{"id":"https://openalex.org/W4415445367","doi":"https://doi.org/10.1109/lca.2025.3624272","title":"PNM Meets Sparse Attention: Enabling Multi-Million Tokens Inference at Scale","display_name":"PNM Meets Sparse Attention: Enabling Multi-Million Tokens Inference at Scale","publication_year":2025,"publication_date":"2025-07-01","ids":{"openalex":"https://openalex.org/W4415445367","doi":"https://doi.org/10.1109/lca.2025.3624272"},"language":"en","primary_location":{"id":"doi:10.1109/lca.2025.3624272","doi":"https://doi.org/10.1109/lca.2025.3624272","is_oa":false,"landing_page_url":"https://doi.org/10.1109/lca.2025.3624272","pdf_url":null,"source":{"id":"https://openalex.org/S17643076","display_name":"IEEE Computer Architecture Letters","issn_l":"1556-6056","issn":["1556-6056","1556-6064","2473-2575"],"is_oa":false,"is_in_doaj":false,"is_core":true,"listed_in":["cwts-core","dongbi-b","norway-1","poland-70","russia-white-list-2"],"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 Computer Architecture Letters","raw_type":"journal-article"},"volume":null,"type":"article","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/A5108548114","display_name":"Sookyung Choi","orcid":"https://orcid.org/0009-0008-2452-7321","observed_orcids":["https://orcid.org/0009-0008-2452-7321"]},"institutions":[{"id":"https://openalex.org/I4405272917","display_name":"SK Hynix (South Korea)","ror":"https://ror.org/04z2p0w98","country_code":"KR","type":"company","lineage":["https://openalex.org/I4405272917"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Sookyung Choi","raw_affiliation_strings":["SK hynix Inc., Icheon-si, South Korea","SK hynix Inc., Icheon-si, Gyeonggi-do, South Korea"],"raw_orcid":"https://orcid.org/0009-0008-2452-7321","affiliations":[{"raw_affiliation_string":"SK hynix Inc., Icheon-si, South Korea","institution_ids":["https://openalex.org/I4405272917"]},{"raw_affiliation_string":"SK hynix Inc., Icheon-si, Gyeonggi-do, South Korea","institution_ids":["https://openalex.org/I4405272917"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5088514623","display_name":"Myunghyun Rhee","orcid":"https://orcid.org/0009-0003-5359-9942","observed_orcids":["https://orcid.org/0009-0003-5359-9942"]},"institutions":[{"id":"https://openalex.org/I4405272917","display_name":"SK Hynix (South Korea)","ror":"https://ror.org/04z2p0w98","country_code":"KR","type":"company","lineage":["https://openalex.org/I4405272917"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Myunghyun Rhee","raw_affiliation_strings":["SK hynix Inc., Icheon-si, South Korea","SK hynix Inc., Icheon-si, Gyeonggi-do, South Korea"],"raw_orcid":"https://orcid.org/0009-0003-5359-9942","affiliations":[{"raw_affiliation_string":"SK hynix Inc., Icheon-si, South Korea","institution_ids":["https://openalex.org/I4405272917"]},{"raw_affiliation_string":"SK hynix Inc., Icheon-si, Gyeonggi-do, South Korea","institution_ids":["https://openalex.org/I4405272917"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5045299073","display_name":"Euiseok Kim","orcid":"https://orcid.org/0000-0002-4223-0221","observed_orcids":["https://orcid.org/0000-0002-4223-0221"]},"institutions":[{"id":"https://openalex.org/I4405272917","display_name":"SK Hynix (South Korea)","ror":"https://ror.org/04z2p0w98","country_code":"KR","type":"company","lineage":["https://openalex.org/I4405272917"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Euiseok Kim","raw_affiliation_strings":["SK hynix Inc., Icheon-si, South Korea","SK hynix Inc., Icheon-si, Gyeonggi-do, South Korea"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"SK hynix Inc., Icheon-si, South Korea","institution_ids":["https://openalex.org/I4405272917"]},{"raw_affiliation_string":"SK hynix Inc., Icheon-si, Gyeonggi-do, South Korea","institution_ids":["https://openalex.org/I4405272917"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5112107904","display_name":"Kwang-Sik Shin","orcid":"https://orcid.org/0009-0000-5144-5673","observed_orcids":["https://orcid.org/0009-0000-5144-5673"]},"institutions":[{"id":"https://openalex.org/I4405272917","display_name":"SK