication_year":2023,"publication_date":"2023-09-14","ids":{"openalex":"https://openalex.org/W4386736489","doi":"https://doi.org/10.1109/tvlsi.2023.3305569"},"language":"en","primary_location":{"id":"doi:10.1109/tvlsi.2023.3305569","is_oa":false,"landing_page_url":"https://doi.org/10.1109/tvlsi.2023.3305569","pdf_url":null,"source":{"id":"https://openalex.org/S37538908","display_name":"IEEE Transactions on Very Large Scale Integration (VLSI) Systems","issn_l":"1063-8210","issn":["1063-8210","1557-9999"],"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 Very Large Scale Integration (VLSI) Systems","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/A5090170161","display_name":"Haikuo Shao","orcid":"https://orcid.org/0009-0008-6965-3436"},"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":"Haikuo Shao","raw_affiliation_strings":["School of Electronic Science and Engineering, Nanjing University, Nanjing, China"],"raw_orcid":"https://orcid.org/0009-0008-6965-3436","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/A5059107361","display_name":"Jinming Lu","orcid":"https://orcid.org/0000-0002-7134-6514"},"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":"Jinming Lu","raw_affiliation_strings":["School of Electronic Science and Engineering, Nanjing University, Nanjing, China"],"raw_orcid":"https://orcid.org/0000-0002-7134-6514","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/A5100785858","display_name":"Meiqi Wang","orcid":"https://orcid.org/0000-0001-9553-3640"},"institutions":[{"id":"https://openalex.org/I157773358","display_name":"Sun Yat-sen University","ror":"https://ror.org/0064kty71","country_code":"CN","type":"education","lineage":["https://openalex.org/I157773358"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Meiqi Wang","raw_affiliation_strings":["School of Integrated Circuits, Sun Yat-sen University, Shenzhen, China"],"raw_orcid":"https://orcid.org/0000-0001-9553-3640","affiliations":[{"raw_affiliation_string":"School of Integrated Circuits, Sun Yat-sen University, Shenzhen, China","institution_ids":["https://openalex.org/I157773358"]}]},{"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":4,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":1.9634,"has_fulltext":false,"cited_by_count":16,"citation_normalized_percentile":{"value":0.8654104,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":{"min":95,"max":99},"biblio":{"volume":"31","issue":"11","first_page":"1788","last_page":"1801"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T12808","display_name":"Ferroelectric and Negative Capacitance Devices","score":0.9980999827384949,"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/T12808","display_name":"Ferroelectric and Negative Capacitance Devices","score":0.9980999827384949,"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/T12611","display_name":"Neural Networks and Reservoir Computing","score":0.996999979019165,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10036","display_name":"Advanced Neural Network Applications","score":0.9948999881744385,"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/computer-science","display_name":"Computer science","score":0.7309882640838623},{"id":"https://openalex.org/keywords/computation","display_name":"Computation","score":0.6254850029945374},{"id":"https://openalex.org/keywords/computer-engineering","display_name":"Computer engineering","score":0.5375409126281738},{"id":"https://openalex.org/keywords/transformer","display_name":"Transformer","score":0.5232021808624268},{"id":"https://openalex.org/keywords/computer-hardware","display_name":"Computer hardware","score":0.44751736521720886},{"id":"https://openalex.org/keywords/parallel-computing","display_name":"Parallel computing","score":0.41585880517959595},{"id":"https://openalex.org/keywords/computer-architecture","display_name":"Computer architecture","score":0.3579757809638977},{"id":"https://openalex.org/keywords/algorithm","display_name":"Algorithm","score":0.34538811445236206},{"id":"https://openalex.org/keywords/voltage","display_name":"Voltage","score":0.16476067900657654},{"id":"https://openalex.org/keywords/electrical-engineering","display_name":"Electrical