{"id":"https://openalex.org/W7127922577","doi":"https://doi.org/10.1145/3748173.3779189","title":"Hummingbird+: Advancing FPGA-based LLM Deployment from Research Prototype to Edge Product","display_name":"Hummingbird+: Advancing FPGA-based LLM Deployment from Research Prototype to Edge Product","publication_year":2026,"publication_date":"2026-02-05","ids":{"openalex":"https://openalex.org/W7127922577","doi":"https://doi.org/10.1145/3748173.3779189"},"language":null,"primary_location":{"id":"doi:10.1145/3748173.3779189","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3748173.3779189","pdf_url":null,"source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 2026 ACM/SIGDA International Symposium on Field Programmable Gate Arrays","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://doi.org/10.1145/3748173.3779189","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5125140323","display_name":"Jindong Li","orcid":null},"institutions":[{"id":"https://openalex.org/I19820366","display_name":"Chinese Academy of Sciences","ror":"https://ror.org/034t30j35","country_code":"CN","type":"government","lineage":["https://openalex.org/I19820366"]},{"id":"https://openalex.org/I4210094879","display_name":"Shandong Institute of Automation","ror":"https://ror.org/00qdtba35","country_code":"CN","type":"facility","lineage":["https://openalex.org/I4210094879","https://openalex.org/I4210142748"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Jindong Li","raw_affiliation_strings":["Institute of Automation, Chinese Academy of Sciences, Beijing, China"],"raw_orcid":"https://orcid.org/0000-0002-4009-916X","affiliations":[{"raw_affiliation_string":"Institute of Automation, Chinese Academy of Sciences, Beijing, China","institution_ids":["https://openalex.org/I19820366","https://openalex.org/I4210094879"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5119309865","display_name":"Tenglong Li","orcid":"https://orcid.org/0009-0007-3266-2075"},"institutions":[{"id":"https://openalex.org/I19820366","display_name":"Chinese Academy of Sciences","ror":"https://ror.org/034t30j35","country_code":"CN","type":"government","lineage":["https://openalex.org/I19820366"]},{"id":"https://openalex.org/I4210094879","display_name":"Shandong Institute of Automation","ror":"https://ror.org/00qdtba35","country_code":"CN","type":"facility","lineage":["https://openalex.org/I4210094879","https://openalex.org/I4210142748"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Tenglong Li","raw_affiliation_strings":["Institute of Automation, Chinese Academy of Sciences, Beijing, China"],"raw_orcid":"https://orcid.org/0009-0007-3266-2075","affiliations":[{"raw_affiliation_string":"Institute of Automation, Chinese Academy of Sciences, Beijing, China","institution_ids":["https://openalex.org/I19820366","https://openalex.org/I4210094879"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5081220735","display_name":"Guobin Shen","orcid":"https://orcid.org/0000-0002-4069-2107"},"institutions":[{"id":"https://openalex.org/I19820366","display_name":"Chinese Academy of Sciences","ror":"https://ror.org/034t30j35","country_code":"CN","type":"government","lineage":["https://openalex.org/I19820366"]},{"id":"https://openalex.org/I4210094879","display_name":"Shandong Institute of Automation","ror":"https://ror.org/00qdtba35","country_code":"CN","type":"facility","lineage":["https://openalex.org/I4210094879","https://openalex.org/I4210142748"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Guobin Shen","raw_affiliation_strings":["Institute of Automation, Chinese Academy of Sciences, Beijing, China"],"raw_orcid":"https://orcid.org/0000-0002-4069-2107","affiliations":[{"raw_affiliation_string":"Institute of Automation, Chinese Academy of Sciences, Beijing, China","institution_ids":["https://openalex.org/I19820366","https://openalex.org/I4210094879"]}]},{"author_position":"middle","author":{"id":null,"display_name":"Dongcheng