{"id":"https://openalex.org/W7125664763","doi":"https://doi.org/10.1109/tvlsi.2025.3650684","title":"TABv2: A Faster Ternary and Binary Neural Network Inference Library on the Edge","display_name":"TABv2: A Faster Ternary and Binary Neural Network Inference Library on the Edge","publication_year":2026,"publication_date":"2026-01-22","ids":{"openalex":"https://openalex.org/W7125664763","doi":"https://doi.org/10.1109/tvlsi.2025.3650684"},"language":"en","primary_location":{"id":"doi:10.1109/tvlsi.2025.3650684","is_oa":false,"landing_page_url":"https://doi.org/10.1109/tvlsi.2025.3650684","pdf_url":null,"source":{"id":"https://openalex.org/S37538908","display_name":"IEEE Transactions on Very Large Scale Integration (VLSI) Systems","issn_l":"1063-8210","issn":["1063-8210","1557-9999"],"is_oa":false,"is_in_doaj":false,"is_core":true,"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":"journal"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"IEEE Transactions on Very Large Scale Integration (VLSI) Systems","raw_type":"journal-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":null,"display_name":"Guanshujie Fu","orcid":"https://orcid.org/0000-0002-9615-8456"},"institutions":[{"id":"https://openalex.org/I35440088","display_name":"ETH Zurich","ror":"https://ror.org/05a28rw58","country_code":"CH","type":"education","lineage":["https://openalex.org/I2799323385","https://openalex.org/I35440088"]}],"countries":["CH"],"is_corresponding":false,"raw_author_name":"Guanshujie Fu","raw_affiliation_strings":["Department of Computer Science, Systems Group, ETH Zurich, Zurich, Switzerland"],"raw_orcid":"https://orcid.org/0000-0002-9615-8456","affiliations":[{"raw_affiliation_string":"Department of Computer Science, Systems Group, ETH Zurich, Zurich, Switzerland","institution_ids":["https://openalex.org/I35440088"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5112138421","display_name":"Olivier Fischer","orcid":null},"institutions":[{"id":"https://openalex.org/I35440088","display_name":"ETH Zurich","ror":"https://ror.org/05a28rw58","country_code":"CH","type":"education","lineage":["https://openalex.org/I2799323385","https://openalex.org/I35440088"]}],"countries":["CH"],"is_corresponding":false,"raw_author_name":"Olivier Fischer","raw_affiliation_strings":["Department of Computer Science, Systems Group, ETH Zurich, Zurich, Switzerland"],"raw_orcid":"https://orcid.org/0009-0004-5074-6940","affiliations":[{"raw_affiliation_string":"Department of Computer Science, Systems Group, ETH Zurich, Zurich, Switzerland","institution_ids":["https://openalex.org/I35440088"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5087479741","display_name":"Shien Zhu","orcid":"https://orcid.org/0000-0002-2094-7643"},"institutions":[{"id":"https://openalex.org/I35440088","display_name":"ETH Zurich","ror":"https://ror.org/05a28rw58","country_code":"CH","type":"education","lineage":["https://openalex.org/I2799323385","https://openalex.org/I35440088"]}],"countries":["CH"],"is_corresponding":false,"raw_author_name":"Shien Zhu","raw_affiliation_strings":["Department of Computer Science, Systems Group, ETH Zurich, Zurich, Switzerland"],"raw_orcid":"https://orcid.org/0000-0002-2094-7643","affiliations":[{"raw_affiliation_string":"Department of Computer Science, Systems Group, ETH Zurich, Zurich, Switzerland","institution_ids":["https://openalex.org/I35440088"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5123476453","display_name":"Gustavo Alonso","orcid":null},"institutions":[{"id":"https://openalex.org/I35440088","display_name":"ETH Zurich","ror":"https://ror.org/05a28rw58","country_code":"CH","type":"education","lineage":["https://openalex.org/I2799323385","https://openalex.org/I35440088"]}],"countries":["CH"],"is_corresponding":false,"raw_author_name":"Gustavo