{"id":"https://openalex.org/W4393578447","doi":"https://doi.org/10.1145/3626202.3637600","title":"A Comprehensive Evaluation of FPGA-Based Spatial Acceleration of LLMs","display_name":"A Comprehensive Evaluation of FPGA-Based Spatial Acceleration of LLMs","publication_year":2024,"publication_date":"2024-04-01","ids":{"openalex":"https://openalex.org/W4393578447","doi":"https://doi.org/10.1145/3626202.3637600"},"language":"en","primary_location":{"id":"doi:10.1145/3626202.3637600","is_oa":false,"landing_page_url":"https://doi.org/10.1145/3626202.3637600","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 2024 ACM/SIGDA International Symposium on Field Programmable Gate Arrays","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/A5008694159","display_name":"Hongzheng Chen","orcid":"https://orcid.org/0000-0002-6617-0075"},"institutions":[{"id":"https://openalex.org/I205783295","display_name":"Cornell University","ror":"https://ror.org/05bnh6r87","country_code":"US","type":"education","lineage":["https://openalex.org/I205783295"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Hongzheng Chen","raw_affiliation_strings":["Cornell University, Ithaca, USA"],"raw_orcid":"https://orcid.org/0000-0002-6617-0075","affiliations":[{"raw_affiliation_string":"Cornell University, Ithaca, USA","institution_ids":["https://openalex.org/I205783295"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100445508","display_name":"Jiahao Zhang","orcid":"https://orcid.org/0009-0000-8379-7489"},"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":"Jiahao Zhang","raw_affiliation_strings":["Tsinghua University, Beijing, China"],"raw_orcid":"https://orcid.org/0009-0000-8379-7489","affiliations":[{"raw_affiliation_string":"Tsinghua University, Beijing, China","institution_ids":["https://openalex.org/I99065089"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5039033697","display_name":"Yixiao Du","orcid":"https://orcid.org/0000-0002-6106-1283"},"institutions":[{"id":"https://openalex.org/I205783295","display_name":"Cornell University","ror":"https://ror.org/05bnh6r87","country_code":"US","type":"education","lineage":["https://openalex.org/I205783295"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Yixiao Du","raw_affiliation_strings":["Cornell University, Ithaca, USA"],"raw_orcid":"https://orcid.org/0000-0002-6106-1283","affiliations":[{"raw_affiliation_string":"Cornell University, Ithaca, USA","institution_ids":["https://openalex.org/I205783295"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5032824854","display_name":"Shaojie Xiang","orcid":"https://orcid.org/0000-0002-6901-8837"},"institutions":[{"id":"https://openalex.org/I205783295","display_name":"Cornell University","ror":"https://ror.org/05bnh6r87","country_code":"US","type":"education","lineage":["https://openalex.org/I205783295"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Shaojie Xiang","raw_affiliation_strings":["Cornell University, Ithaca, USA"],"raw_orcid":"https://orcid.org/0000-0002-6901-8837","affiliations":[{"raw_affiliation_string":"Cornell University, Ithaca, USA","institution_ids":["https://openalex.org/I205783295"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5029785504","display_name":"Zichao Yue","orcid":"https://orcid.org/0009-0003-8585-5947"},"institutions":[{"id":"https://openalex.org/I205783295","display_name":"Cornell University","ror":"https://ror.org/05bnh6r87","country_code":"US","type":"education","lineage":["https://openalex.org/I205783295"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Zichao Yue","raw_affiliation_strings":["Cornell University, Ithaca, USA"],"raw_orcid":"https://orcid.org/0009-0003-8585-5947","affiliations":[{"raw_affiliation_string":"Cornell University, Ithaca, USA","institution_ids":["https://openalex.org/I205783295"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5059293268","display_name":"Niansong Zhang","orcid":"https://orcid.org/0000-0002-2850-0176"},"institutions":[{"id":"https://openalex.org/I205783295","display_name":"Cornell University","ror":"https://ror.org/05bnh6r87","country_code":"US","type":"education","lineage":["https://openalex.org/I205783295"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Niansong