{"id":"https://openalex.org/W4408563999","doi":"https://doi.org/10.1109/tcsi.2025.3549060","title":"Topkima-Former: Low-Energy, Low-Latency Inference for Transformers Using Top- k In-Memory ADC","display_name":"Topkima-Former: Low-Energy, Low-Latency Inference for Transformers Using Top- k In-Memory ADC","publication_year":2025,"publication_date":"2025-03-18","ids":{"openalex":"https://openalex.org/W4408563999","doi":"https://doi.org/10.1109/tcsi.2025.3549060"},"language":"en","primary_location":{"id":"doi:10.1109/tcsi.2025.3549060","is_oa":false,"landing_page_url":"https://doi.org/10.1109/tcsi.2025.3549060","pdf_url":null,"source":{"id":"https://openalex.org/S116977442","display_name":"IEEE Transactions on Circuits and Systems I Regular Papers","issn_l":"1549-8328","issn":["1549-8328","1558-0806"],"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 Circuits and Systems I: Regular Papers","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":"Shuai Dong","orcid":"https://orcid.org/0009-0007-4807-5094"},"institutions":[{"id":"https://openalex.org/I168719708","display_name":"City University of Hong Kong","ror":"https://ror.org/03q8dnn23","country_code":"HK","type":"education","lineage":["https://openalex.org/I168719708"]}],"countries":["HK"],"is_corresponding":false,"raw_author_name":"Shuai Dong","raw_affiliation_strings":["Department of Electrical Engineering, City University of Hong Kong, Kowloon, Hong Kong","Department of Electrical Engineering, City University of Hong Kong, Hong Kong, Hong Kong"],"raw_orcid":"https://orcid.org/0009-0007-4807-5094","affiliations":[{"raw_affiliation_string":"Department of Electrical Engineering, City University of Hong Kong, Kowloon, Hong Kong","institution_ids":["https://openalex.org/I168719708"]},{"raw_affiliation_string":"Department of Electrical Engineering, City University of Hong Kong, Hong Kong, Hong Kong","institution_ids":["https://openalex.org/I168719708"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5001439194","display_name":"Junyi Yang","orcid":"https://orcid.org/0000-0002-5867-4943"},"institutions":[{"id":"https://openalex.org/I168719708","display_name":"City University of Hong Kong","ror":"https://ror.org/03q8dnn23","country_code":"HK","type":"education","lineage":["https://openalex.org/I168719708"]}],"countries":["HK"],"is_corresponding":false,"raw_author_name":"Junyi Yang","raw_affiliation_strings":["Department of Electrical Engineering, City University of Hong Kong, Kowloon, Hong Kong","Department of Electrical Engineering, City University of Hong Kong, Hong Kong, Hong Kong"],"raw_orcid":"https://orcid.org/0000-0002-5867-4943","affiliations":[{"raw_affiliation_string":"Department of Electrical Engineering, City University of Hong Kong, Kowloon, Hong Kong","institution_ids":["https://openalex.org/I168719708"]},{"raw_affiliation_string":"Department of Electrical Engineering, City University of Hong Kong, Hong Kong, Hong Kong","institution_ids":["https://openalex.org/I168719708"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5102547113","display_name":"Xiaoqi Peng","orcid":"https://orcid.org/0009-0005-7716-5318"},"institutions":[{"id":"https://openalex.org/I168719708","display_name":"City University of Hong Kong","ror":"https://ror.org/03q8dnn23","country_code":"HK","type":"education","lineage":["https://openalex.org/I168719708"]}],"countries":["HK"],"is_corresponding":false,"raw_author_name":"Xiaoqi Peng","raw_affiliation_strings":["Department of Electrical Engineering, City University of Hong Kong, Kowloon, Hong Kong","Department of Electrical Engineering, City University of Hong Kong, Hong Kong, Hong Kong"],"raw_orcid":"https://orcid.org/0009-0005-7716-5318","affiliations":[{"raw_affiliation_string":"Department of Electrical Engineering, City University of Hong Kong, Kowloon, Hong Kong","institution_ids":["https://openalex.org/I168719708"]},{"raw_affiliation_string":"Department of Electrical Engineering, City University of Hong Kong, Hong Kong, Hong