{"id":"https://openalex.org/W7106124743","doi":"https://doi.org/10.1109/iccad66269.2025.11240713","title":"H <sup>2</sup> EAL: Hybrid-Bonding Architecture with Hybrid Sparse Attention for Efficient Long-Context LLM Inference","display_name":"H <sup>2</sup> EAL: Hybrid-Bonding Architecture with Hybrid Sparse Attention for Efficient Long-Context LLM Inference","publication_year":2025,"publication_date":"2025-10-26","ids":{"openalex":"https://openalex.org/W7106124743","doi":"https://doi.org/10.1109/iccad66269.2025.11240713"},"language":null,"primary_location":{"id":"doi:10.1109/iccad66269.2025.11240713","is_oa":false,"landing_page_url":"https://doi.org/10.1109/iccad66269.2025.11240713","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2025 IEEE/ACM International Conference On Computer Aided Design (ICCAD)","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":null,"display_name":"Zizhuo Fu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zizhuo Fu","raw_affiliation_strings":["Institute for Artificial Intelligence"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Institute for Artificial Intelligence","institution_ids":[]}]},{"author_position":"middle","author":{"id":null,"display_name":"Xiaotian Guo","orcid":null},"institutions":[{"id":"https://openalex.org/I20231570","display_name":"Peking University","ror":"https://ror.org/02v51f717","country_code":"CN","type":"education","lineage":["https://openalex.org/I20231570"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Xiaotian Guo","raw_affiliation_strings":["Peking University,School of Integrated Circuits,Beijing,China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Peking University,School of Integrated Circuits,Beijing,China","institution_ids":["https://openalex.org/I20231570"]}]},{"author_position":"middle","author":{"id":null,"display_name":"Wenxuan Zeng","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wenxuan Zeng","raw_affiliation_strings":["Institute for Artificial Intelligence"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Institute for Artificial Intelligence","institution_ids":[]}]},{"author_position":"middle","author":{"id":null,"display_name":"Shuzhang Zhong","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Shuzhang Zhong","raw_affiliation_strings":["Institute for Artificial Intelligence"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Institute for Artificial Intelligence","institution_ids":[]}]},{"author_position":"middle","author":{"id":null,"display_name":"Yadong Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yadong Zhang","raw_affiliation_strings":["Nano Core Chip Electronic Technology,Hangzhou,China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Nano Core Chip Electronic Technology,Hangzhou,China","institution_ids":[]}]},{"author_position":"middle","author":{"id":null,"display_name":"Peiyu Chen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Peiyu Chen","raw_affiliation_strings":["Nano Core Chip Electronic Technology,Hangzhou,China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Nano Core Chip Electronic Technology,Hangzhou,China","institution_ids":[]}]},{"author_position":"middle","author":{"id":null,"display_name":"Runsheng Wang","orcid":null},"institutions":[{"id":"https://openalex.org/I20231570","display_name":"Peking University","ror":"https://ror.org/02v51f717","country_code":"CN","type":"education","lineage":["https://openalex.org/I20231570"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Runsheng Wang","raw_affiliation_strings":["Peking University,School of Integrated Circuits,Beijing,China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Peking University,School of Integrated Circuits,Beijing,China","institution_ids":["https://openalex.org/I20231570"]}]},{"author_position":"middle","author":{"id":null,"display_name":"Le Ye","orcid":null},"institutions":[{"id":"https://openalex.org/I20231570","display_name":"Peking University","ror":"https://ror.org/02v51f717","country_code":"CN","type":"education","lineage":["https://openalex.org/I20231570"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Le Ye","raw_affiliation_strings":["Peking University,School of Integrated Circuits,Beijing,China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Peking University,School of Integrated Circuits,Beijing,China","institution_ids":["https://openalex.org/I20231570"]}]},{"author_position":"last","author":{"id":null,"display_name":"Meng Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Meng Li","raw_affiliation_strings":["Institute