{"id":"https://openalex.org/W4403919776","doi":"https://doi.org/10.1109/tcad.2024.3488572","title":"Improving Transformer Inference Through Optimized Nonlinear Operations With Quantization-Approximation-Based Strategy","display_name":"Improving Transformer Inference Through Optimized Nonlinear Operations With Quantization-Approximation-Based Strategy","publication_year":2024,"publication_date":"2024-10-30","ids":{"openalex":"https://openalex.org/W4403919776","doi":"https://doi.org/10.1109/tcad.2024.3488572"},"language":"en","primary_location":{"id":"doi:10.1109/tcad.2024.3488572","is_oa":false,"landing_page_url":"https://doi.org/10.1109/tcad.2024.3488572","pdf_url":null,"source":{"id":"https://openalex.org/S100835903","display_name":"IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems","issn_l":"0278-0070","issn":["0278-0070","1937-4151"],"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 Computer-Aided Design of Integrated Circuits and 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/A5023029393","display_name":"Wenxun Wang","orcid":"https://orcid.org/0009-0007-1999-6441"},"institutions":[{"id":"https://openalex.org/I99065089","display_name":"Tsinghua University","ror":"https://ror.org/03cve4549","country_code":"CN","type":"education","lineage":["https://openalex.org/I99065089"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Wenxun Wang","raw_affiliation_strings":["Department of Electronic Engineering, Tsinghua University, Beijing, China"],"raw_orcid":"https://orcid.org/0009-0007-1999-6441","affiliations":[{"raw_affiliation_string":"Department of Electronic Engineering, Tsinghua University, Beijing, China","institution_ids":["https://openalex.org/I99065089"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5085072777","display_name":"Wenyu Sun","orcid":"https://orcid.org/0000-0002-4793-0972"},"institutions":[{"id":"https://openalex.org/I99065089","display_name":"Tsinghua University","ror":"https://ror.org/03cve4549","country_code":"CN","type":"education","lineage":["https://openalex.org/I99065089"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Wenyu Sun","raw_affiliation_strings":["Department of Electronic Engineering, Tsinghua University, Beijing, China"],"raw_orcid":"https://orcid.org/0000-0002-4793-0972","affiliations":[{"raw_affiliation_string":"Department of Electronic Engineering, Tsinghua University, Beijing, China","institution_ids":["https://openalex.org/I99065089"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5045721867","display_name":"Yongpan Liu","orcid":"https://orcid.org/0000-0002-4892-2309"},"institutions":[{"id":"https://openalex.org/I99065089","display_name":"Tsinghua University","ror":"https://ror.org/03cve4549","country_code":"CN","type":"education","lineage":["https://openalex.org/I99065089"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Yongpan Liu","raw_affiliation_strings":["Department of Electronic Engineering, Tsinghua University, Beijing, China"],"raw_orcid":"https://orcid.org/0000-0002-4892-2309","affiliations":[{"raw_affiliation_string":"Department of Electronic Engineering, Tsinghua University, Beijing, China","institution_ids":["https://openalex.org/I99065089"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I99065089"],"apc_list":null,"apc_paid":null,"fwci":1.1114,"has_fulltext":false,"cited_by_count":4,"citation_normalized_percentile":{"value":0.81802775,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":{"min":94,"max":97},"biblio":{"volume":"44","issue":"4","first_page":"1475","last_page":"1488"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10320","display_name":"Neural Networks and Applications","score":0.9435999989509583,"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"}},"topics":[{"id":"https://openalex.org/T10320","display_name":"Neural Networks and Applications","score":0.9435999989509583,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.7317502498626709},{"id":"https://openalex.org/keywords/transformer","display_name":"Transformer","score":0.5861381888389587},{"id":"https://openalex.org/keywords/quantization","display_name":"Quantization (signal