Kong","institution_ids":["https://openalex.org/I168719708"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5109797096","display_name":"H. Shang","orcid":"https://orcid.org/0009-0007-6276-1947"},"institutions":[{"id":"https://openalex.org/I168719708","display_name":"City University of Hong Kong","ror":"https://ror.org/03q8dnn23","country_code":"HK","type":"education","lineage":["https://openalex.org/I168719708"]}],"countries":["HK"],"is_corresponding":false,"raw_author_name":"Hongyang Shang","raw_affiliation_strings":["Department of Electrical Engineering, City University of Hong Kong, Kowloon, Hong Kong","Department of Electrical Engineering, City University of Hong Kong, Hong Kong, Hong Kong"],"raw_orcid":"https://orcid.org/0009-0007-6276-1947","affiliations":[{"raw_affiliation_string":"Department of Electrical Engineering, City University of Hong Kong, Kowloon, Hong Kong","institution_ids":["https://openalex.org/I168719708"]},{"raw_affiliation_string":"Department of Electrical Engineering, City University of Hong Kong, Hong Kong, Hong Kong","institution_ids":["https://openalex.org/I168719708"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5104284748","display_name":"Ye Ke","orcid":"https://orcid.org/0009-0002-9809-1192"},"institutions":[{"id":"https://openalex.org/I168719708","display_name":"City University of Hong Kong","ror":"https://ror.org/03q8dnn23","country_code":"HK","type":"education","lineage":["https://openalex.org/I168719708"]}],"countries":["HK"],"is_corresponding":false,"raw_author_name":"Ye Ke","raw_affiliation_strings":["Department of Electrical Engineering, City University of Hong Kong, Kowloon, Hong Kong","Department of Electrical Engineering, City University of Hong Kong, Hong Kong, Hong Kong"],"raw_orcid":"https://orcid.org/0009-0002-9809-1192","affiliations":[{"raw_affiliation_string":"Department of Electrical Engineering, City University of Hong Kong, Kowloon, Hong Kong","institution_ids":["https://openalex.org/I168719708"]},{"raw_affiliation_string":"Department of Electrical Engineering, City University of Hong Kong, Hong Kong, Hong Kong","institution_ids":["https://openalex.org/I168719708"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5063285225","display_name":"Xiaofeng Yang","orcid":"https://orcid.org/0000-0003-1400-7994"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xiaofeng Yang","raw_affiliation_strings":["Reexen Technology, Shenzhen, China"],"raw_orcid":"https://orcid.org/0000-0003-1400-7994","affiliations":[{"raw_affiliation_string":"Reexen Technology, Shenzhen, China","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5010273308","display_name":"H. Liu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hongjie Liu","raw_affiliation_strings":["Reexen Technology, Shenzhen, China"],"raw_orcid":"https://orcid.org/0009-0005-0904-5736","affiliations":[{"raw_affiliation_string":"Reexen Technology, Shenzhen, China","institution_ids":[]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5002380437","display_name":"Arindam Basu","orcid":"https://orcid.org/0000-0003-1035-8770"},"institutions":[{"id":"https://openalex.org/I168719708","display_name":"City University of Hong Kong","ror":"https://ror.org/03q8dnn23","country_code":"HK","type":"education","lineage":["https://openalex.org/I168719708"]}],"countries":["HK"],"is_corresponding":false,"raw_author_name":"Arindam Basu","raw_affiliation_strings":["Department of Electrical Engineering, City University of Hong Kong, Kowloon, Hong Kong","Department of Electrical Engineering, City University of Hong Kong, Hong Kong, Hong Kong"],"raw_orcid":"https://orcid.org/0000-0003-1035-8770","affiliations":[{"raw_affiliation_string":"Department of Electrical Engineering, City University of Hong Kong, Kowloon, Hong Kong","institution_ids":["https://openalex.org/I168719708"]},{"raw_affiliation_string":"Department of Electrical Engineering, City University of Hong Kong, Hong Kong, Hong