{"id":"https://openalex.org/W4406783634","doi":"https://doi.org/10.1145/3714416","title":"TATAA: Programmable Mixed-Precision Transformer Acceleration with a Transformable Arithmetic Architecture","display_name":"TATAA: Programmable Mixed-Precision Transformer Acceleration with a Transformable Arithmetic Architecture","publication_year":2025,"publication_date":"2025-01-24","ids":{"openalex":"https://openalex.org/W4406783634","doi":"https://doi.org/10.1145/3714416"},"language":"en","primary_location":{"id":"doi:10.1145/3714416","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3714416","pdf_url":null,"source":{"id":"https://openalex.org/S112809824","display_name":"ACM Transactions on Reconfigurable Technology and Systems","issn_l":"1936-7406","issn":["1936-7406","1936-7414"],"is_oa":false,"is_in_doaj":false,"is_core":true,"host_organization":"https://openalex.org/P4310319798","host_organization_name":"Association for Computing Machinery","host_organization_lineage":["https://openalex.org/P4310319798"],"host_organization_lineage_names":["Association for Computing Machinery"],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"ACM Transactions on Reconfigurable Technology and Systems","raw_type":"journal-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"hybrid","oa_url":"https://doi.org/10.1145/3714416","any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5016445316","display_name":"Jiajun Wu","orcid":"https://orcid.org/0000-0003-2477-3553"},"institutions":[{"id":"https://openalex.org/I889458895","display_name":"University of Hong Kong","ror":"https://ror.org/02zhqgq86","country_code":"HK","type":"education","lineage":["https://openalex.org/I889458895"]}],"countries":["HK"],"is_corresponding":false,"raw_author_name":"Jiajun Wu","raw_affiliation_strings":["Department of Electrical and Electronic Engineering, The University of Hong Kong, Hong Kong, Hong Kong"],"raw_orcid":"https://orcid.org/0000-0003-2477-3553","affiliations":[{"raw_affiliation_string":"Department of Electrical and Electronic Engineering, The University of Hong Kong, Hong Kong, Hong Kong","institution_ids":["https://openalex.org/I889458895"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5102120052","display_name":"Mo Song","orcid":null},"institutions":[{"id":"https://openalex.org/I889458895","display_name":"University of Hong Kong","ror":"https://ror.org/02zhqgq86","country_code":"HK","type":"education","lineage":["https://openalex.org/I889458895"]}],"countries":["HK"],"is_corresponding":false,"raw_author_name":"Mo Song","raw_affiliation_strings":["Department of Electrical and Electronic Engineering, The University of Hong Kong, Hong Kong, Hong Kong"],"raw_orcid":"https://orcid.org/0009-0004-5225-072X","affiliations":[{"raw_affiliation_string":"Department of Electrical and Electronic Engineering, The University of Hong Kong, Hong Kong, Hong Kong","institution_ids":["https://openalex.org/I889458895"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5114197871","display_name":"Jingmin Zhao","orcid":"https://orcid.org/0000-0003-4098-2355"},"institutions":[{"id":"https://openalex.org/I889458895","display_name":"University of Hong Kong","ror":"https://ror.org/02zhqgq86","country_code":"HK","type":"education","lineage":["https://openalex.org/I889458895"]}],"countries":["HK"],"is_corresponding":false,"raw_author_name":"Jingmin Zhao","raw_affiliation_strings":["Department of Electrical and Electronic Engineering, The University of Hong Kong, Hong Kong, Hong Kong"],"raw_orcid":"https://orcid.org/0000-0003-4098-2355","affiliations":[{"raw_affiliation_string":"Department of Electrical and Electronic Engineering, The University of Hong Kong, Hong Kong, Hong Kong","institution_ids":["https://openalex.org/I889458895"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5026079724","display_name":"Yizhao Gao","orcid":"https://orcid.org/0000-0001-5673-3746"},"institutions":[{"id":"https://openalex.org/I889458895","display_name":"University of Hong Kong","ror":"https://ror.org/02zhqgq86","country_code":"HK","type":"education","lineage":["https://openalex.org/I889458895"]}],"countries":["HK"],"is_corresponding":false,"raw_author_name":"Yizhao Gao","raw_affiliation_strings":["Department of Electrical and Electronic Engineering, The University of Hong Kong, Hong Kong, Hong Kong"],"raw_orcid":"https://orcid.org/0000-0001-5673-3746","affiliations":[{"raw_affiliation_string":"Department of Electrical and Electronic Engineering, The University of Hong Kong, Hong Kong, Hong Kong","institution_ids":["https://openalex.org/I889458895"]}]},{"author_position":"middle","author":{"id":null,"display_name":"Jia