for Artificial Intelligence"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Institute for Artificial Intelligence","institution_ids":[]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"cited_by_count":1,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"1","last_page":"9"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10036","display_name":"Advanced Neural Network Applications","score":0.21610000729560852,"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":0.21610000729560852,"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/T14347","display_name":"Big Data and Digital Economy","score":0.13429999351501465,"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/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.10040000081062317,"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/speedup","display_name":"Speedup","score":0.5885999798774719},{"id":"https://openalex.org/keywords/efficient-energy-use","display_name":"Efficient energy use","score":0.5361999869346619},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.5343000292778015},{"id":"https://openalex.org/keywords/leverage","display_name":"Leverage (statistics)","score":0.517300009727478},{"id":"https://openalex.org/keywords/overhead","display_name":"Overhead (engineering)","score":0.5062999725341797},{"id":"https://openalex.org/keywords/energy-consumption","display_name":"Energy consumption","score":0.44850000739097595},{"id":"https://openalex.org/keywords/sparse-matrix","display_name":"Sparse matrix","score":0.41370001435279846},{"id":"https://openalex.org/keywords/cache","display_name":"Cache","score":0.40630000829696655}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8360000252723694},{"id":"https://openalex.org/C68339613","wikidata":"https://www.wikidata.org/wiki/Q1549489","display_name":"Speedup","level":2,"score":0.5885999798774719},{"id":"https://openalex.org/C2742236","wikidata":"https://www.wikidata.org/wiki/Q924713","display_name":"Efficient energy use","level":2,"score":0.5361999869346619},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5343000292778015},{"id":"https://openalex.org/C153083717","wikidata":"https://www.wikidata.org/wiki/Q6535263","display_name":"Leverage (statistics)","level":2,"score":0.517300009727478},{"id":"https://openalex.org/C2779960059","wikidata":"https://www.wikidata.org/wiki/Q7113681","display_name":"Overhead (engineering)","level":2,"score":0.5062999725341797},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.48559999465942383},{"id":"https://openalex.org/C2780165032","wikidata":"https://www.wikidata.org/wiki/Q16869822","display_name":"Energy consumption","level":2,"score":0.44850000739097595},{"id":"https://openalex.org/C56372850","wikidata":"https://www.wikidata.org/wiki/Q1050404","display_name":"Sparse matrix","level":3,"score":0.41370001435279846},{"id":"https://openalex.org/C113775141","wikidata":"https://www.wikidata.org/wiki/Q428691","display_name":"Computer engineering","level":1,"score":0.40630000829696655},{"id":"https://openalex.org/C115537543","wikidata":"https://www.wikidata.org/wiki/Q165596","display_name":"Cache","level":2,"score":0.40630000829696655},{"id":"https://openalex.org/C82876162","wikidata":"https://www.wikidata.org/wiki/Q17096504","display_name":"Latency (audio)","level":2,"score":0.3959999978542328},{"id":"https://openalex.org/C2778476105","wikidata":"https://www.wikidata.org/wiki/Q628539","display_name":"Workload","level":2,"score":0.3874000012874603},{"id":"https://openalex.org/C2984118289","wikidata":"https://www.wikidata.org/wiki/Q29954","display_name":"Power consumption","level":3,"score":0.383899986743927},{"id":"https://openalex.org/C2778915421","wikidata":"https://www.wikidata.org/wiki/Q3643177","display_name":"Performance improvement","level":2,"score":0.3684999942779541},{"id":"https://openalex.org/C77618280","wikidata":"https://www.wikidata.org/wiki/Q1155772","display_name":"Scheme (mathematics)","level":2,"score":0.34880000352859497},{"id":"https://openalex.org/C188045654","wikidata":"https://www.wikidata.org/wiki/Q17148339","display_name":"Memory bandwidth","level":2,"score":0.3425999879837036},{"id":"https://openalex.org/C2776257435","wikidata":"https://www.wikidata.org/wiki/Q1576430","display_name":"Bandwidth (computing)","level":2,"score":0.33239999413490295},{"id":"https://openalex.org/C124066611","wikidata":"https://www.wikidata.org/wiki/Q28684319","display_name":"Sparse