engineering","score":0.11787372827529907},{"id":"https://openalex.org/keywords/engineering","display_name":"Engineering","score":0.10565686225891113}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7309882640838623},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.6254850029945374},{"id":"https://openalex.org/C113775141","wikidata":"https://www.wikidata.org/wiki/Q428691","display_name":"Computer engineering","level":1,"score":0.5375409126281738},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.5232021808624268},{"id":"https://openalex.org/C9390403","wikidata":"https://www.wikidata.org/wiki/Q3966","display_name":"Computer hardware","level":1,"score":0.44751736521720886},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.41585880517959595},{"id":"https://openalex.org/C118524514","wikidata":"https://www.wikidata.org/wiki/Q173212","display_name":"Computer architecture","level":1,"score":0.3579757809638977},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.34538811445236206},{"id":"https://openalex.org/C165801399","wikidata":"https://www.wikidata.org/wiki/Q25428","display_name":"Voltage","level":2,"score":0.16476067900657654},{"id":"https://openalex.org/C119599485","wikidata":"https://www.wikidata.org/wiki/Q43035","display_name":"Electrical engineering","level":1,"score":0.11787372827529907},{"id":"https://openalex.org/C127413603","wikidata":"https://www.wikidata.org/wiki/Q11023","display_name":"Engineering","level":0,"score":0.10565686225891113}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1109/tvlsi.2023.3305569","is_oa":false,"landing_page_url":"https://doi.org/10.1109/tvlsi.2023.3305569","pdf_url":null,"source":{"id":"https://openalex.org/S37538908","display_name":"IEEE Transactions on Very Large Scale Integration (VLSI) Systems","issn_l":"1063-8210","issn":["1063-8210","1557-9999"],"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 Very Large Scale Integration (VLSI) Systems","raw_type":"journal-article"}],"best_oa_location":null,"sustainable_development_goals":[{"id":"https://metadata.un.org/sdg/7","display_name":"Affordable and clean energy","score":0.6800000071525574}],"awards":[{"id":"https://openalex.org/G2250357272","display_name":null,"funder_award_id":"62174084","funder_id":"https://openalex.org/F4320321001","funder_display_name":"National Natural Science Foundation of China"},{"id":"https://openalex.org/G4187776313","display_name":null,"funder_award_id":"2022YFB4400604","funder_id":"https://openalex.org/F4320335777","funder_display_name":"National Key Research and Development Program of China"}],"funders":[{"id":"https://openalex.org/F4320321001","display_name":"National Natural Science Foundation of China","ror":"https://ror.org/01h0zpd94"},{"id":"https://openalex.org/F4320335777","display_name":"National Key Research and Development Program of China","ror":null}],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":55,"referenced_works":["https://openalex.org/W1632114991","https://openalex.org/W2108598243","https://openalex.org/W2112796928","https://openalex.org/W2194775991","https://openalex.org/W2606722458","https://openalex.org/W2763421725","https://openalex.org/W2892009249","https://openalex.org/W2896457183","https://openalex.org/W2898985762","https://openalex.org/W2921918777","https://openalex.org/W2949833565","https://openalex.org/W2949920209","https://openalex.org/W2965373594","https://openalex.org/W2996428491","https://openalex.org/W3016542674","https://openalex.org/W3034368386","https://openalex.org/W3035016149","https://openalex.org/W3035232708","https://openalex.org/W3048705738","https://openalex.org/W3084623993","https://openalex.org/W3089022192","https://openalex.org/W3097528158","https://openalex.org/W3104229680","https://openalex.org/W3118608800","https://openalex.org/W3128120039","https://openalex.org/W3138154797","https://openalex.org/W3138516171","https://openalex.org/W3161873870","https://openalex.org/W3206774775","https://openalex.org/W3208821873","https://openalex.org/W3213775513","https://openalex.org/W4210823275","https://openalex.org/W4220865834","https://openalex.org/W4221149185","https://openalex.org/W4287118909","https://openalex.org/W4288346545","https://openalex.org/W4292779060","https://openalex.org/W4298422451","https://openalex.org/W4319996342","https://openalex.org/W4385245566","https://openalex.org/W6727099177","https://openalex.org/W6739901393","https://openalex.org/W6745245109","https://openalex.org/W6751609549","https://