Alonso","raw_affiliation_strings":["Department of Computer Science, Systems Group, ETH Zurich, Zurich, Switzerland"],"raw_orcid":"https://orcid.org/0000-0002-4396-6695","affiliations":[{"raw_affiliation_string":"Department of Computer Science, Systems Group, ETH Zurich, Zurich, Switzerland","institution_ids":["https://openalex.org/I35440088"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I35440088"],"apc_list":null,"apc_paid":null,"fwci":15.909,"has_fulltext":false,"cited_by_count":1,"citation_normalized_percentile":{"value":0.96940849,"is_in_top_1_percent":false,"is_in_top_10_percent":true},"cited_by_percentile_year":{"min":95,"max":97},"biblio":{"volume":"34","issue":"4","first_page":"1097","last_page":"1109"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10036","display_name":"Advanced Neural Network Applications","score":0.707099974155426,"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"}},"topics":[{"id":"https://openalex.org/T10036","display_name":"Advanced Neural Network Applications","score":0.707099974155426,"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/T10028","display_name":"Topic Modeling","score":0.0333000011742115,"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/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.021400000900030136,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/bitwise-operation","display_name":"Bitwise operation","score":0.7498999834060669},{"id":"https://openalex.org/keywords/scalar-multiplication","display_name":"Scalar multiplication","score":0.6161999702453613},{"id":"https://openalex.org/keywords/simd","display_name":"SIMD","score":0.5724999904632568},{"id":"https://openalex.org/keywords/artificial-neural-network","display_name":"Artificial neural network","score":0.4869000017642975},{"id":"https://openalex.org/keywords/binary-number","display_name":"Binary number","score":0.4595000147819519},{"id":"https://openalex.org/keywords/quantization","display_name":"Quantization (signal processing)","score":0.438400000333786},{"id":"https://openalex.org/keywords/intrinsics","display_name":"Intrinsics","score":0.39410001039505005},{"id":"https://openalex.org/keywords/byte","display_name":"Byte","score":0.3874000012874603},{"id":"https://openalex.org/keywords/ternary-operation","display_name":"Ternary operation","score":0.3783000111579895}],"concepts":[{"id":"https://openalex.org/C134765980","wikidata":"https://www.wikidata.org/wiki/Q879126","display_name":"Bitwise operation","level":2,"score":0.7498999834060669},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7073000073432922},{"id":"https://openalex.org/C171182647","wikidata":"https://www.wikidata.org/wiki/Q126736","display_name":"Scalar multiplication","level":3,"score":0.6161999702453613},{"id":"https://openalex.org/C150552126","wikidata":"https://www.wikidata.org/wiki/Q339387","display_name":"SIMD","level":2,"score":0.5724999904632568},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.4869000017642975},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.48429998755455017},{"id":"https://openalex.org/C48372109","wikidata":"https://www.wikidata.org/wiki/Q3913","display_name":"Binary number","level":2,"score":0.4595000147819519},{"id":"https://openalex.org/C28855332","wikidata":"https://www.wikidata.org/wiki/Q198099","display_name":"Quantization (signal processing)","level":2,"score":0.438400000333786},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.42170000076293945},{"id":"https://openalex.org/C2908650547","wikidata":"https://www.wikidata.org/wiki/Q20999234","display_name":"Intrinsics","level":2,"score":0.39410001039505005},{"id":"https://openalex.org/C43364308","wikidata":"https://www.wikidata.org/wiki/Q8799","display_name":"Byte","level":2,"score":0.3874000012874603},{"id":"https://openalex.org/C64452783","wikidata":"https://www.wikidata.org/wiki/Q1524945","display_name":"Ternary