Li","orcid":"https://orcid.org/0009-0006-6492-727X"},"institutions":[{"id":"https://openalex.org/I889458895","display_name":"University of Hong Kong","ror":"https://ror.org/02zhqgq86","country_code":"HK","type":"education","lineage":["https://openalex.org/I889458895"]}],"countries":["HK"],"is_corresponding":false,"raw_author_name":"Jia Li","raw_affiliation_strings":["Department of Electrical and Electronic Engineering, The University of Hong Kong, Hong Kong, Hong Kong"],"raw_orcid":"https://orcid.org/0009-0006-6492-727X","affiliations":[{"raw_affiliation_string":"Department of Electrical and Electronic Engineering, The University of Hong Kong, Hong Kong, Hong Kong","institution_ids":["https://openalex.org/I889458895"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5020581824","display_name":"Hayden Kwok\u2010Hay So","orcid":"https://orcid.org/0000-0002-6514-0237"},"institutions":[{"id":"https://openalex.org/I889458895","display_name":"University of Hong Kong","ror":"https://ror.org/02zhqgq86","country_code":"HK","type":"education","lineage":["https://openalex.org/I889458895"]}],"countries":["HK"],"is_corresponding":false,"raw_author_name":"Hayden Kwok-Hay So","raw_affiliation_strings":["Department of Electrical and Electronic Engineering, The University of Hong Kong, Hong Kong, Hong Kong"],"raw_orcid":"https://orcid.org/0000-0002-6514-0237","affiliations":[{"raw_affiliation_string":"Department of Electrical and Electronic Engineering, The University of Hong Kong, Hong Kong, Hong Kong","institution_ids":["https://openalex.org/I889458895"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I889458895"],"apc_list":null,"apc_paid":null,"fwci":1.9725,"has_fulltext":false,"cited_by_count":6,"citation_normalized_percentile":{"value":0.84291775,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":{"min":94,"max":98},"biblio":{"volume":"18","issue":"1","first_page":"1","last_page":"31"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10323","display_name":"Analog and Mixed-Signal Circuit Design","score":0.9980999827384949,"subfield":{"id":"https://openalex.org/subfields/2204","display_name":"Biomedical 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/T10323","display_name":"Analog and Mixed-Signal Circuit Design","score":0.9980999827384949,"subfield":{"id":"https://openalex.org/subfields/2204","display_name":"Biomedical 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/T11697","display_name":"Numerical Methods and Algorithms","score":0.9925000071525574,"subfield":{"id":"https://openalex.org/subfields/1703","display_name":"Computational Theory and Mathematics"},"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/T12300","display_name":"Advanced Electrical Measurement Techniques","score":0.9919000267982483,"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/computer-science","display_name":"Computer science","score":0.810075044631958},{"id":"https://openalex.org/keywords/transformer","display_name":"Transformer","score":0.6771044135093689},{"id":"https://openalex.org/keywords/architecture","display_name":"Architecture","score":0.6480975151062012},{"id":"https://openalex.org/keywords/arithmetic","display_name":"Arithmetic","score":0.5273129343986511},{"id":"https://openalex.org/keywords/acceleration","display_name":"Acceleration","score":0.49495118856430054},{"id":"https://openalex.org/keywords/computer-hardware","display_name":"Computer hardware","score":0.43482500314712524},{"id":"https://openalex.org/keywords/parallel-computing","display_name":"Parallel computing","score":0.42826470732688904},{"id":"https://openalex.org/keywords/embedded-system","display_name":"Embedded system","score":0.3731782138347626},{"id":"https://openalex.org/keywords/computer-architecture","display_name":"Computer architecture","score":0.34452247619628906},{"id":"https://openalex.org/keywords/electrical-engineering","display_name":"Electrical engineering","score":0.21576130390167236},{"id":"https://openalex.org/keywords/mathematics","display_name":"Mathematics","score":0.09508076310157776},{"id":"https://openalex.org/keywords/voltage","display_name":"Voltage","score":0.08298182487487793}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.810075044631958},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.6771044135093689},{"id":"https://openalex.org/C123657996","wikidata":"https://www.wikidata.org/wiki/Q12271","display_name":"Architecture","level":2,"score":0.6480975151062012},{"id":"https://openalex.org/C94375191","wikidata":"https://www.wikidata.org/wiki/Q11205","display_name":"Arithmetic","level":1,"score":0.5273129343986511},{"id":"https://openalex.org/C117896860","wikidata":"https://www.wikidata.org/wiki/Q11376","display_name":"Acceleration","level":2,"score":0.49495118856430054},{"id":"https://openalex.org/C9390403","wikidata":"https://www.wikidata.org/wiki/Q3966","display_name":"Computer