Zhang","raw_affiliation_strings":["Cornell University, Ithaca, USA"],"raw_orcid":"https://orcid.org/0000-0002-2850-0176","affiliations":[{"raw_affiliation_string":"Cornell University, Ithaca, USA","institution_ids":["https://openalex.org/I205783295"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5050643298","display_name":"Yaohui Cai","orcid":"https://orcid.org/0000-0003-3785-3413"},"institutions":[{"id":"https://openalex.org/I205783295","display_name":"Cornell University","ror":"https://ror.org/05bnh6r87","country_code":"US","type":"education","lineage":["https://openalex.org/I205783295"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Yaohui Cai","raw_affiliation_strings":["Cornell University, Ithaca, USA"],"raw_orcid":"https://orcid.org/0000-0003-3785-3413","affiliations":[{"raw_affiliation_string":"Cornell University, Ithaca, USA","institution_ids":["https://openalex.org/I205783295"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5037210004","display_name":"Zhiru Zhang","orcid":"https://orcid.org/0000-0002-0778-0308"},"institutions":[{"id":"https://openalex.org/I205783295","display_name":"Cornell University","ror":"https://ror.org/05bnh6r87","country_code":"US","type":"education","lineage":["https://openalex.org/I205783295"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Zhiru Zhang","raw_affiliation_strings":["Cornell University, Ithaca, USA"],"raw_orcid":"https://orcid.org/0000-0002-0778-0308","affiliations":[{"raw_affiliation_string":"Cornell University, Ithaca, USA","institution_ids":["https://openalex.org/I205783295"]}]}],"institutions":[],"countries_distinct_count":2,"institutions_distinct_count":8,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":1.7478,"has_fulltext":false,"cited_by_count":6,"citation_normalized_percentile":{"value":0.83736927,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":{"min":91,"max":98},"biblio":{"volume":null,"issue":null,"first_page":"185","last_page":"185"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T12810","display_name":"Real-time simulation and control systems","score":0.9164999723434448,"subfield":{"id":"https://openalex.org/subfields/2207","display_name":"Control and Systems 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/T12810","display_name":"Real-time simulation and control systems","score":0.9164999723434448,"subfield":{"id":"https://openalex.org/subfields/2207","display_name":"Control and Systems 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/T10904","display_name":"Embedded Systems Design Techniques","score":0.9158999919891357,"subfield":{"id":"https://openalex.org/subfields/1708","display_name":"Hardware and Architecture"},"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/T11522","display_name":"VLSI and FPGA Design Techniques","score":0.9054999947547913,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/acceleration","display_name":"Acceleration","score":0.7721245288848877},{"id":"https://openalex.org/keywords/field-programmable-gate-array","display_name":"Field-programmable gate array","score":0.7142110466957092},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.5799899101257324},{"id":"https://openalex.org/keywords/embedded-system","display_name":"Embedded system","score":0.25911813974380493},{"id":"https://openalex.org/keywords/physics","display_name":"Physics","score":0.08026444911956787}],"concepts":[{"id":"https://openalex.org/C117896860","wikidata":"https://www.wikidata.org/wiki/Q11376","display_name":"Acceleration","level":2,"score":0.7721245288848877},{"id":"https://openalex.org/C42935608","wikidata":"https://www.wikidata.org/wiki/Q190411","display_name":"Field-programmable gate array","level":2,"score":0.7142110466957092},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5799899101257324},{"id":"https://openalex.org/C149635348","wikidata":"https://www.wikidata.org/wiki/Q193040","display_name":"Embedded