{"id":"https://openalex.org/W7163527459","doi":"https://doi.org/10.23919/date69613.2026.11539471","title":"Efficient LLM Decoding on Ryzen AI NPUs","display_name":"Efficient LLM Decoding on Ryzen AI NPUs","publication_year":2026,"publication_date":"2026-04-20","ids":{"openalex":"https://openalex.org/W7163527459","doi":"https://doi.org/10.23919/date69613.2026.11539471"},"language":null,"primary_location":{"id":"doi:10.23919/date69613.2026.11539471","is_oa":false,"landing_page_url":"https://doi.org/10.23919/date69613.2026.11539471","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2026 Design, Automation & 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/A5137857736","display_name":"Zhenyu Xu","orcid":null},"institutions":[{"id":"https://openalex.org/I8078737","display_name":"Clemson University","ror":"https://ror.org/037s24f05","country_code":"US","type":"education","lineage":["https://openalex.org/I8078737"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Zhenyu Xu","raw_affiliation_strings":["Clemson University,Electrical and Computer Engineering,Clemson,USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Clemson University,Electrical and Computer Engineering,Clemson,USA","institution_ids":["https://openalex.org/I8078737"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5013631588","display_name":"Miaoxiang Yu","orcid":"https://orcid.org/0000-0002-4382-9009"},"institutions":[{"id":"https://openalex.org/I8078737","display_name":"Clemson University","ror":"https://ror.org/037s24f05","country_code":"US","type":"education","lineage":["https://openalex.org/I8078737"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Miaoxiang Yu","raw_affiliation_strings":["Clemson University,Electrical and Computer Engineering,Clemson,USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Clemson University,Electrical and Computer Engineering,Clemson,USA","institution_ids":["https://openalex.org/I8078737"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5020091845","display_name":"Jillian Cai","orcid":"https://orcid.org/0009-0001-5006-223X"},"institutions":[{"id":"https://openalex.org/I8078737","display_name":"Clemson University","ror":"https://ror.org/037s24f05","country_code":"US","type":"education","lineage":["https://openalex.org/I8078737"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Jillian Cai","raw_affiliation_strings":["Clemson University,Electrical and Computer Engineering,Clemson,USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Clemson University,Electrical and Computer Engineering,Clemson,USA","institution_ids":["https://openalex.org/I8078737"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137875488","display_name":"Qing Yang","orcid":null},"institutions":[{"id":"https://openalex.org/I17626003","display_name":"University of Rhode Island","ror":"https://ror.org/013ckk937","country_code":"US","type":"education","lineage":["https://openalex.org/I17626003"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Qing Yang","raw_affiliation_strings":["University of Rhode Island,Electrical and Computer Engineering,Kingston,USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Rhode Island,Electrical and Computer Engineering,Kingston,USA","institution_ids":["https://openalex.org/I17626003"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5107952348","display_name":"Tao Wei","orcid":"https://orcid.org/0009-0000-4027-0310"},"institutions":[{"id":"https://openalex.org/I8078737","display_name":"Clemson University","ror":"https://ror.org/037s24f05","country_code":"US","type":"education","lineage":["https://openalex.org/I8078737"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Tao Wei","raw_affiliation_strings":["Clemson University,Electrical and Computer Engineering,Clemson,USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Clemson University,Electrical and Computer Engineering,Clemson,USA","institution_ids":["https://openalex.org/I8078737"]}]}],"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":0.62594425,"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/T10237","display_name":"Cryptography