hardware","level":1,"score":0.43482500314712524},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.42826470732688904},{"id":"https://openalex.org/C149635348","wikidata":"https://www.wikidata.org/wiki/Q193040","display_name":"Embedded system","level":1,"score":0.3731782138347626},{"id":"https://openalex.org/C118524514","wikidata":"https://www.wikidata.org/wiki/Q173212","display_name":"Computer architecture","level":1,"score":0.34452247619628906},{"id":"https://openalex.org/C119599485","wikidata":"https://www.wikidata.org/wiki/Q43035","display_name":"Electrical engineering","level":1,"score":0.21576130390167236},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.09508076310157776},{"id":"https://openalex.org/C165801399","wikidata":"https://www.wikidata.org/wiki/Q25428","display_name":"Voltage","level":2,"score":0.08298182487487793},{"id":"https://openalex.org/C127413603","wikidata":"https://www.wikidata.org/wiki/Q11023","display_name":"Engineering","level":0,"score":0.0},{"id":"https://openalex.org/C153349607","wikidata":"https://www.wikidata.org/wiki/Q36649","display_name":"Visual arts","level":1,"score":0.0},{"id":"https://openalex.org/C142362112","wikidata":"https://www.wikidata.org/wiki/Q735","display_name":"Art","level":0,"score":0.0},{"id":"https://openalex.org/C74650414","wikidata":"https://www.wikidata.org/wiki/Q11397","display_name":"Classical mechanics","level":1,"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":2,"locations":[{"id":"doi:10.1145/3714416","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3714416","pdf_url":null,"source":{"id":"https://openalex.org/S112809824","display_name":"ACM Transactions on Reconfigurable Technology and Systems","issn_l":"1936-7406","issn":["1936-7406","1936-7414"],"is_oa":false,"is_in_doaj":false,"is_core":true,"host_organization":"https://openalex.org/P4310319798","host_organization_name":"Association for Computing Machinery","host_organization_lineage":["https://openalex.org/P4310319798"],"host_organization_lineage_names":["Association for Computing Machinery"],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"ACM Transactions on Reconfigurable Technology and Systems","raw_type":"journal-article"},{"id":"pmh:oai:hub.hku.hk:10722/366101","is_oa":false,"landing_page_url":"http://hdl.handle.net/10722/366101","pdf_url":null,"source":{"id":"https://openalex.org/S4377196271","display_name":"The HKU Scholars Hub (University of Hong Kong)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I889458895","host_organization_name":"University of Hong Kong","host_organization_lineage":["https://openalex.org/I889458895"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":null,"raw_type":"Article"}],"best_oa_location":{"id":"doi:10.1145/3714416","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3714416","pdf_url":null,"source":{"id":"https://openalex.org/S112809824","display_name":"ACM Transactions on Reconfigurable Technology and Systems","issn_l":"1936-7406","issn":["1936-7406","1936-7414"],"is_oa":false,"is_in_doaj":false,"is_core":true,"host_organization":"https://openalex.org/P4310319798","host_organization_name":"Association for Computing Machinery","host_organization_lineage":["https://openalex.org/P4310319798"],"host_organization_lineage_names":["Association for Computing Machinery"],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"ACM Transactions on Reconfigurable Technology and Systems","raw_type":"journal-article"},"sustainable_development_goals":[],"awards":[],"funders":[{"id":"https://openalex.org/F4320307285","display_name":"Impact Fund","ror":"https://ror.org/00jb20j87"}],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":52,"referenced_works":["https://openalex.org/W1583837637","https://openalex.org/W2108598243","https://openalex.org/W2140766383","https://openalex.org/W2585720638","https://openalex.org/W2604319603","https://openalex.org/W2791673912","https://openalex.org/W2906917284","https://openalex.org/W2923014074","https://openalex.org/W2963015836","https://openalex.org/W2981040094","https://openalex.org/W2990138404","https://openalex.org/W3017024317","https://openalex.org/W3021972173","https://openalex.org/W3047848469","https://openalex.org/W3093982621","https://openalex.org/W3101543398","https://openalex.org/W3112948415","https://openalex.org/W3130240120","https://openalex.org/W3138516171","https://openalex.org/W3155487259","https