{"id":"https://openalex.org/W4308083513","doi":"https://doi.org/10.1109/micro56248.2022.00051","title":"DFX: A Low-latency Multi-FPGA Appliance for Accelerating Transformer-based Text Generation","display_name":"DFX: A Low-latency Multi-FPGA Appliance for Accelerating Transformer-based Text Generation","publication_year":2022,"publication_date":"2022-10-01","ids":{"openalex":"https://openalex.org/W4308083513","doi":"https://doi.org/10.1109/micro56248.2022.00051"},"language":"en","primary_location":{"id":"doi:10.1109/micro56248.2022.00051","doi":"https://doi.org/10.1109/micro56248.2022.00051","is_oa":false,"landing_page_url":"https://doi.org/10.1109/micro56248.2022.00051","pdf_url":null,"source":{"id":"https://openalex.org/S4210171338","display_name":"IEEE/ACM International Symposium on Microarchitecture (MICRO)","issn_l":"1072-4451","issn":["1072-4451","2379-3155"],"is_oa":false,"is_in_doaj":false,"is_core":true,"listed_in":["cwts-core","norway-1"],"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":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2022 55th IEEE/ACM International Symposium on Microarchitecture (MICRO)","raw_type":"proceedings-article"},"volume":null,"type":"conference-paper","is_traditional":true,"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/A5075135160","display_name":"Seongmin Hong","orcid":"https://orcid.org/0000-0001-6904-8094","observed_orcids":["https://orcid.org/0000-0001-6904-8094"]},"institutions":[{"id":"https://openalex.org/I157485424","display_name":"Korea Advanced Institute of Science and Technology","ror":"https://ror.org/05apxxy63","country_code":"KR","type":"education","lineage":["https://openalex.org/I157485424"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Seongmin Hong","raw_affiliation_strings":["KAIST, Daejeon, South Korea"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"KAIST, Daejeon, South Korea","institution_ids":["https://openalex.org/I157485424"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5030449068","display_name":"Seungjae Moon","orcid":"https://orcid.org/0009-0002-5924-7000","observed_orcids":["https://orcid.org/0009-0002-5924-7000"]},"institutions":[{"id":"https://openalex.org/I157485424","display_name":"Korea Advanced Institute of Science and Technology","ror":"https://ror.org/05apxxy63","country_code":"KR","type":"education","lineage":["https://openalex.org/I157485424"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Seungjae Moon","raw_affiliation_strings":["KAIST, Daejeon, South Korea"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"KAIST, Daejeon, South Korea","institution_ids":["https://openalex.org/I157485424"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100636164","display_name":"Junsoo Kim","orcid":"https://orcid.org/0000-0001-6680-2602","observed_orcids":["https://orcid.org/0000-0001-6680-2602"]},"institutions":[{"id":"https://openalex.org/I157485424","display_name":"Korea Advanced Institute of Science and Technology","ror":"https://ror.org/05apxxy63","country_code":"KR","type":"education","lineage":["https://openalex.org/I157485424"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Junsoo Kim","raw_affiliation_strings":["KAIST, Daejeon, South Korea"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"KAIST, Daejeon, South Korea","institution_ids":["https://openalex.org/I157485424"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5041595994","display_name":"Sungjae Lee","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I4405260609","display_name":"NAVER Cloud (South Korea)","ror":"https://ror.org/041051y78","country_code":"KR","type":"company","lineage":["https://openalex.org/I4405260609","https://openalex.org/I60922564"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Sungjae Lee","raw_affiliation_strings":["NAVER CLOVA, Seongnam, South Korea"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"NAVER CLOVA, Seongnam, South Korea","institution_ids":["https://openalex.org/I4405260609"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5103139088","display_name":"Minsub