Hynix (South Korea)","ror":"https://ror.org/04z2p0w98","country_code":"KR","type":"company","lineage":["https://openalex.org/I4405272917"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Kwangsik Shin","raw_affiliation_strings":["SK hynix Inc., Icheon-si, South Korea","SK hynix Inc., Icheon-si, Gyeonggi-do, South Korea"],"raw_orcid":"https://orcid.org/0009-0000-5144-5673","affiliations":[{"raw_affiliation_string":"SK hynix Inc., Icheon-si, South Korea","institution_ids":["https://openalex.org/I4405272917"]},{"raw_affiliation_string":"SK hynix Inc., Icheon-si, Gyeonggi-do, South Korea","institution_ids":["https://openalex.org/I4405272917"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5063948587","display_name":"Youngpyo Joo","orcid":"https://orcid.org/0000-0003-4391-6361","observed_orcids":["https://orcid.org/0000-0003-4391-6361"]},"institutions":[{"id":"https://openalex.org/I4405272917","display_name":"SK Hynix (South Korea)","ror":"https://ror.org/04z2p0w98","country_code":"KR","type":"company","lineage":["https://openalex.org/I4405272917"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Youngpyo Joo","raw_affiliation_strings":["SK hynix Inc., Icheon-si, South Korea","SK hynix Inc., Icheon-si, Gyeonggi-do, South Korea"],"raw_orcid":"https://orcid.org/0000-0003-4391-6361","affiliations":[{"raw_affiliation_string":"SK hynix Inc., Icheon-si, South Korea","institution_ids":["https://openalex.org/I4405272917"]},{"raw_affiliation_string":"SK hynix Inc., Icheon-si, Gyeonggi-do, South Korea","institution_ids":["https://openalex.org/I4405272917"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5008944433","display_name":"Hoshik Kim","orcid":"https://orcid.org/0000-0002-4017-8124","observed_orcids":["https://orcid.org/0000-0002-4017-8124"]},"institutions":[{"id":"https://openalex.org/I4405272917","display_name":"SK Hynix (South Korea)","ror":"https://ror.org/04z2p0w98","country_code":"KR","type":"company","lineage":["https://openalex.org/I4405272917"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Hoshik Kim","raw_affiliation_strings":["SK hynix Inc., Icheon-si, South Korea","SK hynix Inc., Icheon-si, Gyeonggi-do, South Korea"],"raw_orcid":"https://orcid.org/0000-0002-4017-8124","affiliations":[{"raw_affiliation_string":"SK hynix Inc., Icheon-si, South Korea","institution_ids":["https://openalex.org/I4405272917"]},{"raw_affiliation_string":"SK hynix Inc., Icheon-si, Gyeonggi-do, South Korea","institution_ids":["https://openalex.org/I4405272917"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I4405272917"],"apc_list":{"value":2645,"currency":"USD","value_usd":2645},"apc_paid":null,"fwci":2.3324,"has_fulltext":false,"cited_by_count":5,"citation_normalized_percentile":{"value":0.86936262,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":{"min":97,"max":98},"biblio":{"volume":"24","issue":"2","first_page":"353","last_page":"356"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.9028000235557556,"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/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.9028000235557556,"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/T10502","display_name":"Advanced Memory and Neural Computing","score":0.0215000007301569,"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/T10317","display_name":"Advanced Database Systems and Queries","score":0.008700000122189522,"subfield":{"id":"https://openalex.org/subfields/1710","display_name":"Information Systems"},"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/sparse-attention","display_name":"sparse attention","score":0.8479999899864197},{"id":"https://openalex.org/keywords/dynamic-sparse-attention","display_name":"dynamic sparse attention","score":0.7149999737739563},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.7300000190734863},{"id":"https://openalex.org/keywords/memory-bottleneck","display_name":"memory bottleneck","score":0.7369999885559082},{"id":"https://openalex.org/keywords/capacity-scaling","display_name":"capacity