{"id":"https://openalex.org/W7164188446","doi":"https://doi.org/10.1109/fccm68464.2026.00027","title":"LUT-LLM: Efficient Language Model Inference with Memory-based Computations on FPGAs","display_name":"LUT-LLM: Efficient Language Model Inference with Memory-based Computations on FPGAs","publication_year":2026,"publication_date":"2026-05-13","ids":{"openalex":"https://openalex.org/W7164188446","doi":"https://doi.org/10.1109/fccm68464.2026.00027"},"language":null,"primary_location":{"id":"doi:10.1109/fccm68464.2026.00027","is_oa":false,"landing_page_url":"https://doi.org/10.1109/fccm68464.2026.00027","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2026 IEEE 34th Annual International Symposium on Field-Programmable Custom Computing Machines (FCCM)","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/A5108929023","display_name":"Zifan He","orcid":null},"institutions":[{"id":"https://openalex.org/I161318765","display_name":"University of California, Los Angeles","ror":"https://ror.org/046rm7j60","country_code":"US","type":"education","lineage":["https://openalex.org/I161318765"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Zifan He","raw_affiliation_strings":["University of California,Los Angeles"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of California,Los Angeles","institution_ids":["https://openalex.org/I161318765"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5010141774","display_name":"Shengyu Ye","orcid":null},"institutions":[{"id":"https://openalex.org/I4210164937","display_name":"Microsoft Research (United Kingdom)","ror":"https://ror.org/05k87vq12","country_code":"GB","type":"company","lineage":["https://openalex.org/I1290206253","https://openalex.org/I4210164937"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Shengyu Ye","raw_affiliation_strings":["Microsoft Research"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Microsoft Research","institution_ids":["https://openalex.org/I4210164937"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5138322906","display_name":"Rui Ma","orcid":null},"institutions":[{"id":"https://openalex.org/I4210164937","display_name":"Microsoft Research (United Kingdom)","ror":"https://ror.org/05k87vq12","country_code":"GB","type":"company","lineage":["https://openalex.org/I1290206253","https://openalex.org/I4210164937"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Rui Ma","raw_affiliation_strings":["Microsoft Research"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Microsoft Research","institution_ids":["https://openalex.org/I4210164937"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5138292985","display_name":"Yang Wang","orcid":null},"institutions":[{"id":"https://openalex.org/I4210164937","display_name":"Microsoft Research (United Kingdom)","ror":"https://ror.org/05k87vq12","country_code":"GB","type":"company","lineage":["https://openalex.org/I1290206253","https://openalex.org/I4210164937"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Yang Wang","raw_affiliation_strings":["Microsoft Research"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Microsoft Research","institution_ids":["https://openalex.org/I4210164937"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5138334489","display_name":"Jason Cong","orcid":null},"institutions":[{"id":"https://openalex.org/I161318765","display_name":"University of California, Los Angeles","ror":"https://ror.org/046rm7j60","country_code":"US","type":"education","lineage":["https://openalex.org/I161318765"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Jason Cong","raw_affiliation_strings":["University of California,Los Angeles"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of California,Los Angeles","institution_ids":["https://openalex.org/I161318765"]}]}],"institutions":[],"countries_distinct_count":2,"institutions_distinct_count":5,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":{"value":0.94126226,"is_in_top_1_percent":false,"is_in_top_10_percent":true},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"109","last_page":"118"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.44670000672340393,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.44670000672340393,"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/T10201","display_name":"Speech Recognition and Synthesis","score":0.05310000106692314,"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/T10028","display_name":"Topic Modeling","score":0.05260000005364418,"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/computation","display_name":"Computation","score":0.5169000029563904},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.4603999853134155},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.36149999499320984},{"id":"https://openalex.org/keywords/field-programmable-gate-array","display_name":"Field-programmable gate array","score":0.3569999933242798},{"id":"https://openalex.org/keywords/model-of-computation","display_name":"Model of computation","score":0.3059000074863434}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7002000212669373},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.5169000029563904},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.4603999853134155},{"id":"https://openalex.org/C199360897","wikidata":"https://www.wikidata.org/wiki/Q9143","display_name":"Programming