{"id":"https://openalex.org/W4410582779","doi":"https://doi.org/10.23919/date64628.2025.10992868","title":"Integer Unit-Based Outlier-Aware LLM Accelerator Preserving Numerical Accuracy of FP-FP GEMM","display_name":"Integer Unit-Based Outlier-Aware LLM Accelerator Preserving Numerical Accuracy of FP-FP GEMM","publication_year":2025,"publication_date":"2025-03-31","ids":{"openalex":"https://openalex.org/W4410582779","doi":"https://doi.org/10.23919/date64628.2025.10992868"},"language":"en","primary_location":{"id":"doi:10.23919/date64628.2025.10992868","is_oa":false,"landing_page_url":"https://doi.org/10.23919/date64628.2025.10992868","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2025 Design, Automation & Test in Europe Conference (DATE)","raw_type":"proceedings-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/A5110045339","display_name":"Je\u2010Hun Lee","orcid":null},"institutions":[{"id":"https://openalex.org/I139264467","display_name":"Seoul National University","ror":"https://ror.org/04h9pn542","country_code":"KR","type":"education","lineage":["https://openalex.org/I139264467"]}],"countries":["KR"],"is_corresponding":true,"raw_author_name":"Jehun Lee","raw_affiliation_strings":["Seoul National University,Seoul,Korea"],"affiliations":[{"raw_affiliation_string":"Seoul National University,Seoul,Korea","institution_ids":["https://openalex.org/I139264467"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5003219699","display_name":"Jae\u2010Joon Kim","orcid":"https://orcid.org/0000-0001-5175-8258"},"institutions":[{"id":"https://openalex.org/I139264467","display_name":"Seoul National University","ror":"https://ror.org/04h9pn542","country_code":"KR","type":"education","lineage":["https://openalex.org/I139264467"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Jae-Joon Kim","raw_affiliation_strings":["Seoul National University,Seoul,Korea"],"affiliations":[{"raw_affiliation_string":"Seoul National University,Seoul,Korea","institution_ids":["https://openalex.org/I139264467"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"corresponding_author_ids":["https://openalex.org/A5110045339"],"corresponding_institution_ids":["https://openalex.org/I139264467"],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":{"value":0.11515462,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"1","last_page":"7"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10781","display_name":"Plasma Diagnostics and Applications","score":0.9847000241279602,"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/T10781","display_name":"Plasma Diagnostics and Applications","score":0.9847000241279602,"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/T10558","display_name":"Advancements in Semiconductor Devices and Circuit Design","score":0.9797000288963318,"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/T10346","display_name":"Magnetic confinement fusion research","score":0.9767000079154968,"subfield":{"id":"https://openalex.org/subfields/3106","display_name":"Nuclear and High Energy Physics"},"field":{"id":"https://openalex.org/fields/31","display_name":"Physics and Astronomy"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/outlier","display_name":"Outlier","score":0.6284822225570679},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.6059939861297607},{"id":"https://openalex.org/keywords/integer","display_name":"Integer (computer science)","score":0.5897879004478455},{"id":"https://openalex.org/keywords/unit","display_name":"Unit (ring theory)","score":0.463672399520874},{"id":"https://openalex.org/keywords/algorithm","display_name":"Algorithm","score":0.43399882316589355},{"id":"https://openalex.org/keywords/computational-science","display_name":"Computational science","score":0.40966954827308655},{"id":"https://openalex.org/keywords/parallel-computing","display_name":"Parallel computing","score":0.35617589950561523},{"id":"https://openalex.org/keywords/mathematics","display_name":"Mathematics","score":0.26220741868019104},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.1608242392539978},{"id":"https://openalex.org/keywords/programming-language","display_name":"Programming language","score":0.1452740728855133}],"concepts":[{"id":"https://openalex.org/C79337645","wikidata":"https://www.wikidata.org/wiki/Q779824","display_name":"Outlier","level":2,"score":0.6284822225570679},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6059939861297607},{"id":"https://openalex.org/C97137487","wikidata":"https://www.wikidata.org/wiki/Q729138","display_name":"Integer (computer science)","level":2,"score":0.5897879004478455},{"id":"https://openalex.org/C122637931","wikidata":"https://www.wikidata.org/wiki/Q118084","display_name":"Unit (ring theory)","level":2,"score":0.463672399520874},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.43399882316589355},{"id":"https://openalex.org/C459310","wikidata":"https://www.wikidata.org/wiki/Q117801","display_name":"Computational