{"id":"https://openalex.org/W4417085234","doi":"https://doi.org/10.1145/3779212.3790226","title":"STARC: Selective Token Access with Remapping and Clustering for Efficient LLM Decoding on PIM Systems","display_name":"STARC: Selective Token Access with Remapping and Clustering for Efficient LLM Decoding on PIM Systems","publication_year":2026,"publication_date":"2026-03-10","ids":{"openalex":"https://openalex.org/W4417085234","doi":"https://doi.org/10.1145/3779212.3790226"},"language":"en","primary_location":{"id":"doi:10.1145/3779212.3790226","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3779212.3790226","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 31st ACM International Conference on Architectural Support for Programming Languages and Operating Systems, Volume 2","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["arxiv","crossref","datacite"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://doi.org/10.1145/3779212.3790226","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5078919511","display_name":"Zehao Fan","orcid":"https://orcid.org/0000-0003-4513-9674"},"institutions":[{"id":"https://openalex.org/I165799507","display_name":"Rensselaer Polytechnic Institute","ror":"https://ror.org/01rtyzb94","country_code":"US","type":"education","lineage":["https://openalex.org/I165799507"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Zehao Fan","raw_affiliation_strings":["Rensselaer Polytechnic Institute, Troy, NY, USA"],"raw_orcid":"https://orcid.org/0009-0001-0021-2578","affiliations":[{"raw_affiliation_string":"Rensselaer Polytechnic Institute, Troy, NY, USA","institution_ids":["https://openalex.org/I165799507"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5119381438","display_name":"Garrett Gagnon","orcid":"https://orcid.org/0009-0009-0754-9518"},"institutions":[{"id":"https://openalex.org/I24603500","display_name":"University of Massachusetts Amherst","ror":"https://ror.org/0072zz521","country_code":"US","type":"education","lineage":["https://openalex.org/I24603500"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Yunzhen Liu","raw_affiliation_strings":["University of Massachusetts, Amherst, Amherst, MA, USA"],"raw_orcid":"https://orcid.org/0009-0006-2445-5765","affiliations":[{"raw_affiliation_string":"University of Massachusetts, Amherst, Amherst, MA, USA","institution_ids":["https://openalex.org/I24603500"]}]},{"author_position":"middle","author":{"id":null,"display_name":"Garrett Gagnon","orcid":"https://orcid.org/0009-0009-0754-9518"},"institutions":[{"id":"https://openalex.org/I165799507","display_name":"Rensselaer Polytechnic Institute","ror":"https://ror.org/01rtyzb94","country_code":"US","type":"education","lineage":["https://openalex.org/I165799507"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Garrett Gagnon","raw_affiliation_strings":["Rensselaer Polytechnic Institute, Troy, NY, USA"],"raw_orcid":"https://orcid.org/0009-0009-0754-9518","affiliations":[{"raw_affiliation_string":"Rensselaer Polytechnic Institute, Troy, NY, USA","institution_ids":["https://openalex.org/I165799507"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100324254","display_name":"Liu Liu","orcid":"https://orcid.org/0000-0001-8730-5824"},"institutions":[{"id":"https://openalex.org/I165799507","display_name":"Rensselaer Polytechnic Institute","ror":"https://ror.org/01rtyzb94","country_code":"US","type":"education","lineage":["https://openalex.org/I165799507"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Zhenyu Liu","raw_affiliation_strings":["Rensselaer Polytechnic Institute, Troy, NY, USA"],"raw_orcid":"https://orcid.org/0009-0000-0047-7026","affiliations":[{"raw_affiliation_string":"Rensselaer Polytechnic Institute, Troy, NY, USA","institution_ids":["https://openalex.org/I165799507"]}]},{"author_position":"middle","author":{"id":null,"display_name":"Yayue Hou","orcid":"https://orcid.org/0009-0009-6780-076X"},"institutions":[{"id":"https://openalex.org/I165799507","display_name":"Rensselaer Polytechnic Institute","ror":"https://ror.org/01rtyzb94","country_code":"US","type":"education","lineage":["https://openalex.org/I165799507"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Yayue Hou","raw_affiliation_strings":["Rensselaer Polytechnic Institute, Troy, NY, USA"],"raw_orcid":"https://orcid.org/0009-0009-6780-076X","affiliations":[{"raw_affiliation_string":"Rensselaer Polytechnic Institute, Troy, NY, USA","institution_ids":["https://openalex.org/I165799507"]}]},{"author_position":"middle","author":{"id":null,"display_name":"Hadjer Benmeziane","orcid":"https://orcid.org/0000-0002-5259-0749"},"institutions":[{"id":"https://openalex.org/I4210126328","display_name":"IBM Research - Zurich","ror":"https://ror.org/02js37d36","country_code":"CH","type":"facility","lineage":["https://openalex.org/I1341412227","https://openalex.org/I4210114115","https://openalex.org/I4210126328"]}],"countries":["CH"],"is_corresponding":false,"raw_author_name":"Hadjer Benmeziane","raw_affiliation_strings":["IBM Research -- Ruschlikon, Ruschlikon, Switzerland"],"raw_orcid":"https://orcid.org/0000-0002-5259-0749","affiliations":[{"raw_affiliation_string":"IBM