Zhao","orcid":"https://orcid.org/0000-0002-0593-8650"},"institutions":[{"id":"https://openalex.org/I19820366","display_name":"Chinese Academy of Sciences","ror":"https://ror.org/034t30j35","country_code":"CN","type":"government","lineage":["https://openalex.org/I19820366"]},{"id":"https://openalex.org/I4210094879","display_name":"Shandong Institute of Automation","ror":"https://ror.org/00qdtba35","country_code":"CN","type":"facility","lineage":["https://openalex.org/I4210094879","https://openalex.org/I4210142748"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Dongcheng Zhao","raw_affiliation_strings":["Institute of Automation, Chinese Academy of Sciences, Beijing, China"],"raw_orcid":"https://orcid.org/0000-0002-0593-8650","affiliations":[{"raw_affiliation_string":"Institute of Automation, Chinese Academy of Sciences, Beijing, China","institution_ids":["https://openalex.org/I19820366","https://openalex.org/I4210094879"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5125170433","display_name":"Qian Zhang","orcid":null},"institutions":[{"id":"https://openalex.org/I19820366","display_name":"Chinese Academy of Sciences","ror":"https://ror.org/034t30j35","country_code":"CN","type":"government","lineage":["https://openalex.org/I19820366"]},{"id":"https://openalex.org/I4210094879","display_name":"Shandong Institute of Automation","ror":"https://ror.org/00qdtba35","country_code":"CN","type":"facility","lineage":["https://openalex.org/I4210094879","https://openalex.org/I4210142748"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Qian Zhang","raw_affiliation_strings":["Institute of Automation, Chinese Academy of Sciences, Beijing, China"],"raw_orcid":"https://orcid.org/0000-0001-5314-4233","affiliations":[{"raw_affiliation_string":"Institute of Automation, Chinese Academy of Sciences, Beijing, China","institution_ids":["https://openalex.org/I19820366","https://openalex.org/I4210094879"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5125145361","display_name":"Yi Zeng","orcid":null},"institutions":[{"id":"https://openalex.org/I4210100255","display_name":"Beijing Academy of Artificial Intelligence","ror":"https://ror.org/016a74861","country_code":"CN","type":"other","lineage":["https://openalex.org/I4210100255"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Yi Zeng","raw_affiliation_strings":["Institute of Automation, Chinese Academy of Sciences, Beijing, China, Center for Long-term Artificial Intelligence, Beijing, China, and Institute of AI Safety and Governance, Beijing, China"],"raw_orcid":"https://orcid.org/0000-0002-9595-9091","affiliations":[{"raw_affiliation_string":"Institute of Automation, Chinese Academy of Sciences, Beijing, China, Center for Long-term Artificial Intelligence, Beijing, China, and Institute of AI Safety and Governance, Beijing, China","institution_ids":["https://openalex.org/I4210100255"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":3,"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":0.11330481,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"235","last_page":"246"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.40880000591278076,"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":0.40880000591278076,"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/T10904","display_name":"Embedded Systems Design Techniques","score":0.06859999895095825,"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/T10201","display_name":"Speech Recognition and Synthesis","score":0.06509999930858612,"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/field-programmable-gate-array","display_name":"Field-programmable gate array","score":0.7766000032424927},{"id":"https://openalex.org/keywords/software-deployment","display_name":"Software deployment","score":0.6865000128746033},{"id":"https://openalex.org/keywords/enhanced-data-rates-for-gsm-evolution","display_name":"Enhanced Data Rates for GSM