Kim","orcid":"https://orcid.org/0000-0002-1675-1314","observed_orcids":["https://orcid.org/0000-0002-1675-1314"]},"institutions":[{"id":"https://openalex.org/I4405260609","display_name":"NAVER Cloud (South Korea)","ror":"https://ror.org/041051y78","country_code":"KR","type":"company","lineage":["https://openalex.org/I4405260609","https://openalex.org/I60922564"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Minsub Kim","raw_affiliation_strings":["NAVER CLOVA, Seongnam, South Korea"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"NAVER CLOVA, Seongnam, South Korea","institution_ids":["https://openalex.org/I4405260609"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101652725","display_name":"Dongsoo Lee","orcid":"https://orcid.org/0000-0002-4155-6940","observed_orcids":["https://orcid.org/0000-0002-4155-6940"]},"institutions":[{"id":"https://openalex.org/I4405260609","display_name":"NAVER Cloud (South Korea)","ror":"https://ror.org/041051y78","country_code":"KR","type":"company","lineage":["https://openalex.org/I4405260609","https://openalex.org/I60922564"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Dongsoo Lee","raw_affiliation_strings":["NAVER CLOVA, Seongnam, South Korea"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"NAVER CLOVA, Seongnam, South Korea","institution_ids":["https://openalex.org/I4405260609"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5100447377","display_name":"Joo-Young Kim","orcid":"https://orcid.org/0000-0003-1099-1496","observed_orcids":["https://orcid.org/0000-0003-1099-1496"]},"institutions":[{"id":"https://openalex.org/I157485424","display_name":"Korea Advanced Institute of Science and Technology","ror":"https://ror.org/05apxxy63","country_code":"KR","type":"education","lineage":["https://openalex.org/I157485424"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Joo-Young Kim","raw_affiliation_strings":["KAIST, Daejeon, South Korea"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"KAIST, Daejeon, South Korea","institution_ids":["https://openalex.org/I157485424"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":16.4491,"has_fulltext":false,"cited_by_count":116,"citation_normalized_percentile":{"value":0.99841521,"is_in_top_1_percent":true,"is_in_top_10_percent":true},"cited_by_percentile_year":{"min":88,"max":100},"biblio":{"volume":null,"issue":null,"first_page":"616","last_page":"630"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10904","display_name":"Embedded Systems Design Techniques","score":0.9115999937057495,"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"}},"topics":[{"id":"https://openalex.org/T10904","display_name":"Embedded Systems Design Techniques","score":0.9115999937057495,"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/T10036","display_name":"Advanced Neural Network Applications","score":0.03539999946951866,"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"}},{"id":"https://openalex.org/T11522","display_name":"VLSI and FPGA Design Techniques","score":0.008299999870359898,"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/text-generation","display_name":"text generation","score":0.824999988079071},{"id":"https://openalex.org/keywords/natural-language-generation","display_name":"natural language generation","score":0.925000011920929},{"id":"https://openalex.org/keywords/model-parallelism","display_name":"model parallelism","score":0.8939999938011169},{"id":"https://openalex.org/keywords/custom-instructions","display_name":"custom instructions","score":0.675000011920929},{"id":"https://openalex.org/keywords/high-bandwidth-memory","display_name":"high-bandwidth memory","score":0.7200000286102295},{"id":"https://openalex.org/keywords/gpu-comparison","display_name":"GPU comparison","score":0.5550000071525574},{"id":"https://openalex.org/keywords/energy-efficiency","display_name":"energy efficiency","score":0.7839999794960022}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8417438268661499},{"id":"https://openalex.org/C42935608","wikidata":"https://www.wikidata.org/wiki/Q190411","display_name":"Field-programmable gate array","level":2,"score":0.6526927351951599},{"id":"https://openalex.org/C170858558","wikidata":"https://www.wikidata.org/wiki/Q1394144","display_name":"Automatic