operation","level":2,"score":0.3783000111579895},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.34779998660087585},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.33340001106262207},{"id":"https://openalex.org/C17020691","wikidata":"https://www.wikidata.org/wiki/Q139677","display_name":"Operator (biology)","level":5,"score":0.31859999895095825},{"id":"https://openalex.org/C81363708","wikidata":"https://www.wikidata.org/wiki/Q17084460","display_name":"Convolutional neural network","level":2,"score":0.30649998784065247},{"id":"https://openalex.org/C2779190172","wikidata":"https://www.wikidata.org/wiki/Q4913888","display_name":"Binary data","level":3,"score":0.3052999973297119},{"id":"https://openalex.org/C63435697","wikidata":"https://www.wikidata.org/wiki/Q864135","display_name":"Binary code","level":3,"score":0.29649999737739563},{"id":"https://openalex.org/C2776445388","wikidata":"https://www.wikidata.org/wiki/Q6716998","display_name":"MODTRAN","level":3,"score":0.29159998893737793},{"id":"https://openalex.org/C153338461","wikidata":"https://www.wikidata.org/wiki/Q2651","display_name":"Arithmetic coding","level":4,"score":0.28870001435279846},{"id":"https://openalex.org/C152124472","wikidata":"https://www.wikidata.org/wiki/Q1204361","display_name":"Redundancy (engineering)","level":2,"score":0.2818000018596649},{"id":"https://openalex.org/C124584101","wikidata":"https://www.wikidata.org/wiki/Q1053266","display_name":"Multiplier (economics)","level":2,"score":0.2791999876499176},{"id":"https://openalex.org/C138236772","wikidata":"https://www.wikidata.org/wiki/Q25098575","display_name":"Edge device","level":3,"score":0.2761000096797943},{"id":"https://openalex.org/C94375191","wikidata":"https://www.wikidata.org/wiki/Q11205","display_name":"Arithmetic","level":1,"score":0.267300009727478},{"id":"https://openalex.org/C159423971","wikidata":"https://www.wikidata.org/wiki/Q177251","display_name":"Associative property","level":2,"score":0.2671000063419342},{"id":"https://openalex.org/C2780595030","wikidata":"https://www.wikidata.org/wiki/Q3860309","display_name":"Multiplication (music)","level":2,"score":0.26100000739097595},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.2605000138282776}],"mesh":[],"locations_count":2,"locations":[{"id":"doi:10.1109/tvlsi.2025.3650684","is_oa":false,"landing_page_url":"https://doi.org/10.1109/tvlsi.2025.3650684","pdf_url":null,"source":{"id":"https://openalex.org/S37538908","display_name":"IEEE Transactions on Very Large Scale Integration (VLSI) Systems","issn_l":"1063-8210","issn":["1063-8210","1557-9999"],"is_oa":false,"is_in_doaj":false,"is_core":true,"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":"journal"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"IEEE Transactions on Very Large Scale Integration (VLSI) Systems","raw_type":"journal-article"},{"id":"pmh:doi:10.3929/ethz-c-000795257","is_oa":false,"landing_page_url":"http://hdl.handle.net/20.500.11850/795257","pdf_url":null,"source":{"id":"https://openalex.org/S4406922384","display_name":"Open MIND","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Journal