and Data Security","score":0.07460000365972519,"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"}},"topics":[{"id":"https://openalex.org/T10237","display_name":"Cryptography and Data Security","score":0.07460000365972519,"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/T11321","display_name":"Error Correcting Code Techniques","score":0.050200000405311584,"subfield":{"id":"https://openalex.org/subfields/1705","display_name":"Computer Networks and Communications"},"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/T10320","display_name":"Neural Networks and Applications","score":0.04820000007748604,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/scalability","display_name":"Scalability","score":0.7791000008583069},{"id":"https://openalex.org/keywords/speedup","display_name":"Speedup","score":0.763700008392334},{"id":"https://openalex.org/keywords/robustness","display_name":"Robustness (evolution)","score":0.7062000036239624},{"id":"https://openalex.org/keywords/decoding-methods","display_name":"Decoding methods","score":0.6884999871253967},{"id":"https://openalex.org/keywords/benchmark","display_name":"Benchmark (surveying)","score":0.6302000284194946},{"id":"https://openalex.org/keywords/cache","display_name":"Cache","score":0.6047000288963318},{"id":"https://openalex.org/keywords/efficient-energy-use","display_name":"Efficient energy use","score":0.5375999808311462},{"id":"https://openalex.org/keywords/central-processing-unit","display_name":"Central processing unit","score":0.4805999994277954}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7918000221252441},{"id":"https://openalex.org/C48044578","wikidata":"https://www.wikidata.org/wiki/Q727490","display_name":"Scalability","level":2,"score":0.7791000008583069},{"id":"https://openalex.org/C68339613","wikidata":"https://www.wikidata.org/wiki/Q1549489","display_name":"Speedup","level":2,"score":0.763700008392334},{"id":"https://openalex.org/C63479239","wikidata":"https://www.wikidata.org/wiki/Q7353546","display_name":"Robustness (evolution)","level":3,"score":0.7062000036239624},{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.6884999871253967},{"id":"https://openalex.org/C185798385","wikidata":"https://www.wikidata.org/wiki/Q1161707","display_name":"Benchmark (surveying)","level":2,"score":0.6302000284194946},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.6256999969482422},{"id":"https://openalex.org/C115537543","wikidata":"https://www.wikidata.org/wiki/Q165596","display_name":"Cache","level":2,"score":0.6047000288963318},{"id":"https://openalex.org/C2742236","wikidata":"https://www.wikidata.org/wiki/Q924713","display_name":"Efficient energy use","level":2,"score":0.5375999808311462},{"id":"https://openalex.org/C49154492","wikidata":"https://www.wikidata.org/wiki/Q5300","display_name":"Central processing unit","level":2,"score":0.4805999994277954},{"id":"https://openalex.org/C2780513914","wikidata":"https://www.wikidata.org/wiki/Q18210350","display_name":"Bottleneck","level":2,"score":0.446399986743927},{"id":"https://openalex.org/C189783530","wikidata":"https://www.wikidata.org/wiki/Q352090","display_name":"CPU cache","level":3,"score":0.38999998569488525},{"id":"https://openalex.org/C113775141","wikidata":"https://www.wikidata.org/wiki/Q428691","display_name":"Computer engineering","level":1,"score":0.382099986076355},{"id":"https://openalex.org/C149635348","wikidata":"https://www.wikidata.org/wiki/Q193040","display_name":"Embedded system","level":1,"score":0.35429999232292175},{"id":"https://openalex.org/C9390403","wikidata":"https://www.wikidata.org/wiki/Q3966","display_name":"Computer hardware","level":1,"score":0.34950000047683716},{"id":"https://openalex.org/C79403827","wikidata":"https://www.wikidata.org/wiki/Q3988","display_name":"Real-time computing","level":1,"score":0.3483999967575073},{"id":"https://openalex.org/C186370098","wikidata":"https://www.wikidata.org/wiki/Q442787","display_name":"Energy (signal