scaling","score":0.7200000286102295},{"id":"https://openalex.org/keywords/memory-bandwidth","display_name":"memory bandwidth","score":0.4729999899864197}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8999999761581421},{"id":"https://openalex.org/C2780513914","wikidata":"https://www.wikidata.org/wiki/Q18210350","display_name":"Bottleneck","level":2,"score":0.8223999738693237},{"id":"https://openalex.org/C188045654","wikidata":"https://www.wikidata.org/wiki/Q17148339","display_name":"Memory bandwidth","level":2,"score":0.6644999980926514},{"id":"https://openalex.org/C68339613","wikidata":"https://www.wikidata.org/wiki/Q1549489","display_name":"Speedup","level":2,"score":0.6065000295639038},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5587999820709229},{"id":"https://openalex.org/C98045186","wikidata":"https://www.wikidata.org/wiki/Q205663","display_name":"Process (computing)","level":2,"score":0.5216000080108643},{"id":"https://openalex.org/C176649486","wikidata":"https://www.wikidata.org/wiki/Q2308807","display_name":"Memory management","level":3,"score":0.4763000011444092},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.46700000762939453},{"id":"https://openalex.org/C2776257435","wikidata":"https://www.wikidata.org/wiki/Q1576430","display_name":"Bandwidth (computing)","level":2,"score":0.42809998989105225},{"id":"https://openalex.org/C120314980","wikidata":"https://www.wikidata.org/wiki/Q180634","display_name":"Distributed computing","level":1,"score":0.42329999804496765},{"id":"https://openalex.org/C123657996","wikidata":"https://www.wikidata.org/wiki/Q12271","display_name":"Architecture","level":2,"score":0.41530001163482666},{"id":"https://openalex.org/C2779602883","wikidata":"https://www.wikidata.org/wiki/Q15544750","display_name":"Memory architecture","level":2,"score":0.41359999775886536},{"id":"https://openalex.org/C118524514","wikidata":"https://www.wikidata.org/wiki/Q173212","display_name":"Computer architecture","level":1,"score":0.4122999906539917},{"id":"https://openalex.org/C62611344","wikidata":"https://www.wikidata.org/wiki/Q1062658","display_name":"Node (physics)","level":2,"score":0.38019999861717224},{"id":"https://openalex.org/C118702147","wikidata":"https://www.wikidata.org/wiki/Q189396","display_name":"Dynamic random-access memory","level":3,"score":0.3434999883174896},{"id":"https://openalex.org/C2778755073","wikidata":"https://www.wikidata.org/wiki/Q10858537","display_name":"Scale (ratio)","level":2,"score":0.31380000710487366},{"id":"https://openalex.org/C99844830","wikidata":"https://www.wikidata.org/wiki/Q102441924","display_name":"Scaling","level":2,"score":0.31299999356269836},{"id":"https://openalex.org/C57863822","wikidata":"https://www.wikidata.org/wiki/Q905488","display_name":"Flat memory model","level":4,"score":0.3109999895095825},{"id":"https://openalex.org/C107598950","wikidata":"https://www.wikidata.org/wiki/Q259864","display_name":"Microarchitecture","level":2,"score":0.3107999861240387},{"id":"https://openalex.org/C2781357197","wikidata":"https://www.wikidata.org/wiki/Q5757597","display_name":"High memory","level":2,"score":0.3084999918937683},{"id":"https://openalex.org/C12186640","wikidata":"https://www.wikidata.org/wiki/Q6815743","display_name":"Memory model","level":3,"score":0.3019999861717224},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.295199990272522},{"id":"https://openalex.org/C202491316","wikidata":"https://www.wikidata.org/wiki/Q272683","display_name":"Instruction set","level":2,"score":0.2948000133037567},{"id":"https://openalex.org/C113775141","wikidata":"https://www.wikidata.org/wiki/Q428691","display_name":"Computer engineering","level":1,"score":0.29179999232292175},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.28369998931884766},{"id":"https://openalex.org/C39528615","wikidata":"https://www.wikidata.org/wiki/Q1229610","display_name":"Distributed shared memory","level":5,"score":0.2824999988079071},{"id":"https://openalex.org/C206345919","wikidata":"https://www.wikidata.org/wiki/Q20380951","display_name":"Resource (disambiguation)","level":2,"score":0.27810001373291016},{"id":"https://openalex.org/C106515295","wikidata":"https://www.wikidata.org/wiki/Q26806595","display_name":"Parallel processing","level":2,"score":0.2736000120639801},{"id":"https://openalex.org/C199360897","wikidata":"https://www.wikidata.org/wiki/Q9143","display_name":"Programming language","level":1,"score":0.27230000495910645},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.2549999952316284},{"id":"https://openalex.org/C46743427","wikidata":"https://www.wikidata.org/wiki/Q1341685","display_name":"Inference engine","level":3,"score":0.2531999945640564},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.25209999084472656},{"id":"https://openalex.org/C174998907","wikidata":"https://www.wikidata.org/wiki/Q357662","display_name":"Work in process","level":2,"score":0.25099998712539673}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1109/lca.2025.3624272","doi":"https://doi.org/10.1109/lca.2025.3624272","is_oa":false,"landing_page_url":"https://doi.org/10.1109/lca.2025.3624272","pdf_url":null,"source":{"id":"https://openalex.org/S17643076","display_name":"IEEE Computer Architecture Letters","issn_l":"1556-6056","issn":["1556-6056","1556-6064","2473-2575"],"is_oa":false,"is_in_doaj":false,"is_core":true,"listed_in":["cwts-core","dongbi-b","norway-1","poland-70","russia-white-list-2"],"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 Computer Architecture Letters","raw_type":"journal-article"}],"best_oa_location":null,"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":8,"referenced_works":["https://openalex.org/W2997517014","https://openalex.org/W3191222816","https://openalex.org/W4312304862","https://openalex.org/W4393407316","https://openalex.org/W4407197060","https://openalex.org/W4409248815","https://openalex.org/W4410636796","https://openalex.org/W4415798413"],"related_works":["https://openalex.org/W7140680812","https://openalex.org/W4417403290","https://openalex.org/W7134111884","https://openalex.org/W4410086824","https://openalex.org/W7168290336","https://openalex.org/W4405029226","https://openalex.org/W7128803222","https://openalex.org/W4405254006","https://openalex.org/W7143433589","https://openalex.org/W7133490529"],"abstract_inverted_index":{"Processing":[0,53],"multi-million":[1],"tokens":[2],"for":[3,14,120],"advanced":[4],"Large":[5],"Language":[6],"Models":[7],"(LLMs)":[8],"poses":[9],"a":[10,22,87,98],"significant":[11],"memory":[12,28],"bottleneck":[13,19],"existing":[15,108],"AI":[16,122],"systems.":[17],"This":[18,68],"stems":[20],"from":[21],"fundamental":[23],"resource":[24],"imbalance,":[25],"where":[26],"enormous":[27],"capacity":[29,71],"and":[30,76,115],"bandwidth":[31],"are":[32],"required,":[33],"yet":[34],"the":[35,51,58,118],"computational":[36],"load":[37],"is":[38],"minimal.":[39],"We":[40],"proposeNELSSA(ProcessingNear":[41],"Memory":[42,55],"forExtremelyLongSequences":[43],"withSparseAttention),":[44],"an":[45,92],"architectural":[46],"platform":[47],"that":[48],"synergistically":[49],"combines":[50],"high-capacity":[52],"Near":[54],"(PNM)":[56],"with":[57],"principles":[59],"of":[60],"dynamic":[61],"sparse":[62],"attention":[63],"to":[64,83,94],"address":[65],"this":[66],"issue.":[67],"approach":[69],"enables":[70],"scaling":[72],"without":[73],"performance":[74],"degradation,":[75],"our":[77],"evaluation":[78],"shows":[79],"thatNELSSAcan":[80],"process":[81],"up":[82],"20M-token":[84],"sequences":[85],"on":[86],"single":[88],"node":[89],"(Llama-2-70B),":[90],"achieving":[91],"11\u00d7":[93],"40\u00d7":[95],"speedup":[96],"over":[97],"representative":[99],"DIMM-based":[100],"PNM":[101],"system.":[102],"The":[103],"proposed":[104],"architecture":[105],"radically":[106],"resolves":[107],"inefficiencies,":[109],"enabling":[110],"previously":[111],"impractical":[112],"multi-million-token":[113],"processing":[114],"thus":[116],"laying":[117],"foundation":[119],"next-generation":[121],"applications.":[123]},"counts_by_year":[{"year":2026,"cited_by_count":5}],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2025-10-24T00:00:00"}