{"id":"https://openalex.org/W7163519015","doi":"https://doi.org/10.23919/date69613.2026.11539662","title":"From Buffers to Registers: Unlocking Fine-Grained FlashAttention with Hybrid-Bonded 3D NPU Co-Design","display_name":"From Buffers to Registers: Unlocking Fine-Grained FlashAttention with Hybrid-Bonded 3D NPU Co-Design","publication_year":2026,"publication_date":"2026-04-20","ids":{"openalex":"https://openalex.org/W7163519015","doi":"https://doi.org/10.23919/date69613.2026.11539662"},"language":null,"primary_location":{"id":"doi:10.23919/date69613.2026.11539662","is_oa":false,"landing_page_url":"https://doi.org/10.23919/date69613.2026.11539662","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2026 Design, Automation &amp;amp; Test in Europe Conference (DATE)","raw_type":"proceedings-article"},"type":"conference-paper","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/A5137901078","display_name":"Jinxin Yu","orcid":null},"institutions":[{"id":"https://openalex.org/I19820366","display_name":"Chinese Academy of Sciences","ror":"https://ror.org/034t30j35","country_code":"CN","type":"government","lineage":["https://openalex.org/I19820366"]},{"id":"https://openalex.org/I4210090176","display_name":"Institute of Computing Technology","ror":"https://ror.org/0090r4d87","country_code":"CN","type":"facility","lineage":["https://openalex.org/I19820366","https://openalex.org/I4210090176"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Jinxin Yu","raw_affiliation_strings":["SKLP, Institute of Computing Technology,Chinese Academy of Sciences,Beijing,China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"SKLP, Institute of Computing Technology,Chinese Academy of Sciences,Beijing,China","institution_ids":["https://openalex.org/I19820366","https://openalex.org/I4210090176"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101320367","display_name":"Yudong Pan","orcid":"https://orcid.org/0009-0001-0012-4113"},"institutions":[{"id":"https://openalex.org/I19820366","display_name":"Chinese Academy of Sciences","ror":"https://ror.org/034t30j35","country_code":"CN","type":"government","lineage":["https://openalex.org/I19820366"]},{"id":"https://openalex.org/I4210090176","display_name":"Institute of Computing Technology","ror":"https://ror.org/0090r4d87","country_code":"CN","type":"facility","lineage":["https://openalex.org/I19820366","https://openalex.org/I4210090176"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Yudong Pan","raw_affiliation_strings":["SKLP, Institute of Computing Technology,Chinese Academy of Sciences,Beijing,China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"SKLP, Institute of Computing Technology,Chinese Academy of Sciences,Beijing,China","institution_ids":["https://openalex.org/I19820366","https://openalex.org/I4210090176"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5079264255","display_name":"M Wang","orcid":null},"institutions":[{"id":"https://openalex.org/I19820366","display_name":"Chinese Academy of Sciences","ror":"https://ror.org/034t30j35","country_code":"CN","type":"government","lineage":["https://openalex.org/I19820366"]},{"id":"https://openalex.org/I4210090176","display_name":"Institute of Computing Technology","ror":"https://ror.org/0090r4d87","country_code":"CN","type":"facility","lineage":["https://openalex.org/I19820366","https://openalex.org/I4210090176"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Mengdi Wang","raw_affiliation_strings":["SKLP, Institute of Computing Technology,Chinese Academy of Sciences,Beijing,China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"SKLP, Institute of Computing Technology,Chinese Academy of Sciences,Beijing,China","institution_ids":["https://openalex.org/I19820366","https://openalex.org/I4210090176"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5065859286","display_name":"Hui Li","orcid":"https://orcid.org/0000-0001-9198-3951"},"institutions":[{"id":"https://openalex.org/I19820366","display_name":"Chinese Academy of Sciences","ror":"https://ror.org/034t30j35","country_code":"CN","type":"government","lineage":["https://openalex.org/I19820366"]},{"id":"https://openalex.org/I4210090176","display_name":"Institute of Computing Technology","ror":"https://ror.org/0090r4d87","country_code":"CN","type":"facility","lineage":["https://openalex.org/I19820366","https://openalex.org/I4210090176"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Huawei Li","raw_affiliation_strings":["SKLP, Institute of Computing Technology,Chinese Academy of Sciences,Beijing,China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"SKLP, Institute of Computing Technology,Chinese Academy of Sciences,Beijing,China","institution_ids":["https://openalex.org/I19820366","https://openalex.org/I4210090176"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5111302581","display_name":"Y