processing)","level":2,"score":0.3188999891281128},{"id":"https://openalex.org/C3017489831","wikidata":"https://www.wikidata.org/wiki/Q2393193","display_name":"Running time","level":2,"score":0.3111000061035156},{"id":"https://openalex.org/C2989134064","wikidata":"https://www.wikidata.org/wiki/Q288510","display_name":"Execution time","level":2,"score":0.3077999949455261},{"id":"https://openalex.org/C163258240","wikidata":"https://www.wikidata.org/wiki/Q25342","display_name":"Power (physics)","level":2,"score":0.3068000078201294},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.3009999990463257},{"id":"https://openalex.org/C46686674","wikidata":"https://www.wikidata.org/wiki/Q466303","display_name":"Boosting (machine learning)","level":2,"score":0.2806999981403351},{"id":"https://openalex.org/C118993495","wikidata":"https://www.wikidata.org/wiki/Q5042828","display_name":"Electrical efficiency","level":3,"score":0.27410000562667847},{"id":"https://openalex.org/C57493831","wikidata":"https://www.wikidata.org/wiki/Q3134666","display_name":"Projection (relational algebra)","level":2,"score":0.27309998869895935},{"id":"https://openalex.org/C82876162","wikidata":"https://www.wikidata.org/wiki/Q17096504","display_name":"Latency (audio)","level":2,"score":0.25850000977516174},{"id":"https://openalex.org/C2778915421","wikidata":"https://www.wikidata.org/wiki/Q3643177","display_name":"Performance improvement","level":2,"score":0.25279998779296875},{"id":"https://openalex.org/C2984118289","wikidata":"https://www.wikidata.org/wiki/Q29954","display_name":"Power consumption","level":3,"score":0.25130000710487366}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.23919/date69613.2026.11539471","is_oa":false,"landing_page_url":"https://doi.org/10.23919/date69613.2026.11539471","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2026 Design, Automation & Test in Europe Conference (DATE)","raw_type":"proceedings-article"}],"best_oa_location":null,"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":6,"referenced_works":["https://openalex.org/W4407196790","https://openalex.org/W4407953488","https://openalex.org/W4407953492","https://openalex.org/W4410810249","https://openalex.org/W4410810610","https://openalex.org/W7133227460"],"related_works":[],"abstract_inverted_index":{"We":[0,100],"propose":[1],"an":[2],"efficient":[3],"and":[4,20,31,57,71,89,97,105,115,120,132,136],"scalable":[5],"LLM":[6],"decoding":[7],"framework":[8],"optimized":[9],"for":[10,40,134],"AMD":[11],"Ryzen":[12],"AI":[13],"NPUs,":[14],"leveraging":[15],"two":[16],"novel":[17],"techniques:":[18],"FusedDQP":[19,22],"FlowKV.":[21],"fuses":[23],"dequantization":[24],"with":[25,86],"projection":[26],"to":[27,68,77,118,130],"minimize":[28],"memory":[29],"operations":[30],"latency,":[32],"while":[33,123],"FlowKV":[34],"introduces":[35],"a":[36],"pipelined,":[37],"bandwidth-optimized":[38],"approach":[39],"KV":[41],"cache":[42],"access":[43],"across":[44,91],"compute":[45],"tiles":[46],"(CT).":[47],"Together,":[48],"these":[49],"methods":[50],"deliver":[51],"substantial":[52],"improvements":[53],"in":[54],"both":[55],"speed":[56],"energy":[58,127],"efficiency":[59,74,128],"without":[60],"altering":[61],"model":[62,93],"accuracy.":[63],"Our":[64],"solution":[65],"achieves":[66],"up":[67],"14.2\u00d7":[69],"speedup":[70],"2.66\u00d7":[72],"power":[73],"gains":[75],"compared":[76],"existing":[78],"state-of-the-art":[79],"(SOTA)":[80],"NPU":[81],"baselines,":[82],"demonstrating":[83],"linear":[84],"scalability":[85],"CT":[87],"count":[88],"robustness":[90],"LLaMA-3.1/3.2":[92],"variants":[94],"(1B,":[95],"3B,":[96],"8B":[98],"parameters).":[99],"also":[101],"benchmark":[102],"against":[103],"CPU":[104,114,135],"iGPU":[106,116],"on":[107],"the":[108],"same":[109],"platform,":[110],"our":[111],"performance":[112],"surpasses":[113],"(up":[117,129],"1.8x":[119],"16.2x":[121],"speedup),":[122],"delivering":[124],"substantially":[125],"improved":[126],"3.63x":[131],"11.38x":[133],"iGPU,":[137],"respectively).":[138]},"counts_by_year":[],"updated_date":"2026-07-29T14:22:42.915294","created_date":"2026-06-05T00:00:00"}