Kong","institution_ids":["https://openalex.org/I168719708"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":{"value":2645,"currency":"USD","value_usd":2645},"apc_paid":null,"fwci":1.8943,"has_fulltext":false,"cited_by_count":7,"citation_normalized_percentile":{"value":0.84983323,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":{"min":96,"max":99},"biblio":{"volume":"72","issue":"6","first_page":"2509","last_page":"2519"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10502","display_name":"Advanced Memory and Neural Computing","score":0.9994999766349792,"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"}},"topics":[{"id":"https://openalex.org/T10502","display_name":"Advanced Memory and Neural Computing","score":0.9994999766349792,"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"}},{"id":"https://openalex.org/T12808","display_name":"Ferroelectric and Negative Capacitance Devices","score":0.9961000084877014,"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"}},{"id":"https://openalex.org/T10472","display_name":"Semiconductor materials and devices","score":0.9958999752998352,"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":0.5513524413108826},{"id":"https://openalex.org/keywords/transformer","display_name":"Transformer","score":0.5100651383399963},{"id":"https://openalex.org/keywords/successive-approximation-adc","display_name":"Successive approximation ADC","score":0.48008662462234497},{"id":"https://openalex.org/keywords/electronic-engineering","display_name":"Electronic engineering","score":0.4511508047580719},{"id":"https://openalex.org/keywords/electrical-engineering","display_name":"Electrical engineering","score":0.34638431668281555},{"id":"https://openalex.org/keywords/capacitor","display_name":"Capacitor","score":0.24647632241249084},{"id":"https://openalex.org/keywords/engineering","display_name":"Engineering","score":0.20862668752670288},{"id":"https://openalex.org/keywords/voltage","display_name":"Voltage","score":0.19944453239440918}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5513524413108826},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.5100651383399963},{"id":"https://openalex.org/C60154766","wikidata":"https://www.wikidata.org/wiki/Q2650458","display_name":"Successive approximation ADC","level":4,"score":0.48008662462234497},{"id":"https://openalex.org/C24326235","wikidata":"https://www.wikidata.org/wiki/Q126095","display_name":"Electronic engineering","level":1,"score":0.4511508047580719},{"id":"https://openalex.org/C119599485","wikidata":"https://www.wikidata.org/wiki/Q43035","display_name":"Electrical engineering","level":1,"score":0.34638431668281555},{"id":"https://openalex.org/C52192207","wikidata":"https://www.wikidata.org/wiki/Q5322","display_name":"Capacitor","level":3,"score":0.24647632241249084},{"id":"https://openalex.org/C127413603","wikidata":"https://www.wikidata.org/wiki/Q11023","display_name":"Engineering","level":0,"score":0.20862668752670288},{"id":"https://openalex.org/C165801399","wikidata":"https://www.wikidata.org/wiki/Q25428","display_name":"Voltage","level":2,"score":0.19944453239440918}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1109/tcsi.2025.3549060","is_oa":false,"landing_page_url":"https://doi.org/10.1109/tcsi.2025.3549060","pdf_url":null,"source":{"id":"https://openalex.org/S116977442","display_name":"IEEE Transactions on Circuits and Systems I Regular Papers","issn_l":"1549-8328","issn":["1549-8328","1558-0806"],"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 Circuits and Systems I: Regular Papers","raw_type":"journal-article"}],"best_oa_location":null,"sustainable_development_goals":[{"display_name":"Affordable and clean energy","score":0.8299999833106995,"id":"https://metadata.un.org/sdg/7"}],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":38,"referenced_works":["https://openalex.org/W206948248","https://openalex.org/W2020003503","https://openalex.org/W2042519306","https://openalex.org/W2114833655","https://openalex.org/W2167148127","https://openalex.org/W2906917284","https://openalex.org/W2946480111","https://openalex.org/W2963899139","https://openalex.org/W2964176059","https://openalex.org/W3017968097","https://openalex.org/W3044192339","https://openalex.org/W3094502228","https://openalex.org/W3112740243","https://openalex.org/W3115909741","https