processing)","score":0.5499125719070435},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.48045796155929565},{"id":"https://openalex.org/keywords/algorithm","display_name":"Algorithm","score":0.40551817417144775},{"id":"https://openalex.org/keywords/mathematics","display_name":"Mathematics","score":0.36558806896209717},{"id":"https://openalex.org/keywords/mathematical-optimization","display_name":"Mathematical optimization","score":0.34366995096206665},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.3269667625427246},{"id":"https://openalex.org/keywords/engineering","display_name":"Engineering","score":0.15479031205177307},{"id":"https://openalex.org/keywords/electrical-engineering","display_name":"Electrical engineering","score":0.12571829557418823},{"id":"https://openalex.org/keywords/voltage","display_name":"Voltage","score":0.06532654166221619}],"concepts":[{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.7317502498626709},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.5861381888389587},{"id":"https://openalex.org/C28855332","wikidata":"https://www.wikidata.org/wiki/Q198099","display_name":"Quantization (signal processing)","level":2,"score":0.5499125719070435},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.48045796155929565},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.40551817417144775},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.36558806896209717},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.34366995096206665},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3269667625427246},{"id":"https://openalex.org/C127413603","wikidata":"https://www.wikidata.org/wiki/Q11023","display_name":"Engineering","level":0,"score":0.15479031205177307},{"id":"https://openalex.org/C119599485","wikidata":"https://www.wikidata.org/wiki/Q43035","display_name":"Electrical engineering","level":1,"score":0.12571829557418823},{"id":"https://openalex.org/C165801399","wikidata":"https://www.wikidata.org/wiki/Q25428","display_name":"Voltage","level":2,"score":0.06532654166221619}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1109/tcad.2024.3488572","is_oa":false,"landing_page_url":"https://doi.org/10.1109/tcad.2024.3488572","pdf_url":null,"source":{"id":"https://openalex.org/S100835903","display_name":"IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems","issn_l":"0278-0070","issn":["0278-0070","1937-4151"],"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 Computer-Aided Design of Integrated Circuits and Systems","raw_type":"journal-article"}],"best_oa_location":null,"sustainable_development_goals":[],"awards":[{"id":"https://openalex.org/G2021969373","display_name":null,"funder_award_id":"92267203","funder_id":"https://openalex.org/F4320321001","funder_display_name":"National Natural Science Foundation of China"},{"id":"https://openalex.org/G5685581577","display_name":null,"funder_award_id":"2021ZD0114402","funder_id":"https://openalex.org/F4320329860","funder_display_name":"National Science and Technology Major Project"}],"funders":[{"id":"https://openalex.org/F4320321001","display_name":"National Natural Science Foundation of China","ror":"https://ror.org/01h0zpd94"},{"id":"https://openalex.org/F4320329860","display_name":"National Science and Technology Major Project","ror":null}],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":44,"referenced_works":["https://openalex.org/W2108598243","https://openalex.org/W2111013824","https://openalex.org/W2511929605","https://openalex.org/W2903972532","https://openalex.org/W2965373594","https://openalex.org/W2998183051","https://openalex.org/W2998218113","https://openalex.org/W3004659153","https://openalex.org/W3017024317","https://openalex.org/W3138516171","https://openalex.org/W3159727696","https://openalex.org/W3170841864","https://openalex.org/W3196923642","https://openalex.org/W3206453033","https://openalex.org/W3211525823","https://openalex.org/W4293025109","https://openalex.org/W4385245566","https://openalex.org/W4386764151","https://openalex.org/W6637373629","https://openalex.org/W6638318767","https://openalex.org/W6677580257","https://openalex.org/W6718053083","https://openalex.org/W6720242923","https://openalex.org/W6727099177","https://openalex.org/W6750615492","https://openalex.org/W6751068846","https://openalex.org/W6755207826","https://openalex.org/W6767298317","https://openalex.org/W6776320331","https://openalex.org/W6778883912","https://openalex.org/W6784094891","https://openalex.org/W6788001715","https://openalex