summarization","level":2,"score":0.5630018711090088},{"id":"https://openalex.org/C133162039","wikidata":"https://www.wikidata.org/wiki/Q1061077","display_name":"Code generation","level":3,"score":0.5177847146987915},{"id":"https://openalex.org/C2776221188","wikidata":"https://www.wikidata.org/wiki/Q21072556","display_name":"Design space exploration","level":2,"score":0.5073768496513367},{"id":"https://openalex.org/C82876162","wikidata":"https://www.wikidata.org/wiki/Q17096504","display_name":"Latency (audio)","level":2,"score":0.4875895380973816},{"id":"https://openalex.org/C68339613","wikidata":"https://www.wikidata.org/wiki/Q1549489","display_name":"Speedup","level":2,"score":0.47702786326408386},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.4666787385940552},{"id":"https://openalex.org/C2780513914","wikidata":"https://www.wikidata.org/wiki/Q18210350","display_name":"Bottleneck","level":2,"score":0.4458944797515869},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.42676153779029846},{"id":"https://openalex.org/C9390403","wikidata":"https://www.wikidata.org/wiki/Q3966","display_name":"Computer hardware","level":1,"score":0.4243490993976593},{"id":"https://openalex.org/C118524514","wikidata":"https://www.wikidata.org/wiki/Q173212","display_name":"Computer architecture","level":1,"score":0.4077393710613251},{"id":"https://openalex.org/C149635348","wikidata":"https://www.wikidata.org/wiki/Q193040","display_name":"Embedded system","level":1,"score":0.3805030882358551},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.17901095747947693},{"id":"https://openalex.org/C111919701","wikidata":"https://www.wikidata.org/wiki/Q9135","display_name":"Operating system","level":1,"score":0.102885901927948},{"id":"https://openalex.org/C76155785","wikidata":"https://www.wikidata.org/wiki/Q418","display_name":"Telecommunications","level":1,"score":0.0},{"id":"https://openalex.org/C62520636","wikidata":"https://www.wikidata.org/wiki/Q944","display_name":"Quantum mechanics","level":1,"score":0.0},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.0},{"id":"https://openalex.org/C165801399","wikidata":"https://www.wikidata.org/wiki/Q25428","display_name":"Voltage","level":2,"score":0.0},{"id":"https://openalex.org/C121332964","wikidata":"https://www.wikidata.org/wiki/Q413","display_name":"Physics","level":0,"score":0.0}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1109/micro56248.2022.00051","doi":"https://doi.org/10.1109/micro56248.2022.00051","is_oa":false,"landing_page_url":"https://doi.org/10.1109/micro56248.2022.00051","pdf_url":null,"source":{"id":"https://openalex.org/S4210171338","display_name":"IEEE/ACM International Symposium on Microarchitecture (MICRO)","issn_l":"1072-4451","issn":["1072-4451","2379-3155"],"is_oa":false,"is_in_doaj":false,"is_core":true,"listed_in":["cwts-core","norway-1"],"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":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2022 55th IEEE/ACM International Symposium on Microarchitecture (MICRO)","raw_type":"proceedings-article"}],"best_oa_location":null,"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[{"id":"https://openalex.org/F4320316108","display_name":"NAV","ror":"https://ror.org/00qghd717"}],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":42,"referenced_works":["https://openalex.org/W1599016936","https://openalex.org/W2019013398","https://openalex.org/W2064675550","https://openalex.org/W2082375193","https://openalex.org/W2157331557","https://openalex.org/W2194775991","https://openalex.org/W2525778437","https://openalex.org/W2528904340","https://openalex.org/W2606722458","https://openalex.org/W2626778328","https://openalex.org/W2807925339","https://openalex.org/W2883929540","https://openalex.org/W2896457183","https://openalex.org/W2899663614","https://openalex.org/W2900096133","https://openalex.org/W2901299405","https://openalex.org/W2914204778","https://openalex.org/W2926767350","https://openalex.org/W2952889708","https://openalex.org/W2956901422","https://openalex.org/W2963341956","https://openalex.org/W2963542740","https://openalex.org/W2973727699","