Han","orcid":null},"institutions":[{"id":"https://openalex.org/I19820366","display_name":"Chinese Academy of Sciences","ror":"https://ror.org/034t30j35","country_code":"CN","type":"government","lineage":["https://openalex.org/I19820366"]},{"id":"https://openalex.org/I4210090176","display_name":"Institute of Computing Technology","ror":"https://ror.org/0090r4d87","country_code":"CN","type":"facility","lineage":["https://openalex.org/I19820366","https://openalex.org/I4210090176"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Yinhe Han","raw_affiliation_strings":["SKLP, Institute of Computing Technology,Chinese Academy of Sciences,Beijing,China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"SKLP, Institute of Computing Technology,Chinese Academy of Sciences,Beijing,China","institution_ids":["https://openalex.org/I19820366","https://openalex.org/I4210090176"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5041607266","display_name":"Li X","orcid":"https://orcid.org/0000-0002-5981-2762"},"institutions":[{"id":"https://openalex.org/I19820366","display_name":"Chinese Academy of Sciences","ror":"https://ror.org/034t30j35","country_code":"CN","type":"government","lineage":["https://openalex.org/I19820366"]},{"id":"https://openalex.org/I4210090176","display_name":"Institute of Computing Technology","ror":"https://ror.org/0090r4d87","country_code":"CN","type":"facility","lineage":["https://openalex.org/I19820366","https://openalex.org/I4210090176"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Xiaowei Li","raw_affiliation_strings":["SKLP, Institute of Computing Technology,Chinese Academy of Sciences,Beijing,China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"SKLP, Institute of Computing Technology,Chinese Academy of Sciences,Beijing,China","institution_ids":["https://openalex.org/I19820366","https://openalex.org/I4210090176"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5137829485","display_name":"Ying Wang","orcid":null},"institutions":[{"id":"https://openalex.org/I19820366","display_name":"Chinese Academy of Sciences","ror":"https://ror.org/034t30j35","country_code":"CN","type":"government","lineage":["https://openalex.org/I19820366"]},{"id":"https://openalex.org/I4210090176","display_name":"Institute of Computing Technology","ror":"https://ror.org/0090r4d87","country_code":"CN","type":"facility","lineage":["https://openalex.org/I19820366","https://openalex.org/I4210090176"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Ying Wang","raw_affiliation_strings":["SKLP, Institute of Computing Technology,Chinese Academy of Sciences,Beijing,China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"SKLP, Institute of Computing Technology,Chinese Academy of Sciences,Beijing,China","institution_ids":["https://openalex.org/I19820366","https://openalex.org/I4210090176"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":{"value":0.63542271,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"1","last_page":"7"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11527","display_name":"3D IC and TSV technologies","score":0.8094000220298767,"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/T11527","display_name":"3D IC and TSV technologies","score":0.8094000220298767,"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/T11522","display_name":"VLSI and FPGA Design Techniques","score":0.02879999950528145,"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/T10460","display_name":"Electronic Packaging and Soldering Technologies","score":0.01140000019222498,"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/component","display_name":"Component (thermodynamics)","score":0.28299999237060547},{"id":"https://openalex.org/keywords/work","display_name":"Work (physics)","score":0.2337999939918518},{"id":"https://openalex.org/keywords/diafiltration","display_name":"Diafiltration","score":0.20990000665187836},{"id":"https://openalex.org/keywords/production","display_name":"Production (economics)","score":0.2094999998807907}],"concepts":[{"id":"https://openalex.org/C185592680","wikidata":"https://www.wikidata.org/wiki/Q2329","display_name":"Chemistry","level":0,"score":0.3815000057220459},{"id":"https://openalex.org/C168167062","wikidata":"https://www.wikidata.org/wiki/Q1117970","display_name":"Component (thermodynamics)","level":2,"score":0.28299999237060547},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.2745000123977661},{"id":"https://openalex.org/C21880701","wikidata":"https://www.wikidata.org/wiki/Q2144042","display_name":"Process engineering","level":1,"score":0.2619999945163727},{"id":"https://openalex.org/C39432304","wikidata":"https://www.wikidata.org/wiki/Q188847","display_name":"Environmental