openalex.org/W6755207826","https://openalex.org/W6755406732","https://openalex.org/W6763509872","https://openalex.org/W6763653508","https://openalex.org/W6766673545","https://openalex.org/W6768021236","https://openalex.org/W6778883912","https://openalex.org/W6796815506","https://openalex.org/W6808152985","https://openalex.org/W6809871064","https://openalex.org/W6810218207"],"related_works":["https://openalex.org/W2115140794","https://openalex.org/W2041120224","https://openalex.org/W4312275919","https://openalex.org/W2067902980","https://openalex.org/W1504320321","https://openalex.org/W3154635860","https://openalex.org/W2338293335","https://openalex.org/W2039378765","https://openalex.org/W2613115449","https://openalex.org/W2548529098"],"abstract_inverted_index":{"Transformers":[0,10,28],"have":[1,148],"achieved":[2],"significant":[3],"success":[4],"in":[5,142,209],"deep":[6],"learning,":[7],"and":[8,23,43,77,86,92,138,150,174,206,213,234],"training":[9,29,69,160,185],"efficiently":[11],"on":[12,30,74,125,176],"resource-constrained":[13],"platforms":[14,32],"has":[15],"been":[16],"attracting":[17],"continuous":[18],"attention":[19],"for":[20,57,134],"domain":[21],"adaptions":[22],"privacy":[24],"concerns.":[25],"However,":[26],"deploying":[27],"these":[31,49],"is":[33,71,95,164],"still":[34],"challenging":[35],"due":[36],"to":[37,97,119,153],"its":[38],"dynamic":[39],"workloads,":[40],"intensive":[41],"computations,":[42],"massive":[44],"memory":[45,87],"accesses.":[46],"To":[47],"address":[48],"issues,":[50],"we":[51,130],"propose":[52],"an":[53],"Efficient":[54],"Training":[55],"Accelerator":[56],"TRansformers":[58],"(TRETA)":[59],"through":[60],"a":[61,66,75,90],"hardware-algorithm":[62],"co-optimization":[63],"strategy.":[64],"First,":[65],"hardware-friendly":[67],"mixed-precision":[68],"algorithm":[70],"presented":[72],"based":[73],"compact":[76],"efficient":[78],"data":[79],"format,":[80],"which":[81],"significantly":[82],"reduces":[83],"the":[84,135,188,192,221],"computation":[85,156],"requirements.":[88],"Second,":[89],"flexible":[91,151],"scalable":[93],"architecture":[94],"proposed":[96,222],"achieve":[98,199],"high":[99],"utilization":[100],"of":[101,158,211],"computing":[102],"resources":[103],"when":[104,122],"processing":[105],"arbitrary":[106],"irregular":[107,116],"general":[108],"matrix":[109],"multiplication":[110],"(GEMM)":[111],"operations":[112,202],"during":[113],"training.":[114],"These":[115,145],"GEMMs":[117],"lead":[118],"severe":[120],"under-utilization":[121],"simply":[123],"mapped":[124],"traditional":[126],"systolic":[127],"architectures.":[128],"Third,":[129],"develop":[131],"training-oriented":[132],"architectures":[133],"crucial":[136],"Softmax":[137],"layer":[139],"normalization":[140],"functions":[141],"Transformers,":[143],"respectively.":[144,216],"area-efficient":[146],"modules":[147],"unified":[149],"microarchitectures":[152],"meet":[154],"various":[155],"requirements":[157],"different":[159],"phases.":[161],"Finally,":[162],"TRETA":[163,197],"implemented":[165],"under":[166],"Taiwan":[167],"Semiconductor":[168],"Manufacturing":[169],"Company":[170],"(TSMC)":[171],"28-nm":[172],"technology":[173],"evaluated":[175],"multiple":[177],"benchmarks.":[178],"The":[179],"experimental":[180],"results":[181],"show":[182],"that":[183],"our":[184],"framework":[186],"achieves":[187],"same":[189],"accuracy":[190],"as":[191],"full":[193],"precision":[194],"baseline.":[195],"Moreover,":[196],"can":[198],"14.71":[200],"tera":[201],"per":[203],"second":[204],"(TOPS)":[205],"3.31":[207],"TOPS/W":[208],"terms":[210],"throughput":[212],"energy":[214,243],"efficiency,":[215],"Compared":[217],"with":[218],"prior":[219],"arts,":[220],"design":[223],"shows":[224],"1.4\u2013":[225],"<inline-formula":[226,236],"xmlns:mml=\"http://www.w3.org/1998/Math/MathML\"":[227,237],"xmlns:xlink=\"http://www.w3.org/1999/xlink\">":[228,238],"<tex-math":[229,239],"notation=\"LaTeX\">$24.5\\times":[230],"$":[231,241],"</tex-math></inline-formula>":[232,242],"speedup":[233],"1.5\u2013":[235],"notation=\"LaTeX\">$25.4\\times":[240],"efficiency":[244],"improvement.":[245]},"counts_by_year":[{"year":2026,"cited_by_count":1},{"year":2025,"cited_by_count":12},{"year":2024,"cited_by_count":3}],"updated_date":"2026-06-11T09:08:48.828518","created_date":"2025-10-10T00:00:00"}