https://openalex.org/W3007303655","https://openalex.org/W3008905965","https://openalex.org/W3014153848","https://openalex.org/W3022810465","https://openalex.org/W3030030185","https://openalex.org/W3030163527","https://openalex.org/W3036878841","https://openalex.org/W3037932933","https://openalex.org/W3038115231","https://openalex.org/W3039050620","https://openalex.org/W3099872554","https://openalex.org/W3103995860","https://openalex.org/W3159727696","https://openalex.org/W3177865674","https://openalex.org/W3189877953","https://openalex.org/W3190062760","https://openalex.org/W3203324590","https://openalex.org/W3206453033","https://openalex.org/W3207622241"],"related_works":["https://openalex.org/W4403279065","https://openalex.org/W4403278343","https://openalex.org/W4411019820","https://openalex.org/W4411171608","https://openalex.org/W4417285552","https://openalex.org/W4400985603","https://openalex.org/W7124154102","https://openalex.org/W4311591782","https://openalex.org/W4385834084","https://openalex.org/W4396653497"],"abstract_inverted_index":{"Transformer":[0,22],"is":[1,100,151,226,238],"a":[2,41,56,60,121,239],"deep":[3],"learning":[4],"language":[5,11,33],"model":[6,128,145],"widely":[7],"used":[8],"for":[9,70,153,203,242],"natural":[10,32],"processing":[12,39,73],"(NLP)":[13],"services":[14],"in":[15,28,45,77,86,137,246],"datacenters.":[16,248],"Among":[17],"transformer":[18],"models,":[19],"Generative":[20],"Pretrained":[21],"(GPT)":[23],"has":[24],"achieved":[25],"remarkable":[26],"performance":[27,83],"text":[29,113,243],"generation,":[30],"or":[31],"generation":[34,52,88,141,244],"(NLG),":[35],"which":[36],"needs":[37],"the":[38,46,51,71,78,87,104,109,174,188,191,197,221,232],"of":[40,74,112,187,190,200],"large":[42,75],"input":[43],"context":[44],"summarization":[47,79,139],"stage,":[48,80],"followed":[49],"by":[50,108],"stage":[53,89],"that":[54,125,150,236],"produces":[55],"single":[57],"word":[58],"at":[59],"time.":[61],"The":[62],"conventional":[63],"platforms":[64],"such":[65],"as":[66],"GPU":[67,233],"are":[68],"specialized":[69],"parallel":[72],"inputs":[76],"but":[81],"their":[82],"significantly":[84],"degrades":[85],"due":[90],"to":[91,102],"its":[92],"sequential":[93,110],"characteristic.":[94],"Therefore,":[95],"an":[96],"efficient":[97],"hardware":[98,176,205],"platform":[99],"required":[101],"address":[103],"high":[105,135,192,204],"latency":[106,133],"caused":[107],"characteristic":[111],"generation.":[114],"In":[115],"this":[116],"paper,":[117],"we":[118],"present":[119],"DFX,":[120],"multi-FPGA":[122],"acceleration":[123],"appliance":[124],"executes":[126],"GPT-2":[127,169,223],"inference":[129],"end-to-end":[130],"with":[131],"low":[132],"and":[134,140,147,167,184,196,211],"throughput":[136],"both":[138],"stages.":[142],"DFX":[143,207,225],"uses":[144],"parallelism":[146],"optimized":[148],"dataflow":[149],"model-and-hardware-aware":[152],"fast":[154],"simultaneous":[155],"workload":[156],"execution":[157],"among":[158],"devices.":[159],"Its":[160],"compute":[161,201],"cores":[162],"operate":[163],"on":[164,178,220],"custom":[165],"instructions":[166],"provide":[168],"operations":[170],"end-to-end.":[171],"We":[172],"implement":[173],"proposed":[175],"architecture":[177],"four":[179,216],"Xilinx":[180],"Alveo":[181],"U280":[182],"FPGAs":[183],"utilize":[185],"all":[186],"channels":[189],"bandwidth":[193],"memory":[194],"(HBM)":[195],"maximum":[198],"number":[199],"resources":[202],"efficiency.":[206],"achieves":[208],"5.58$\\times$":[209],"speedup":[210],"3.99$\\times$":[212],"energy":[213],"efficiency":[214],"over":[215],"NVIDIA":[217],"V100":[218],"GPUs":[219],"modern":[222],"model.":[224],"also":[227],"8.21$\\times$":[228],"more":[229],"cost-effective":[230],"than":[231],"appliance,":[234],"suggesting":[235],"it":[237],"promising":[240],"solution":[241],"workloads":[245],"cloud":[247]},"counts_by_year":[{"year":2026,"cited_by_count":21},{"year":2025,"cited_by_count":55},{"year":2024,"cited_by_count":34},{"year":2023,"cited_by_count":5},{"year":2022,"cited_by_count":1}],"updated_date":"2026-11-10T11:39:43.021295","created_date":"2022-11-08T00:00:00"}