://openalex.org/W3184454880","https://openalex.org/W3196923642","https://openalex.org/W3199934250","https://openalex.org/W3203149535","https://openalex.org/W3211525823","https://openalex.org/W3211787299","https://openalex.org/W4212902668","https://openalex.org/W4254300870","https://openalex.org/W4281660701","https://openalex.org/W4285601701","https://openalex.org/W4286001027","https://openalex.org/W4297948009","https://openalex.org/W4300865759","https://openalex.org/W4308083513","https://openalex.org/W4308128267","https://openalex.org/W4308479898","https://openalex.org/W4383749624","https://openalex.org/W4387789586","https://openalex.org/W4388214757","https://openalex.org/W4388214800","https://openalex.org/W4388979610","https://openalex.org/W4389162736","https://openalex.org/W4390873361","https://openalex.org/W4391125526","https://openalex.org/W4391429130","https://openalex.org/W4393141090","https://openalex.org/W4393578753","https://openalex.org/W4393949386","https://openalex.org/W4393973515","https://openalex.org/W4401017962","https://openalex.org/W6767440493","https://openalex.org/W6797854001"],"related_works":["https://openalex.org/W2565094479","https://openalex.org/W2390829436","https://openalex.org/W1989791859","https://openalex.org/W602859758","https://openalex.org/W1971289376","https://openalex.org/W2379101322","https://openalex.org/W1992553864","https://openalex.org/W2776207444","https://openalex.org/W2726447019","https://openalex.org/W2006439817"],"abstract_inverted_index":{"Modern":[0],"transformer-based":[1],"deep":[2],"neural":[3],"networks":[4],"present":[5],"unique":[6],"technical":[7],"challenges":[8],"for":[9,49,81,129,137,212],"effective":[10],"acceleration":[11],"in":[12,63,222,227],"real-world":[13],"applications.":[14,189],"Apart":[15],"from":[16,150],"the":[17,155,175,194],"vast":[18],"amount":[19],"of":[20,100,183,209],"linear":[21,59,83,204],"operations":[22,62,85],"needed":[23],"due":[24],"to":[25,56,96,122,146,154,168,220],"their":[26],"sizes,":[27],"modern":[28,237],"transformer":[29,102,152,178],"models":[30,153,179],"are":[31],"increasingly":[32],"reliance":[33],"on":[34,92,193,203],"precise":[35],"non-linear":[36,61,98,213],"computations":[37],"that":[38,111,161],"make":[39],"traditional":[40],"low-bitwidth":[41],"quantization":[42],"methods":[43],"and":[44,60,66,133,186,206,225],"fixed-dataflow":[45],"matrix":[46,131],"accelerators":[47],"ineffective":[48],"end-to-end":[50,142,223],"acceleration.":[51],"To":[52],"address":[53],"this":[54,69],"need":[55],"accelerate":[57],"both":[58,113],"a":[64,101,107,125,134,181,207],"unified":[65],"programmable":[67],"framework,":[68],"article":[70],"introduces":[71],"TATAA.":[72],"TATAA":[73,104],"employs":[74],"8-bit":[75],"integer":[76],"(":[77],"int8":[78,130],")":[79],"arithmetic":[80,95,109],"quantized":[82],"layer":[84],"through":[86],"post-training":[87],"quantization,":[88],"while":[89,230],"it":[90,121],"relies":[91],"bfloat16":[93,139],"floating-point":[94],"approximate":[97],"layers":[99,205],"model.":[103],"hardware":[105],"features":[106],"transformable":[108],"architecture":[110],"supports":[112],"formats":[114],"during":[115],"runtime":[116],"with":[117,174],"minimal":[118],"overhead,":[119],"enabling":[120],"switch":[123],"between":[124],"systolic":[126],"array":[127],"mode":[128,136],"multiplications":[132],"SIMD":[135],"vectorized":[138],"operations.":[140],"An":[141],"compiler":[143],"is":[144],"presented":[145],"enable":[147],"flexible":[148],"mapping":[149],"emerging":[151],"proposed":[156],"hardware.":[157],"Experimental":[158],"results":[159],"indicate":[160],"our":[162],"mixed-precision":[163],"design":[164],"incurs":[165],"only":[166],"0.14%":[167],"1.16%":[169],"accuracy":[170],"drop":[171],"when":[172],"compared":[173],"pre-trained":[176],"single-precision":[177],"across":[180],"range":[182],"vision,":[184],"language,":[185],"generative":[187],"text":[188],"Our":[190],"prototype":[191],"implementation":[192],"Alveo":[195],"U280":[196],"FPGA":[197],"currently":[198],"achieves":[199],"2,935.2":[200],"GOPS":[201],"throughput":[202,224],"maximum":[208],"189.5":[210],"GFLOPS":[211],"operations,":[214],"outperforming":[215],"related":[216],"works":[217],"by":[218],"up":[219],"\\(1.45\\times\\)":[221],"\\(2.29\\times\\)":[226],"DSP":[228],"efficiency,":[229],"achieving":[231],"\\(2.19\\times\\)":[232],"higher":[233],"power":[234],"efficiency":[235],"than":[236],"NVIDIA":[238],"RTX4090":[239],"GPU.":[240]},"counts_by_year":[{"year":2026,"cited_by_count":1},{"year":2025,"cited_by_count":5}],"updated_date":"2026-07-31T08:31:51.225901","created_date":"2025-10-10T00:00:00"}