Evolution","score":0.6195999979972839},{"id":"https://openalex.org/keywords/edge-device","display_name":"Edge device","score":0.5748999714851379},{"id":"https://openalex.org/keywords/resource","display_name":"Resource (disambiguation)","score":0.40880000591278076},{"id":"https://openalex.org/keywords/key","display_name":"Key (lock)","score":0.4068000018596649},{"id":"https://openalex.org/keywords/edge-computing","display_name":"Edge computing","score":0.37380000948905945},{"id":"https://openalex.org/keywords/product","display_name":"Product (mathematics)","score":0.3727000057697296}],"concepts":[{"id":"https://openalex.org/C42935608","wikidata":"https://www.wikidata.org/wiki/Q190411","display_name":"Field-programmable gate array","level":2,"score":0.7766000032424927},{"id":"https://openalex.org/C105339364","wikidata":"https://www.wikidata.org/wiki/Q2297740","display_name":"Software deployment","level":2,"score":0.6865000128746033},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6680999994277954},{"id":"https://openalex.org/C149635348","wikidata":"https://www.wikidata.org/wiki/Q193040","display_name":"Embedded system","level":1,"score":0.647599995136261},{"id":"https://openalex.org/C162307627","wikidata":"https://www.wikidata.org/wiki/Q204833","display_name":"Enhanced Data Rates for GSM Evolution","level":2,"score":0.6195999979972839},{"id":"https://openalex.org/C138236772","wikidata":"https://www.wikidata.org/wiki/Q25098575","display_name":"Edge device","level":3,"score":0.5748999714851379},{"id":"https://openalex.org/C9390403","wikidata":"https://www.wikidata.org/wiki/Q3966","display_name":"Computer hardware","level":1,"score":0.4447999894618988},{"id":"https://openalex.org/C118524514","wikidata":"https://www.wikidata.org/wiki/Q173212","display_name":"Computer architecture","level":1,"score":0.4099000096321106},{"id":"https://openalex.org/C206345919","wikidata":"https://www.wikidata.org/wiki/Q20380951","display_name":"Resource (disambiguation)","level":2,"score":0.40880000591278076},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.4068000018596649},{"id":"https://openalex.org/C2778456923","wikidata":"https://www.wikidata.org/wiki/Q5337692","display_name":"Edge computing","level":3,"score":0.37380000948905945},{"id":"https://openalex.org/C90673727","wikidata":"https://www.wikidata.org/wiki/Q901718","display_name":"Product (mathematics)","level":2,"score":0.3727000057697296},{"id":"https://openalex.org/C2780395129","wikidata":"https://www.wikidata.org/wiki/Q1128971","display_name":"Rapid prototyping","level":2,"score":0.3522000014781952},{"id":"https://openalex.org/C120793396","wikidata":"https://www.wikidata.org/wiki/Q173350","display_name":"Printed circuit board","level":2,"score":0.3337000012397766},{"id":"https://openalex.org/C13164978","wikidata":"https://www.wikidata.org/wiki/Q600158","display_name":"Hardware acceleration","level":3,"score":0.3197000026702881},{"id":"https://openalex.org/C530198007","wikidata":"https://www.wikidata.org/wiki/Q80831","display_name":"Integrated circuit","level":2,"score":0.30720001459121704},{"id":"https://openalex.org/C206729178","wikidata":"https://www.wikidata.org/wiki/Q2271896","display_name":"Scheduling (production processes)","level":2,"score":0.28790000081062317},{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.27410000562667847},{"id":"https://openalex.org/C68339613","wikidata":"https://www.wikidata.org/wiki/Q1549489","display_name":"Speedup","level":2,"score":0.26980000734329224},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.25690001249313354},{"id":"https://openalex.org/C2779229675","wikidata":"https://www.wikidata.org/wiki/Q445235","display_name":"Time to market","level":2,"score":0.2554999887943268}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1145/3748173.3779189","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3748173.3779189","pdf_url":null,"source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 2026 