science","level":1,"score":0.40966954827308655},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.35617589950561523},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.26220741868019104},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.1608242392539978},{"id":"https://openalex.org/C199360897","wikidata":"https://www.wikidata.org/wiki/Q9143","display_name":"Programming language","level":1,"score":0.1452740728855133},{"id":"https://openalex.org/C145420912","wikidata":"https://www.wikidata.org/wiki/Q853077","display_name":"Mathematics education","level":1,"score":0.0}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.23919/date64628.2025.10992868","is_oa":false,"landing_page_url":"https://doi.org/10.23919/date64628.2025.10992868","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2025 Design, Automation & Test in Europe Conference (DATE)","raw_type":"proceedings-article"}],"best_oa_location":null,"sustainable_development_goals":[],"awards":[],"funders":[{"id":"https://openalex.org/F4320321292","display_name":"Seoul National University","ror":"https://ror.org/04h9pn542"},{"id":"https://openalex.org/F4320322202","display_name":"IC Design Education Center","ror":"https://ror.org/005v57z85"}],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":34,"referenced_works":["https://openalex.org/W2001966371","https://openalex.org/W2159211021","https://openalex.org/W2883920103","https://openalex.org/W2923014074","https://openalex.org/W2945652047","https://openalex.org/W2963323070","https://openalex.org/W2998617917","https://openalex.org/W3100985894","https://openalex.org/W3189877953","https://openalex.org/W3190062760","https://openalex.org/W3194676777","https://openalex.org/W4308083739","https://openalex.org/W4366341968","https://openalex.org/W4380874652","https://openalex.org/W4393407021","https://openalex.org/W4400489024","https://openalex.org/W4404133534","https://openalex.org/W6751609549","https://openalex.org/W6755207826","https://openalex.org/W6758823024","https://openalex.org/W6763018175","https://openalex.org/W6782879696","https://openalex.org/W6785530804","https://openalex.org/W6789240164","https://openalex.org/W6838633097","https://openalex.org/W6842258392","https://openalex.org/W6842758154","https://openalex.org/W6846164622","https://openalex.org/W6847420240","https://openalex.org/W6847478871","https://openalex.org/W6852337893","https://openalex.org/W6853048723","https://openalex.org/W6854866820","https://openalex.org/W6857288518"],"related_works":["https://openalex.org/W3006513224","https://openalex.org/W2046456988","https://openalex.org/W2357409937","https://openalex.org/W2510582230","https://openalex.org/W2978674666","https://openalex.org/W2074430941","https://openalex.org/W2113096305","https://openalex.org/W1977636359","https://openalex.org/W2772305933","https://openalex.org/W2580722822"],"abstract_inverted_index":{"The":[0,132],"proliferation":[1],"of":[2,12,25,33,46,89,110,117,145,186],"large":[3],"language":[4],"models":[5],"(LLMs)":[6],"has":[7],"significantly":[8],"heightened":[9],"the":[10,16,20,23,39,59,86,108,143,161,165,183,187],"importance":[11],"quantization":[13,28],"to":[14,50,56,141],"alleviate":[15],"computational":[17],"burden":[18],"given":[19],"surge":[21],"in":[22],"number":[24,130,134],"parameters.":[26],"However,":[27],"often":[29],"targets":[30],"a":[31,34,111,128],"subset":[32],"LLM":[35,81,168],"and":[36,52,73,114,138,149,160,176],"relies":[37],"on":[38],"floating-point":[40],"(FP)":[41],"arithmetic":[42,91,100],"for":[43,58,102],"matrix":[44,104],"multiplication":[45,105],"specific":[47],"subsets,":[48],"leading":[49],"performance":[51,174],"energy":[53,179],"overhead.":[54],"Additionally,":[55],"compensate":[57],"quality":[60],"degradation":[61],"incurred":[62],"by":[63,126],"quantization,":[64],"retraining":[65],"methods":[66],"are":[67],"frequently":[68],"employed,":[69],"demanding":[70],"significant":[71],"efforts":[72],"resources.":[74],"This":[75],"paper":[76],"proposes":[77],"OwL-P,":[78],"an":[79,97,171],"outlier-aware":[80],"inference":[82],"accelerator":[83,169],"which":[84],"preserves":[85],"numerical":[87,184],"accuracy":[88,185],"FP":[90,188],"while":[92,181],"enhancing":[93],"hardware":[94],"efficiency":[95],"with":[96],"integer":[98],"(INT)-based":[99],"unit":[101],"general":[103],"(GEMM),":[106],"through":[107],"use":[109],"shared":[112,139],"exponent":[113],"efficient":[115],"management":[116],"outlier":[118],"data.":[119],"It":[120],"also":[121],"mitigates":[122],"off-chip":[123],"bandwidth":[124],"requirements":[125],"employing":[127],"compressed":[129],"format.":[131],"proposed":[133,166],"format":[135],"leverages":[136],"outliers":[137],"exponents":[140],"facilitate":[142],"compression":[144],"both":[146],"model":[147],"weights":[148],"activations.":[150],"We":[151],"evaluate":[152],"this":[153],"work":[154],"across":[155],"10":[156],"different":[157],"transformer-based":[158],"benchmarks,":[159],"results":[162],"demonstrate":[163],"that":[164],"integer-based":[167],"achieves":[170],"average":[172],"2.70x":[173],"gain":[175],"3.57":[177],"x":[178],"savings":[180],"maintaining":[182],"arithmetic.":[189]},"counts_by_year":[],"updated_date":"2025-11-06T03:46:38.306776","created_date":"2025-10-10T00:00:00"}