://openalex.org/W3139521791","https://openalex.org/W3159727696","https://openalex.org/W3189877953","https://openalex.org/W3211525823","https://openalex.org/W4205141919","https://openalex.org/W4214833843","https://openalex.org/W4226402784","https://openalex.org/W4281874271","https://openalex.org/W4293023328","https://openalex.org/W4293768835","https://openalex.org/W4312847929","https://openalex.org/W4360832001","https://openalex.org/W4378801105","https://openalex.org/W4381233128","https://openalex.org/W4387042316","https://openalex.org/W4389166707","https://openalex.org/W4389459141","https://openalex.org/W4390017976","https://openalex.org/W4400726540","https://openalex.org/W4405267264","https://openalex.org/W6691766336","https://openalex.org/W6755207826","https://openalex.org/W6788001715","https://openalex.org/W6810643285"],"related_works":["https://openalex.org/W4391375266","https://openalex.org/W2899084033","https://openalex.org/W2748952813","https://openalex.org/W2390279801","https://openalex.org/W4391913857","https://openalex.org/W2358668433","https://openalex.org/W4396701345","https://openalex.org/W2376932109","https://openalex.org/W2001405890","https://openalex.org/W2864363823"],"abstract_inverted_index":{"Transformer":[0],"has":[1],"emerged":[2],"as":[3,122],"a":[4,26,97,110,131,169,174],"leading":[5],"architecture":[6,50,127],"in":[7,202],"neural":[8],"language":[9],"processing":[10],"(NLP)":[11],"and":[12,32,49,139,173,205],"computer":[13],"vision":[14,204],"(CV).":[15],"However,":[16],"the":[17,38,46,54,57,87,94,105,114,126],"extensive":[18],"use":[19],"of":[20,37,91],"nonlinear":[21],"operations,":[22],"like":[23],"softmax,":[24],"poses":[25],"performance":[27],"bottleneck":[28],"during":[29,104],"transformer":[30],"inference":[31],"comprises":[33],"up":[34],"to":[35,52,69,82,112,168,195],"40%":[36],"total":[39],"latency.":[40],"Hence,":[41],"we":[42,60,129],"propose":[43,61],"innovations":[44],"at":[45],"circuit,":[47],"algorithm":[48],"levels":[51],"accelerate":[53],"transformer.":[55],"At":[56,93,125],"circuit":[58],"level,":[59,96,128],"Topkima\u2014combining":[62],"top-kactivation":[63],"selection":[64],"with":[65,109],"in-memory":[66],"ADC":[67],"(IMA)":[68],"implement":[70],"efficient":[71],"softmax":[72,83,171,177,182],"without":[73],"any":[74],"sorting":[75],"overhead.":[76],"Only":[77],"theklargest":[78],"activations":[79],"are":[80],"sent":[81],"calculation":[84],"block,":[85],"reducing":[86],"huge":[88],"computational":[89],"cost":[90],"softmax.":[92],"algorithmic":[95],"modified":[98],"training":[99],"scheme":[100],"utilizes":[101],"top-kactivations":[102],"only":[103],"forward":[106],"pass,":[107],"combined":[108,149],"sub-top-kmethod":[111],"address":[113],"crossbar":[115],"size":[116],"limitation":[117],"by":[118],"aggregating":[119],"each":[120],"sub-top-kvalues":[121],"global":[123],"top-k.":[124],"introduce":[130],"fine":[132],"pipeline":[133],"for":[134,144,199],"efficiently":[135],"scheduling":[136],"data":[137],"flows":[138],"an":[140],"improved":[141],"scale-free":[142],"technique":[143],"removing":[145],"scaling":[146],"cost.":[147],"The":[148],"system,":[150],"dubbed":[151],"Topkima-Former,":[152],"enhances$1.8\\times":[153],"-84\\times":[154],"$speedup":[155],"and$1.2\\times":[156],"-36\\times":[157],"$energy":[158],"efficiency":[159],"(EE)":[160],"over":[161],"prior":[162],"In-memory":[163],"computing":[164],"(IMC)":[165],"accelerators.":[166],"Compared":[167],"conventional":[170],"macro":[172,183],"digital":[175],"top-k(Dtopk)":[176],"macro,":[178],"our":[179],"proposed":[180],"Topkima":[181],"achieves":[184],"about$15\\times":[185],"$and$8\\times":[186],"$faster":[187],"speed":[188],"respectively.":[189],"Experimental":[190],"evaluations":[191],"demonstrate":[192],"minimal":[193],"(0.42%":[194],"1.60%)":[196],"accuracy":[197],"loss":[198],"different":[200],"models":[201],"both":[203],"NLP":[206],"tasks.":[207]},"counts_by_year":[{"year":2026,"cited_by_count":4},{"year":2025,"cited_by_count":3}],"updated_date":"2026-08-28T12:50:07.497085","created_date":"2025-03-19T00:00:00"}