system","level":1,"score":0.25911813974380493},{"id":"https://openalex.org/C121332964","wikidata":"https://www.wikidata.org/wiki/Q413","display_name":"Physics","level":0,"score":0.08026444911956787},{"id":"https://openalex.org/C74650414","wikidata":"https://www.wikidata.org/wiki/Q11397","display_name":"Classical mechanics","level":1,"score":0.0}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1145/3626202.3637600","is_oa":false,"landing_page_url":"https://doi.org/10.1145/3626202.3637600","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 2024 ACM/SIGDA International Symposium on Field Programmable Gate Arrays","raw_type":"proceedings-article"}],"best_oa_location":null,"sustainable_development_goals":[{"score":0.8899999856948853,"display_name":"Affordable and clean energy","id":"https://metadata.un.org/sdg/7"}],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W2748952813","https://openalex.org/W2111241003","https://openalex.org/W2390279801","https://openalex.org/W4200391368","https://openalex.org/W2358668433","https://openalex.org/W2355315220","https://openalex.org/W2210979487","https://openalex.org/W2376932109","https://openalex.org/W2074043759","https://openalex.org/W2001405890"],"abstract_inverted_index":{"Recent":[0],"advancements":[1],"in":[2,14,40,123,219,232,241],"large":[3],"language":[4],"models":[5],"(LLMs)":[6],"have":[7,172],"generated":[8],"significant":[9],"demands":[10],"for":[11,28,61,75,98,200],"efficient":[12],"deployment":[13],"inference":[15,63],"workloads.":[16],"Most":[17],"existing":[18],"approaches":[19],"rely":[20],"on":[21,64,112,142,177],"temporal":[22],"architectures":[23],"that":[24,154],"reuse":[25],"hardware":[26,73,110],"units":[27,74],"different":[29],"network":[30],"layers":[31],"and":[32,55,157,168,175,228],"operators.":[33],"However,":[34],"these":[35],"methods":[36],"often":[37],"encounter":[38],"challenges":[39],"achieving":[41,224],"low":[42],"latency":[43],"due":[44],"to":[45,191,196,214,236],"considerable":[46],"memory":[47,90],"access":[48],"overhead.":[49],"This":[50],"paper":[51],"investigates":[52],"the":[53,69,100,108,121,161,201,220,237,242],"feasibility":[54],"potential":[56],"of":[57,71,102,138,149,163],"model-specific":[58],"spatial":[59,104,126],"acceleration":[60,127],"LLM":[62,105,140],"FPGAs.":[65],"Our":[66],"approach":[67,187],"involves":[68],"specialization":[70],"distinct":[72],"specific":[76],"operators,":[77],"facilitating":[78],"direct":[79],"communication":[80],"between":[81],"them":[82],"through":[83],"a":[84,94,103,147,210,225,229],"dataflow":[85],"architecture":[86],"while":[87,223],"minimizing":[88],"off-chip":[89],"accesses.":[91],"We":[92],"introduce":[93],"comprehensive":[95],"analytical":[96,166],"model":[97,141,167],"estimating":[99],"performance":[101],"accelerator,":[106],"considering":[107],"available":[109],"resources":[111],"an":[113,139,178,216],"FPGA.":[114,182],"Through":[115],"our":[116,165,186],"analysis,":[117],"we":[118,144,171,208],"can":[119,128,188],"determine":[120],"scenarios":[122],"which":[124],"FPGA-based":[125,198],"outperform":[129],"its":[130],"GPU-based":[131],"counterpart.":[132],"To":[133,159],"enable":[134],"more":[135],"productive":[136],"implementations":[137],"FPGAs,":[143],"further":[145],"provide":[146],"library":[148],"high-level":[150],"synthesis":[151],"(HLS)":[152],"kernels":[153],"are":[155],"composable":[156],"reusable.":[158],"validate":[160],"effectiveness":[162],"both":[164],"HLS":[169],"library,":[170],"implemented":[173],"BERT":[174,202],"GPT2":[176],"AMD":[179],"Alveo":[180],"U280":[181],"Experimental":[183],"results":[184],"demonstrate":[185],"achieve":[189],"up":[190],"16.1x":[192],"speedup":[193,212,227],"when":[194],"compared":[195,213,235],"previous":[197],"accelerators":[199],"model.":[203],"For":[204],"GPT":[205],"generative":[206],"inference,":[207],"attain":[209],"2.2x":[211],"DFX,":[215],"FPGA":[217],"overlay,":[218],"prefill":[221],"stage,":[222],"1.9x":[226],"5.7x":[230],"improvement":[231],"energy":[233],"efficiency":[234],"NVIDIA":[238],"A100":[239],"GPU":[240],"decode":[243],"stage.":[244]},"counts_by_year":[{"year":2025,"cited_by_count":1},{"year":2024,"cited_by_count":5}],"updated_date":"2026-06-11T09:08:48.828518","created_date":"2025-10-10T00:00:00"}