language","level":1,"score":0.42590001225471497},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.39879998564720154},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.3806999921798706},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.36149999499320984},{"id":"https://openalex.org/C42935608","wikidata":"https://www.wikidata.org/wiki/Q190411","display_name":"Field-programmable gate array","level":2,"score":0.3569999933242798},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3409000039100647},{"id":"https://openalex.org/C184596265","wikidata":"https://www.wikidata.org/wiki/Q2651576","display_name":"Model of computation","level":3,"score":0.3059000074863434},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.3009999990463257}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1109/fccm68464.2026.00027","is_oa":false,"landing_page_url":"https://doi.org/10.1109/fccm68464.2026.00027","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2026 IEEE 34th Annual International Symposium on Field-Programmable Custom Computing Machines (FCCM)","raw_type":"proceedings-article"}],"best_oa_location":null,"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":39,"referenced_works":["https://openalex.org/W2002555321","https://openalex.org/W2036816056","https://openalex.org/W2759259332","https://openalex.org/W2923014074","https://openalex.org/W2963122961","https://openalex.org/W2963748441","https://openalex.org/W3133395503","https://openalex.org/W3190062760","https://openalex.org/W4211022608","https://openalex.org/W4244017338","https://openalex.org/W4308083513","https://openalex.org/W4384886635","https://openalex.org/W4385245566","https://openalex.org/W4386830976","https://openalex.org/W4387212331","https://openalex.org/W4387321091","https://openalex.org/W4389518760","https://openalex.org/W4391125526","https://openalex.org/W4392972248","https://openalex.org/W4393578753","https://openalex.org/W4393949386","https://openalex.org/W4394661856","https://openalex.org/W4394931310","https://openalex.org/W4401414043","https://openalex.org/W4402167889","https://openalex.org/W4402671013","https://openalex.org/W4404782943","https://openalex.org/W4407953477","https://openalex.org/W4408029577","https://openalex.org/W4408891426","https://openalex.org/W4409248495","https://openalex.org/W4409506669","https://openalex.org/W4410810291","https://openalex.org/W4410810423","https://openalex.org/W4411119596","https://openalex.org/W4412965816","https://openalex.org/W4415308626","https://openalex.org/W4415796274","https://openalex.org/W4415797506"],"related_works":[],"abstract_inverted_index":{"The":[0],"rapid":[1],"development":[2],"of":[3,64],"large":[4],"language":[5,111,131],"models":[6,112,204],"(LLM)":[7],"has":[8],"greatly":[9],"enhanced":[10],"everyday":[11],"applications.":[12],"While":[13],"many":[14],"FPGA-based":[15],"accelerators,":[16],"with":[17,98,133,209],"flexibility":[18],"for":[19,190,215],"fine-grained":[20],"data":[21,188],"control,":[22],"exhibit":[23],"superior":[24],"speed":[25,236],"and":[26,109,116,148,180,212,230,237],"energy":[27,241],"efficiency":[28,93,242],"compared":[29],"to":[30,41,49,84,96,107,114,171,186,205],"GPUs,":[31],"recent":[32],"GPU-specific":[33],"optimizations":[34],"have":[35],"diminished":[36],"this":[37,57,161],"advantage.":[38],"When":[39],"limited":[40],"arithmetic-based":[42,83],"computation,":[43,135],"FPGAs":[44,65,95],"often":[45],"underperform":[46],"GPUs":[47],"due":[48,113],"their":[50],"comparatively":[51],"fewer":[52],"computational":[53,74],"resources.":[54],"To":[55,159],"address":[56],"challenge,":[58],"we":[59],"exploit":[60],"a":[61,141,166,182,191,198,232,238],"key":[62],"advantage":[63],"over":[66],"GPUs:":[67],"abundant":[68],"distributed":[69],"on-chip":[70],"memory":[71],"embedded":[72],"among":[73],"units.":[75],"We":[76,139,196],"believe":[77],"that":[78,128,201],"shifting":[79],"LLM":[80],"inference":[81],"from":[82],"memory-based":[85,134],"computations":[86],"through":[87],"table":[88,178,194,207],"lookups":[89,208],"can":[90],"improve":[91],"the":[92,124,150,155,221],"on":[94,220],"compete":[97],"GPUs.":[99,244],"However,":[100],"existing":[101,203],"methods":[102],"are":[103],"inefficient":[104],"or":[105],"unable":[106],"scale":[108],"deploy":[110],"algorithm":[115],"architecture":[117],"design":[118,185],"limitations.":[119],"This":[120],"paper":[121],"introduces":[122],"LUT-LLM1,":[123],"first":[125],"FPGA":[126],"accelerator":[127],"deploys":[129],"1B+":[130],"model":[132,219],"leveraging":[136],"vector":[137,152],"quantization.":[138],"construct":[140],"performance":[142],"model,":[143],"evaluate":[144],"multiple":[145],"quantization":[146],"schemes,":[147],"identify":[149],"activation-weight":[151],"co-quantization":[153],"as":[154],"most":[156],"effective":[157],"approach.":[158],"support":[160,206],"scheme,":[162],"LUT-LLM":[163,214],"features":[164],"(1)":[165],"bandwidth-aware":[167],"parallel":[168],"centroid":[169],"search":[170],"reduce":[172,187],"decoding":[173],"latency,":[174],"(2)":[175],"efficient":[176],"2D":[177],"lookups,":[179],"(3)":[181],"spatial-temporal":[183],"hybrid":[184],"caching":[189],"higher":[192,240],"throughput":[193],"lookup.":[195],"develop":[197],"training":[199],"recipe":[200],"converts":[202],"high":[210],"accuracy":[211],"prototype":[213],"Qwen":[216],"3":[217],"1.7B":[218],"AMD":[222],"V80":[223],"FPGA,":[224],"reducing":[225],"arithmetic":[226],"operations":[227],"by":[228],"4\u00d7":[229],"achieving":[231],"1.10\u223c3.29\u00d7":[233],"faster":[234],"generation":[235],"3.05\u223c6.60\u00d7":[239],"than":[243]},"counts_by_year":[],"updated_date":"2026-06-12T06:20:11.936012","created_date":"2026-06-11T00:00:00"}