ACM/SIGDA International Symposium on Field Programmable Gate Arrays","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.1145/3748173.3779189","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3748173.3779189","pdf_url":null,"source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 2026 ACM/SIGDA International Symposium on Field Programmable Gate Arrays","raw_type":"proceedings-article"},"sustainable_development_goals":[{"display_name":"Decent work and economic growth","id":"https://metadata.un.org/sdg/8","score":0.5517542362213135}],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":15,"referenced_works":["https://openalex.org/W2094756095","https://openalex.org/W2194775991","https://openalex.org/W2798956872","https://openalex.org/W3008905965","https://openalex.org/W4313450353","https://openalex.org/W4379471878","https://openalex.org/W4380874786","https://openalex.org/W4388979610","https://openalex.org/W4391681217","https://openalex.org/W4393140379","https://openalex.org/W4393578753","https://openalex.org/W4393949386","https://openalex.org/W4407953477","https://openalex.org/W4407953495","https://openalex.org/W4411485931"],"related_works":[],"abstract_inverted_index":{"Field-Programmable":[0],"Gate":[1],"Arrays":[2],"(FPGAs)":[3],"have":[4],"been":[5],"shown":[6],"to":[7,45,68,97,103],"be":[8],"viable":[9],"for":[10,26,50,81,213],"Large":[11],"Language":[12],"Model":[13],"(LLM)":[14],"deployment,":[15],"but":[16],"they":[17],"remain":[18],"less":[19],"competitive":[20],"than":[21],"embedded":[22,92,104],"GPUs":[23,105],"and":[24,106,108,128,179,208],"NPUs":[25],"final":[27,78,210],"edge":[28,54,202],"products.":[29],"This":[30],"is":[31,195],"largely":[32],"because":[33],"existing":[34],"FPGA-based":[35,93,201],"LLM":[36,82,94,214],"accelerator":[37,95],"prototypes":[38],"rely":[39],"on":[40,153],"large,":[41],"expensive":[42],"FPGA":[43],"devices":[44],"provide":[46],"sufficient":[47],"hardware":[48],"resources":[49],"satisfactory":[51],"performance,":[52],"whereas":[53],"products":[55],"are":[56],"highly":[57],"cost-sensitive.":[58],"In":[59],"this":[60,161,194],"work,":[61],"we":[62,144,163],"move":[63],"beyond":[64],"pure":[65],"architectural":[66],"prototyping":[67],"evaluate":[69],"the":[70,77,147,166,196],"feasibility":[71],"of":[72,126,132,175,183,199],"using":[73],"low-cost":[74],"FPGAs":[75,155],"as":[76,205],"implementation":[79,211],"medium":[80,212],"deployment.":[83,215],"We":[84],"propose":[85],"Hummingbird+,":[86],"which":[87],"encompasses:":[88],"(1)":[89],"a":[90,110,118,172,180,206],"compact":[91],"designed":[96],"deliver":[98],"comparable":[99],"inference":[100],"performance":[101],"compared":[102],"NPUs,":[107],"(2)":[109],"custom":[111],"Printed":[112],"Circuit":[113],"Board":[114],"(PCB)":[115],"built":[116],"around":[117],"Zynq":[119],"UltraScale":[120],"XCZU2CG/3EG":[121],"SoC,":[122],"equipped":[123],"with":[124,156],"24GB":[125],"memory":[127],"an":[129,200],"expected":[130],"Bill":[131],"Materials":[133],"(BOMs)":[134],"under":[135],"\\150":[136],"in":[137],"mass":[138],"production.":[139],"Through":[140],"extensive":[141],"FPGA-centric":[142],"optimizations,":[143],"significantly":[145],"reduce":[146],"accelerator's":[148],"resource":[149],"consumption,":[150],"enabling":[151],"deployment":[152],"entry-level":[154],"exceptional":[157],"cost":[158],"efficiency.":[159],"On":[160],"platform,":[162],"successfully":[164],"deploy":[165],"GPTQ":[167],"4-bit":[168],"Qwen3-30B-A3B":[169],"LLM,":[170],"achieving":[171],"decoding":[173],"speed":[174,182],"over":[176,184],"18":[177],"token/s":[178,186],"prefill":[181],"50":[185],"without":[187],"further":[188],"model":[189],"compression.":[190],"To":[191],"our":[192],"knowledge,":[193],"first":[197],"demonstration":[198],"product":[203],"serving":[204],"practical":[207],"cost-effective":[209]},"counts_by_year":[],"updated_date":"2026-07-30T17:31:21.811387","created_date":"2026-02-07T00:00:00"}