{"authors":[{"id":"orcid_______::389ace5ef15aa7a601c20e48aa2d2353","fullName":"Marian Verhelst","name":null,"surname":null,"rank":1,"pid":{"id":{"scheme":"orcid","value":"0000-0003-3495-9263"},"provenance":null}},{"id":"orcid_______::4e80f9dece9d037f7ebb614b49598800","fullName":"Luca Benini","name":null,"surname":null,"rank":2,"pid":{"id":{"scheme":"orcid","value":"0000-0001-8068-3806"},"provenance":null}},{"id":"orcid_______::e434e96550b998094d2533f752007843","fullName":"Naveen Verma","name":null,"surname":null,"rank":3,"pid":{"id":{"scheme":"orcid","value":"0000-0002-8208-5030"},"provenance":null}}],"openAccessColor":null,"publiclyFunded":false,"eoscIfGuidelines":null,"type":"publication","language":{"code":"und","label":"Undetermined"},"countries":[{"code":"IT","label":"Italy","provenance":null},{"code":"BE","label":"Belgium","provenance":null}],"subjects":[{"subject":{"scheme":"keyword","value":"FOS: Computer and information sciences"},"provenance":null},{"subject":{"scheme":"keyword","value":"Technology"},"provenance":null},{"subject":{"scheme":"keyword","value":"Electrical & Electronic Engineering"},"provenance":null},{"subject":{"scheme":"keyword","value":"Science & Technology"},"provenance":null},{"subject":{"scheme":"keyword","value":"Chip design"},"provenance":null},{"subject":{"scheme":"keyword","value":"roofline model"},"provenance":null},{"subject":{"scheme":"keyword","value":"Chip design; energy efficiency; machine learning (ML) accelerators; memory hierarchy; processor architectures; quantization; roofline model; sparsity; throughput"},"provenance":null},{"subject":{"scheme":"keyword","value":"processor architectures"},"provenance":null},{"subject":{"scheme":"keyword","value":"sparsity"},"provenance":null},{"subject":{"scheme":"keyword","value":"1099 Other Technology"},"provenance":null},{"subject":{"scheme":"keyword","value":"0204 Condensed Matter Physics"},"provenance":null},{"subject":{"scheme":"keyword","value":"Engineering, Electrical & Electronic"},"provenance":null},{"subject":{"scheme":"keyword","value":"memory hierarchy"},"provenance":null},{"subject":{"scheme":"FOS","value":"02 engineering and technology"},"provenance":null},{"subject":{"scheme":"keyword","value":"machine learning (ML) accelerators"},"provenance":null},{"subject":{"scheme":"keyword","value":"0906 Electrical and Electronic Engineering"},"provenance":null},{"subject":{"scheme":"keyword","value":"Engineering"},"provenance":null},{"subject":{"scheme":"keyword","value":"4009 Electronics, sensors and digital hardware"},"provenance":null},{"subject":{"scheme":"keyword","value":"Hardware Architecture (cs.AR)"},"provenance":null},{"subject":{"scheme":"FOS","value":"0202 electrical engineering, electronic engineering, information engineering"},"provenance":null},{"subject":{"scheme":"keyword","value":"quantization"},"provenance":null},{"subject":{"scheme":"keyword","value":"Computer Science - Hardware Architecture"},"provenance":null},{"subject":{"scheme":"keyword","value":"energy efficiency"},"provenance":null},{"subject":{"scheme":"keyword","value":"throughput"},"provenance":null}],"mainTitle":"How to Keep Pushing ML Accelerator Performance? Know Your Rooflines!","subTitle":null,"descriptions":["The rapidly growing importance of Machine Learning (ML) applications, coupled with their ever-increasing model size and inference energy footprint, has created a strong need for specialized ML hardware architectures. Numerous ML accelerators have been explored and implemented, primarily to increase task-level throughput per unit area and reduce task-level energy consumption. This paper surveys key trends toward these objectives for more efficient ML accelerators and provides a unifying framework to understand how compute and memory technologies/architectures interact to enhance system-level efficiency and performance. To achieve this, the paper introduces an enhanced version of the roofline model and applies it to ML accelerators as an effective tool for understanding where various execution regimes fall within roofline bounds and how to maximize performance and efficiency under the rooline. Key concepts are illustrated with examples from state-of-the-art designs, with a view towards open research opportunities to further advance accelerator performance.","in IEEE Journal of Solid-State Circuits, 2025"],"publicationDate":"2025-06-01","publisher":"Institute of Electrical and Electronics Engineers (IEEE)","embargoEndDate":"2025-05-01","sources":["Crossref","IEEE