.org/W6788135285","https://openalex.org/W6789946416","https://openalex.org/W6790978476","https://openalex.org/W6797854001","https://openalex.org/W6809883161","https://openalex.org/W6811340617","https://openalex.org/W6838322825","https://openalex.org/W6840088539","https://openalex.org/W6847478871","https://openalex.org/W6849928207","https://openalex.org/W6850162387","https://openalex.org/W6850503672"],"related_works":["https://openalex.org/W2051487156","https://openalex.org/W2073681303","https://openalex.org/W2053286651","https://openalex.org/W2055243143","https://openalex.org/W2181743346","https://openalex.org/W2187401768","https://openalex.org/W4206178588","https://openalex.org/W3094491777","https://openalex.org/W3214715529","https://openalex.org/W4287635093"],"abstract_inverted_index":{"Transformers":[0],"have":[1,138],"recently":[2],"shown":[3],"significant":[4],"performance":[5,21],"across":[6,237],"various":[7],"tasks,":[8],"such":[9],"as":[10,68,153],"natural":[11],"language":[12],"processing":[13],"(NLP)":[14],"and":[15,29,45,76,91,114,124,134,143,149,200,223,240],"computer":[16],"vision":[17],"(CV).":[18],"However,":[19],"the":[20,24,49,112,157,177,185],"comes":[22],"at":[23],"cost":[25],"of":[26,51,72,101,111,117,228],"large":[27],"memory":[28,77],"computation":[30,75],"overhead.":[31],"Existing":[32],"researches":[33],"primarily":[34],"focus":[35],"on":[36],"accelerating":[37],"matrix":[38],"multiplication":[39],"(MatMul)":[40],"through":[41],"techniques":[42],"like":[43],"quantization":[44,123,133,137],"pruning,":[46],"notably":[47],"increasing":[48],"proportion":[50],"nonlinear":[52,62,102,118],"operations":[53,63,103],"in":[54,141,225],"inference":[55,236],"runtime.":[56],"Meanwhile,":[57],"previous":[58],"approaches":[59],"designed":[60],"for":[61,128,179,234],"struggle":[64],"with":[65,104],"inefficient":[66],"implementation":[67,100,161],"they":[69],"are":[70,232],"incapable":[71],"achieving":[73],"both":[74,238],"efficiency.":[78],"Additionally,":[79],"these":[80,95],"methods":[81],"often":[82],"require":[83],"retraining":[84,180],"or":[85,181],"fine-tuning":[86],"leading":[87],"to":[88,213,230],"substantial":[89,221],"costs":[90],"inconveniences.":[92],"To":[93],"overcome":[94],"problems,":[96],"we":[97,120],"propose":[98],"efficient":[99,159],"quantization-approximation-based":[105],"strategy.":[106],"Through":[107],"an":[108],"in-depth":[109],"analysis":[110],"dataflow":[113],"data":[115],"distribution":[116],"operations,":[119],"design":[121],"distinct":[122],"approximation":[125,154],"strategies":[126],"tailored":[127],"different":[129],"operations.":[130],"Specifically,":[131],"log2":[132],"power-of-two":[135],"factor":[136],"been":[139],"employed":[140],"Softmax":[142],"LayerNorm,":[144],"complemented":[145],"by":[146],"logarithmic":[147],"function":[148],"low-precision":[150],"statistic":[151],"calculation":[152],"strategies.":[155],"Furthermore,":[156],"proposed":[158],"GeLU":[160],"integrates":[162],"a":[163],"nonuniform":[164],"lookup":[165],"procedure":[166],"alongside":[167],"low-bit-width":[168],"quantization.":[169],"Experimental":[170],"results":[171],"demonstrate":[172],"negligible":[173],"accuracy":[174],"drops":[175],"without":[176],"need":[178],"fine-tuning.":[182],"By":[183],"implementing":[184],"hardware":[186],"design,":[187],"it":[188],"achieves":[189],"<inline-formula":[190,201],"xmlns:mml=\"http://www.w3.org/1998/Math/MathML\"":[191,202],"xmlns:xlink=\"http://www.w3.org/1999/xlink\">":[192,203],"<tex-math":[193,204],"notation=\"LaTeX\">$3.14\\times":[194],"-":[195,206],"6.34\\times":[196],"$":[197,208],"</tex-math></inline-formula>":[198,209],"energy-efficiency":[199],"notation=\"LaTeX\">$3.01\\times":[205],"10.1\\times":[207],"area-efficiency":[210],"improvements":[211],"compared":[212],"state-of-the-art":[214],"application-specific-integrated-circuit":[215],"(ASIC)":[216],"designs.":[217],"In":[218],"system-level":[219],"evaluation,":[220],"speedup":[222],"reductions":[224],"energy":[226],"consumption":[227],"15%":[229],"35%":[231],"achieved":[233],"end-to-end":[235],"GPU":[239],"ASIC":[241],"accelerator":[242],"platforms.":[243]},"counts_by_year":[{"year":2026,"cited_by_count":1},{"year":2025,"cited_by_count":3}],"updated_date":"2025-12-21T01:58:51.020947","created_date":"2025-10-10T00:00:00"}