approximation","level":2,"score":0.3140999972820282},{"id":"https://openalex.org/C138236772","wikidata":"https://www.wikidata.org/wiki/Q25098575","display_name":"Edge device","level":3,"score":0.287200003862381},{"id":"https://openalex.org/C118524514","wikidata":"https://www.wikidata.org/wiki/Q173212","display_name":"Computer architecture","level":1,"score":0.28679999709129333},{"id":"https://openalex.org/C123657996","wikidata":"https://www.wikidata.org/wiki/Q12271","display_name":"Architecture","level":2,"score":0.2854999899864197},{"id":"https://openalex.org/C46637626","wikidata":"https://www.wikidata.org/wiki/Q6693015","display_name":"Low latency (capital markets)","level":2,"score":0.25119999051094055}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1109/iccad66269.2025.11240713","is_oa":false,"landing_page_url":"https://doi.org/10.1109/iccad66269.2025.11240713","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2025 IEEE/ACM International Conference On Computer Aided Design (ICCAD)","raw_type":"proceedings-article"}],"best_oa_location":null,"sustainable_development_goals":[{"display_name":"Affordable and clean energy","score":0.8541390299797058,"id":"https://metadata.un.org/sdg/7"}],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":19,"referenced_works":["https://openalex.org/W2982008795","https://openalex.org/W3134526034","https://openalex.org/W3136346557","https://openalex.org/W4213060883","https://openalex.org/W4220943275","https://openalex.org/W4220972538","https://openalex.org/W4283830961","https://openalex.org/W4312847929","https://openalex.org/W4389518760","https://openalex.org/W4392427708","https://openalex.org/W4394998968","https://openalex.org/W4401211878","https://openalex.org/W4402671766","https://openalex.org/W4404411111","https://openalex.org/W4404955001","https://openalex.org/W4411485931","https://openalex.org/W4412944812","https://openalex.org/W4415689913","https://openalex.org/W4415795655"],"related_works":[],"abstract_inverted_index":{"Large":[0],"language":[1,14],"models":[2],"(LLMs)":[3],"have":[4],"demonstrated":[5],"remarkable":[6],"proficiency":[7],"in":[8],"a":[9,46,95,159,196],"wide":[10],"range":[11],"of":[12,66,201],"natural":[13],"processing":[15,52],"applications.":[16],"However,":[17],"the":[18,26,30,87,90,111,117,122,135,148,153,172],"high":[19,114],"energy":[20,188],"and":[21,59,102,129,147,170,184],"latency":[22],"overhead":[23],"induced":[24],"by":[25],"KV":[27],"cache":[28],"limits":[29],"edge":[31],"deployment,":[32],"especially":[33],"for":[34,82,105],"long":[35],"contexts.":[36],"Emerging":[37],"hybrid":[38,96,126],"bonding":[39],"(HB)":[40],"technology":[41],"has":[42],"been":[43],"proposed":[44],"as":[45],"promising":[47],"alternative":[48],"to":[49,108,124,133,166],"conventional":[50],"near-memory":[51],"(NMP)":[53],"architectures,":[54],"offering":[55],"improved":[56],"bandwidth":[57],"efficiency":[58,189],"lower":[60],"power":[61],"consumption":[62],"while":[63],"exhibiting":[64],"characteristics":[65],"distributed":[67,136],"memory.In":[68],"this":[69],"paper,":[70],"we":[71,93,120,156],"propose":[72,94,130],"H2EAL,":[73],"an":[74],"HB-based":[75],"accelerator":[76],"with":[77,100,113,162,195],"sparse":[78,97,127,145],"attention":[79,98,128,141,149,165],"algorithm-hardware":[80],"co-design":[81,121],"efficient":[83],"LLM":[84],"inference":[85],"at":[86],"edge.":[88],"At":[89,116],"algorithm":[91],"level,":[92,119],"scheme":[99],"static":[101],"dynamic":[103],"sparsity":[104,112],"different":[106,140,144],"heads":[107,142],"fully":[109],"leverage":[110],"accuracy.":[115],"hardware":[118,123],"support":[125],"memory-compute":[131],"co-placement":[132],"address":[134,167],"memory":[137],"bottleneck.":[138],"Since":[139],"exhibit":[143],"patterns":[146],"structure":[150],"often":[151],"mismatches":[152],"HB":[154,193],"architecture,":[155],"further":[157],"develop":[158],"load-balancing":[160],"scheduler":[161],"parallel":[163],"tiled":[164],"workload":[168],"imbalance":[169],"optimize":[171],"mapping":[173],"strategy.":[174],"Extensive":[175],"experiments":[176],"demonstrate":[177],"H2EAL":[178],"achieves":[179],"5.20":[180],"\u223c":[181,186],"48.21\u00d7":[182],"speedup":[183],"6.22":[185],"73.48\u00d7":[187],"improvement":[190],"over":[191],"baseline":[192],"implementation,":[194],"negligible":[197],"average":[198],"accuracy":[199],"drop":[200],"0.87%":[202],"on":[203],"multiple":[204],"benchmarks.":[205]},"counts_by_year":[{"year":2026,"cited_by_count":1}],"updated_date":"2026-07-29T09:40:50.615796","created_date":"2025-11-20T00:00:00"}