Research -- Ruschlikon, Ruschlikon, Switzerland","institution_ids":["https://openalex.org/I4210126328"]}]},{"author_position":"middle","author":{"id":null,"display_name":"Kaoutar El Maghraoui","orcid":"https://orcid.org/0000-0002-1967-8749"},"institutions":[{"id":"https://openalex.org/I4210114115","display_name":"IBM Research - Thomas J. Watson Research Center","ror":"https://ror.org/0265w5591","country_code":"US","type":"facility","lineage":["https://openalex.org/I1341412227","https://openalex.org/I4210114115"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Kaoutar El Maghraoui","raw_affiliation_strings":["IBM T. J. Watson Research Center, Yorktown Heights, NY, USA"],"raw_orcid":"https://orcid.org/0000-0002-1967-8749","affiliations":[{"raw_affiliation_string":"IBM T. J. Watson Research Center, Yorktown Heights, NY, USA","institution_ids":["https://openalex.org/I4210114115"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5100324254","display_name":"Liu Liu","orcid":"https://orcid.org/0000-0001-8730-5824"},"institutions":[{"id":"https://openalex.org/I165799507","display_name":"Rensselaer Polytechnic Institute","ror":"https://ror.org/01rtyzb94","country_code":"US","type":"education","lineage":["https://openalex.org/I165799507"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Liu Liu","raw_affiliation_strings":["Rensselaer Polytechnic Institute, Troy, NY, USA"],"raw_orcid":"https://orcid.org/0000-0003-0792-8146","affiliations":[{"raw_affiliation_string":"Rensselaer Polytechnic Institute, Troy, NY, USA","institution_ids":["https://openalex.org/I165799507"]}]}],"institutions":[],"countries_distinct_count":2,"institutions_distinct_count":4,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"1863","last_page":"1879"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T14347","display_name":"Big Data and Digital Economy","score":0.21850000321865082,"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"}},"topics":[{"id":"https://openalex.org/T14347","display_name":"Big Data and Digital Economy","score":0.21850000321865082,"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"}},{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.1736000031232834,"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/T10036","display_name":"Advanced Neural Network Applications","score":0.08590000122785568,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/cache","display_name":"Cache","score":0.6039999723434448},{"id":"https://openalex.org/keywords/decoding-methods","display_name":"Decoding methods","score":0.531000018119812},{"id":"https://openalex.org/keywords/cluster-analysis","display_name":"Cluster analysis","score":0.5160999894142151},{"id":"https://openalex.org/keywords/bottleneck","display_name":"Bottleneck","score":0.4510999917984009},{"id":"https://openalex.org/keywords/memory-bandwidth","display_name":"Memory bandwidth","score":0.4016000032424927},{"id":"https://openalex.org/keywords/latency","display_name":"Latency (audio)","score":0.36890000104904175},{"id":"https://openalex.org/keywords/context","display_name":"Context (archaeology)","score":0.36469998955726624},{"id":"https://openalex.org/keywords/granularity","display_name":"Granularity","score":0.33320000767707825}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8815000057220459},{"id":"https://openalex.org/C115537543","wikidata":"https://www.wikidata.org/wiki/Q165596","display_name":"Cache","level":2,"score":0.6039999723434448},{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.531000018119812},{"id":"https://openalex.org/C73555534","wikidata":"https://www.wikidata.org/wiki/Q622825","display_name":"Cluster analysis","level":2,"score":0.5160999894142151},{"id":"https://openalex.org/C2780513914","wikidata":"https://www.wikidata.org/wiki/Q18210350","display_name":"Bottleneck","level":2,"score":0.4510999917984009},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.4489000141620636},{"id":"https://openalex.org/C188045654","wikidata":"https://www.wikidata.org/wiki/Q17148339","display_name":"Memory bandwidth","level":2,"score":0.4016000032424927},{"id":"https://openalex.org/C82876162","wikidata":"https://www.wikidata.org/wiki/Q17096504","display_name":"Latency (audio)","level":2,"score":0.36890000104904175},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.36469998955726624},{"id":"https://openalex.org/C177774035","wikidata":"https://www.wikidata.org/wiki/Q1246948","display_name":"Granularity","level":2,"score":0.33320000767707825},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.328000009059906},{"id":"https://openalex.org/C2781357197","wikidata":"https://www.wikidata.org/wiki/Q5757597","display_name":"High memory","level":2,"score":0.32760000228881836},{"id":"https://openalex.org/C111335779","wikidata":"https://www.wikidata.org/wiki/Q3454686","display_name":"Reduction (mathematics)","level":2,"score":0.32269999384880066},{"id":"https://openalex.org/C189783530","wikidata":"https://www.wikidata.org/wiki/Q352090","display_name":"CPU cache","level":3,"score":0.3224000036716461},{"id":"https://openalex.org/C2780165032","wikidata":"https://www.wikidata.org/wiki/Q16869822","display_name":"Energy