science","level":0,"score":0.24459999799728394},{"id":"https://openalex.org/C192562407","wikidata":"https://www.wikidata.org/wiki/Q228736","display_name":"Materials science","level":0,"score":0.24009999632835388},{"id":"https://openalex.org/C18762648","wikidata":"https://www.wikidata.org/wiki/Q42213","display_name":"Work (physics)","level":2,"score":0.2337999939918518},{"id":"https://openalex.org/C18743360","wikidata":"https://www.wikidata.org/wiki/Q1208096","display_name":"Diafiltration","level":4,"score":0.20990000665187836},{"id":"https://openalex.org/C2778348673","wikidata":"https://www.wikidata.org/wiki/Q739302","display_name":"Production (economics)","level":2,"score":0.2094999998807907},{"id":"https://openalex.org/C2779530757","wikidata":"https://www.wikidata.org/wiki/Q1207505","display_name":"Quality (philosophy)","level":2,"score":0.20669999718666077}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.23919/date69613.2026.11539662","is_oa":false,"landing_page_url":"https://doi.org/10.23919/date69613.2026.11539662","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2026 Design, Automation &amp;amp; Test in Europe Conference (DATE)","raw_type":"proceedings-article"}],"best_oa_location":null,"sustainable_development_goals":[],"awards":[],"funders":[{"id":"https://openalex.org/F4320321001","display_name":"National Natural Science Foundation of China","ror":"https://ror.org/01h0zpd94"}],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":23,"referenced_works":["https://openalex.org/W1999085092","https://openalex.org/W2067523571","https://openalex.org/W2289252105","https://openalex.org/W2622148586","https://openalex.org/W2907477236","https://openalex.org/W2935331687","https://openalex.org/W2940862705","https://openalex.org/W2952221193","https://openalex.org/W2998732502","https://openalex.org/W3157657667","https://openalex.org/W3163293080","https://openalex.org/W3177358108","https://openalex.org/W4220943275","https://openalex.org/W4312561048","https://openalex.org/W4318541578","https://openalex.org/W4384826052","https://openalex.org/W4386763770","https://openalex.org/W4386763835","https://openalex.org/W4388757726","https://openalex.org/W4391744705","https://openalex.org/W4402715152","https://openalex.org/W4404955120","https://openalex.org/W7133227460"],"related_works":[],"abstract_inverted_index":{"Transformer-based":[0],"models":[1],"dominate":[2],"modern":[3],"AI":[4],"workloads":[5,144],"but":[6],"exacerbate":[7],"memory":[8],"bottlenecks":[9],"due":[10],"to":[11,42,107,164],"their":[12],"quadratic":[13],"attention":[14,46],"complexity":[15],"and":[16,25,146,159,167],"ever-growing":[17],"model":[18],"sizes.":[19],"Existing":[20],"accelerators,":[21],"such":[22,37],"as":[23,38,49],"Groq":[24],"Cerebras,":[26],"mitigate":[27],"off-chip":[28,50],"traffic":[29,51],"with":[30,112],"large":[31,45],"on-chip":[32,57,138],"caches,":[33],"while":[34],"algorithmic":[35],"innovations":[36],"FlashAttention":[39],"fuse":[40],"operators":[41],"avoid":[43],"materializing":[44],"matrices.":[47],"However,":[48],"decreases,":[52],"our":[53,151],"measurements":[54],"show":[55,149],"that":[56,83,127,150],"SRAM":[58,139],"accesses":[59],"account":[60],"for":[61],"over":[62],"60%":[63],"of":[64,117],"energy":[65,157],"in":[66],"long-sequence":[67],"workloads,":[68],"making":[69],"cache":[70],"access":[71],"the":[72],"new":[73],"bottleneck.":[74],"We":[75],"propose":[76],"3D-Flow,":[77],"a":[78,123,133],"hybrid-bonded,":[79],"3D-stacked":[80],"spatial":[81,153],"accelerator":[82,154],"enables":[84],"register-to-register":[85],"communication":[86],"across":[87,130],"vertically":[88],"partitioned":[89],"PE":[90],"tiers.":[91],"Unlike":[92],"2D":[93,166],"multi-array":[94],"architectures":[95],"limited":[96],"by":[97],"NoC-based":[98],"router-to-router":[99],"transfers,":[100],"3D-Flow":[101],"leverages":[102],"sub-10":[103],"\u03bcm":[104],"vertical":[105,135],"TSVs":[106],"sustain":[108],"cycle-level":[109],"operator":[110],"pipelining":[111],"minimal":[113],"overhead.":[114],"On":[115],"top":[116],"this":[118],"architecture,":[119],"we":[120],"design":[121],"3D-FlashAttention,":[122],"fine-grained":[124],"scheduling":[125],"method":[126],"balances":[128],"latency":[129],"tiers,":[131],"forming":[132],"bubble-free":[134],"dataflow":[136],"without":[137],"round-trips.":[140],"Evaluations":[141],"on":[142],"Transformer":[143],"(OPT":[145],"QWEN":[147],"models)":[148],"3D":[152,168],"reduces":[155],"46\u201393%":[156],"consumption":[158],"achieves":[160],"1.4\u00d7\u20137.6\u00d7":[161],"speedups":[162],"compared":[163],"state-of-the-art":[165],"designs.":[169]},"counts_by_year":[],"updated_date":"2026-07-29T14:22:42.915294","created_date":"2026-06-05T00:00:00"}