Article"}],"best_oa_location":null,"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Modern":[0],"deep":[1],"neural":[2,42],"networks":[3,43,54],"and":[4,11,18,21,35,40,46,51,56,67,86,92,103,111,122,160,164,174,180,213,247,286,312,325,335,343,349,363],"large":[5],"language":[6],"models":[7],"(LLMs)":[8],"are":[9,70],"trained":[10],"inferred":[12],"using":[13],"low-bitwidth":[14],"representations":[15],"like":[16],"FP8":[17],"FP4.":[19],"Ternary":[20],"binary":[22,41,104,112],"quantization":[23,121,173,193],"can":[24,225],"further":[25,263],"reduce":[26,214,226],"the":[27,47,63,76,172,186,204,215,218,275,318],"computation":[28],"cost":[29],"by":[30,177,196,249],"extreme":[31],"1/2-bit":[32],"data":[33,117,187,198],"representation":[34],"bitwise":[36,233,242,252,269],"operations.":[37],"As":[38],"ternary":[39,102,110,222,232],"(TNNs,":[44],"BNNs,":[45,158],"mixed-precision":[48,93],"ternary-activation":[49],"binary-weight":[50],"binary-activation":[52],"ternary-weight":[53],"(TBNs":[55],"BTNs))":[57],"achieve":[58,153,298,329],"different":[59],"trade-off":[60],"points":[61],"between":[62],"speed,":[64],"model":[65],"size,":[66],"accuracy,":[68],"they":[69,114],"suitable":[71],"for":[72,109,156,245,259,268,321],"certain":[73],"applications":[74],"on":[75,83,128,162,194,207,256,271,305,309,314,341,355,360],"edge.":[77],"However,":[78],"existing":[79],"works":[80,99],"mainly":[81],"focus":[82],"accelerating":[84],"BNN":[85],"DoReFa-Net-style":[87],"bit-serial":[88],"operations,":[89],"leaving":[90],"TNNs":[91,246],"ones":[94],"under-optimized.":[95],"Although":[96],"some":[97],"related":[98],"such":[100],"as":[101],"(TAB)":[105],"provide":[106],"reference":[107],"implementation":[108],"networks,":[113],"encounter":[115],"high":[116],"access":[118,277],"overhead":[119],"during":[120],"suffer":[123],"from":[124],"slow":[125],"scalar":[126,205],"popcount":[127,144,206,220],"advanced":[129],"vector":[130],"extension":[131],"2":[132],"(AVX2)":[133],"central":[134],"processing":[135,166],"unit":[136,167],"(CPU)":[137],"which":[138,224],"have":[139],"no":[140],"single-instruction":[141],"multiple-data":[142],"(SIMD)":[143],"instructions.":[145],"In":[146],"this":[147],"article,":[148],"we":[149,170,202,239,280,297,328],"propose":[150,191,240],"TABv2":[151],"to":[152,184,273,303,358],"faster":[154],"inference":[155],"TNNs,":[157,322],"TBNs,":[159,323],"BTNs":[161],"CPU":[163],"graphic":[165],"(GPU).":[168],"First,":[169],"optimize":[171],"image-to-row":[175],"(img2row)":[176],"operator":[178],"fusion":[179],"better":[181],"address":[182],"calculation":[183],"improve":[185],"locality.":[188],"We":[189,262],"also":[190],"warp-cooperative":[192],"GPU":[195,272,316,356],"utilizing":[197,221,250],"sync":[199],"intrinsics.":[200],"Second,":[201],"replace":[203],"AVX2":[208,306],"with":[209],"equivalent":[210],"SIMD":[211,219],"operations":[212],"complexity":[216],"of":[217,301],"encoding,":[223],"approximately":[227],"15%":[228],"total":[229],"instruction":[230],"in":[231,284],"general":[234],"matrix-matrix":[235],"multiplication":[236,254],"(GEMM).":[237],"Third,":[238],"new":[241],"GEMM":[243,270],"algorithms":[244],"TBNs":[248],"GPU-native":[251],"matrix":[253],"intrinsics":[255],"tensor":[257],"cores":[258],"higher":[260],"efficiency.":[261],"apply":[264],"a":[265],"four-degree":[266],"pipeline":[267],"hide":[274],"memory":[276],"latency.":[278],"Finally,":[279],"implement":[281],"these":[282],"methods":[283],"C++":[285],"combine":[287],"them":[288],"into":[289],"an":[290],"open-source":[291],"library.":[292],"Evaluation":[293],"results":[294],"show":[295],"that":[296],"layer-level":[299],"speedup":[300,334,348],"up":[302],"2.7\u00d7":[304],"CPU,":[307,311],"2.3\u00d7":[308],"ARM":[310],"8.7\u00d7":[313],"Nvidia":[315],"over":[317],"TAB":[319,359],"baseline":[320],"BTNs,":[324],"BNNs.":[326],"Moreover,":[327],"1.3\u00d7":[330,344],"-":[331,337,345,351],"1.9\u00d7":[332],"end-to-end":[333,347],"1.2\u00d7":[336,350],"1.8\u00d7":[338],"energy":[339,353],"efficiency":[340,354],"CPUs":[342],"3.5\u00d7":[346],"3.0\u00d7":[352],"compared":[357],"ResNet,":[361],"Darknet,":[362],"visual":[364],"geometry":[365],"group":[366],"(VGG)":[367],"models.":[368]},"counts_by_year":[{"year":2026,"cited_by_count":1}],"updated_date":"2026-06-26T08:34:08.712188","created_date":"2026-01-26T00:00:00"}