consumption","level":2,"score":0.32109999656677246},{"id":"https://openalex.org/C157764524","wikidata":"https://www.wikidata.org/wiki/Q1383412","display_name":"Throughput","level":3,"score":0.31369999051094055},{"id":"https://openalex.org/C47487241","wikidata":"https://www.wikidata.org/wiki/Q5227230","display_name":"Data access","level":2,"score":0.3100999891757965},{"id":"https://openalex.org/C38556500","wikidata":"https://www.wikidata.org/wiki/Q13404475","display_name":"Cache algorithms","level":4,"score":0.27160000801086426},{"id":"https://openalex.org/C2742236","wikidata":"https://www.wikidata.org/wiki/Q924713","display_name":"Efficient energy use","level":2,"score":0.26170000433921814},{"id":"https://openalex.org/C113775141","wikidata":"https://www.wikidata.org/wiki/Q428691","display_name":"Computer engineering","level":1,"score":0.25859999656677246}],"mesh":[],"locations_count":4,"locations":[{"id":"doi:10.1145/3779212.3790226","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3779212.3790226","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 31st ACM International Conference on Architectural Support for Programming Languages and Operating Systems, Volume 2","raw_type":"proceedings-article"},{"id":"pmh:oai:arXiv.org:2505.05772","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2505.05772","pdf_url":"https://arxiv.org/pdf/2505.05772","source":{"id":"https://openalex.org/S4393918464","display_name":"ArXiv.org","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"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":"text"},{"id":"pmh:oai:arXiv.org:2505.05772","is_oa":true,"landing_page_url":"https://arxiv.org/abs/2505.05772","pdf_url":"https://arxiv.org/pdf/2505.05772","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"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":"text"},{"id":"doi:10.48550/arxiv.2505.05772","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2505.05772","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.1145/3779212.3790226","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3779212.3790226","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 31st ACM International Conference on Architectural Support for Programming Languages and Operating Systems, Volume 2","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":15,"referenced_works":["https://openalex.org/W4402671766","https://openalex.org/W3206328251","https://openalex.org/W2979874885","https://openalex.org/W2949989598","https://openalex.org/W2963963993","https://openalex.org/W4387321091","https://openalex.org/W4411485931","https://openalex.org/W2070246124","https://openalex.org/W4388757726","https://openalex.org/W4394745423","https://openalex.org/W4293731257","https://openalex.org/W2950368691","https://openalex.org/W3190126809","https://openalex.org/W2251818205","https://openalex.org/W4280496502"],"related_works":[],"abstract_inverted_index":{"Serving":[0],"large":[1],"language":[2],"models":[3],"(LLMs)":[4],"places":[5],"significant":[6],"pressure":[7],"on":[8,70],"memory":[9],"systems":[10],"due":[11],"to":[12,125,141,147],"frequent":[13],"accesses":[14],"and":[15,31,41,78,104,127,133,139,143],"growing":[16],"key\u2013value":[17],"(KV)":[18],"caches":[19],"as":[20],"context":[21],"lengths":[22],"increase.":[23],"Processing-in-memory":[24],"(PIM)":[25],"architectures":[26],"offer":[27],"high":[28],"internal":[29],"bandwidth":[30],"near-data":[32],"compute":[33],"parallelism,":[34],"but":[35],"current":[36],"designs":[37],"target":[38],"dense":[39],"attention":[40,103],"perform":[42],"poorly":[43],"under":[44,118],"the":[45,98],"irregular":[46],"access":[47],"patterns":[48],"of":[49],"dynamic":[50],"KV":[51,76,120],"cache":[52],"sparsity.":[53],"To":[54],"mitigate":[55],"this":[56],"limitation,":[57],"we":[58],"propose":[59],"STARC,":[60],"a":[61,114],"sparsity-optimized":[62],"data":[63],"mapping":[64],"scheme":[65],"for":[66],"efficient":[67],"LLM":[68],"decoding":[69],"PIM.":[71],"STARC":[72,122],"clusters":[73],"semantically":[74],"similar":[75],"pairs":[77],"co-locates":[79],"them":[80],"contiguously":[81],"within":[82],"PIM":[83,106],"banks,":[84],"enabling":[85],"retrieval":[86],"at":[87],"cluster":[88],"granularity":[89],"by":[90],"matching":[91],"queries":[92],"against":[93],"precomputed":[94],"centroids.":[95],"This":[96],"bridges":[97],"gap":[99],"between":[100],"fine-grained":[101],"sparse":[102],"row-level":[105],"operations,":[107],"improving":[108],"utilization":[109],"while":[110,150],"minimizing":[111],"overhead.":[112],"On":[113],"simulated":[115],"HBM-PIM":[116],"system,":[117],"constrained":[119],"budgets,":[121],"achieves":[123],"up":[124,140],"78%":[126],"65%":[128],"reductions":[129,145],"in":[130],"attention-layer":[131],"latency":[132],"energy":[134],"over":[135],"token-wise":[136],"sparsity":[137],"methods,":[138],"93%":[142],"92%":[144],"relative":[146],"full":[148],"attention,":[149],"preserving":[151],"model":[152],"accuracy.":[153]},"counts_by_year":[],"updated_date":"2026-07-15T18:14:33.161393","created_date":"2025-10-10T00:00:00"}