Journal of Solid-State Circuits"],"formats":null,"contributors":null,"coverages":null,"bestAccessRight":{"code":"c_abf2","label":"OPEN","scheme":"http://vocabularies.coar-repositories.org/documentation/access_rights/"},"container":{"name":"IEEE Journal of Solid-State Circuits","issnPrinted":"0018-9200","issnOnline":"1558-173X","issnLinking":null,"ep":"1905","iss":null,"sp":"1888","vol":"60","edition":null,"conferencePlace":null,"conferenceDate":null},"documentationUrls":null,"codeRepositoryUrl":null,"programmingLanguage":null,"contactPeople":null,"contactGroups":null,"tools":null,"size":null,"version":null,"geoLocations":null,"id":"doi_dedup___::b7b24e8176eeeec535f7a693c7bfadda","originalIds":["10.1109/jssc.2025.3553765","50|doiboost____|b7b24e8176eeeec535f7a693c7bfadda","50|datacite____::fdc65a93d6e0688c0a9495f55f560b3b","10.48550/arxiv.2505.16346","50|od________18::e0c594605e6c5e960cccad71c8730e96","oai:arXiv.org:2505.16346","50|dblp________::3d8fefa3d590ebf1fdc6f2a604d02dce","50|dblp________::a0d1dcc4b84a0f2f191b85c24bf661a4","oai:lirias2repo.kuleuven.be:20.500.12942/765416","50|od______1131::203b354f3ca1051bf45935c765bb831b","oai:cris.unibo.it:11585/1039394","50|od______4094::888faf6e34ee61fd296714f9877c52e9","10.1109/JSSC.2025.3553765","50|r3c4b2081b22::d00773f48464bcbbb526a0a9202dd857"],"pids":[{"scheme":"doi","value":"10.1109/jssc.2025.3553765"},{"scheme":"doi","value":"10.48550/arxiv.2505.16346"},{"scheme":"arXiv","value":"2505.16346"},{"scheme":"handle","value":"11585/1039394"}],"dateOfCollection":null,"lastUpdateTimeStamp":null,"indicators":{"citationImpact":{"citationCount":1.0,"influence":2.29471E-9,"popularity":3.0193352E-9,"impulse":1.0,"citationClass":"C5","influenceClass":"C5","impulseClass":"C5","popularityClass":"C5"}},"projects":[{"id":"corda_____he::e3602c1969e72fb08f6d6b1a25e881f9","code":"101088865","acronym":"BINGO","title":"Outplaying the hardware lottery for embedded AI","funder":"European Commission","pids":[{"scheme":"doi","value":"10.3030/101088865"}]},{"id":"corda_____he::bbf48969e885458782e7964603eb742d","code":"101070374","acronym":"CONVOLVE","title":"Seamless design of smart edge processors","funder":"European Commission","pids":[{"scheme":"doi","value":"10.3030/101070374"}]}],"organizations":[{"legalName":"College of New Jersey","acronym":"TCNJ","id":"openorgs____::95f2322932e814a1409c7b2de17f9bdf","pids":[{"scheme":"RRID","value":"RRID:SCR_001059"},{"scheme":"FundRef","value":"100006734"},{"scheme":"RRID","value":"RRID:nlx_16293"},{"scheme":"ISNI","value":"0000000404005239"},{"scheme":"FundRef","value":"100008141"},{"scheme":"FundRef","value":"100005589"},{"scheme":"wikidata","value":"Q7726772"},{"scheme":"fundref","value":"100006734"},{"scheme":"fundref","value":"100005664"},{"scheme":"fundref","value":"100005589"},{"scheme":"FundRef","value":"100014564"},{"scheme":"GRID","value":"grid.264500.5"},{"scheme":"fundref","value":"100006025"},{"scheme":"FundRef","value":"100005677"},{"scheme":"FundRef","value":"100007084"},{"scheme":"mag_id","value":"64281891"},{"scheme":"FundRef","value":"100007242"},{"scheme":"ROR","value":"https://ror.org/02nx5r318"},{"scheme":"FundRef","value":"100005664"},{"scheme":"fundref","value":"100014564"},{"scheme":"mag_id","value":"20089843"},{"scheme":"Wikidata","value":"Q7726772"},{"scheme":"ISNI","value":"0000000120975006"},{"scheme":"Wikidata","value":"Q21578"},{"scheme":"GRID","value":"grid.16750.35"},{"scheme":"fundref","value":"100005591"},{"scheme":"fundref","value":"100007242"},{"scheme":"fundref","value":"100007084"},{"scheme":"fundref","value":"100005677"},{"scheme":"wikidata","value":"Q21578"},{"scheme":"fundref","value":"100008141"},{"scheme":"OrgRef","value":"23922"},{"scheme":"OrgRef","value":"179979"},{"scheme":"OrgRef","value":"449646"},{"scheme":"FundRef","value":"100005591"},{"scheme":"FundRef","value":"100006025"},{"scheme":"ROR","value":"https://ror.org/00hx57361"}]},{"legalName":"KU Leuven","acronym":"KUL","id":"openorgs____::ccc0a066b56d2cfaf90c2ae369df16f5","pids":[{"scheme":"OrgReg","value":"BE0056"},{"scheme":"FundRef","value":"501100004040"},{"scheme":"RRID","value":"RRID:nlx_149306"},{"scheme":"GRID","value":"grid.5596.f"},{"scheme":"wikidata","value":"Q833670"},{"scheme":"ROR","value":"https://ror.org/05f950310"},{"scheme":"OrgRef","value":"3086876"},{"scheme":"fundref","value":"501100004040"},{"scheme":"FundRef","value":"501100004497"},{"scheme":"OrgRef","value":"25524686"},{"scheme":"fundref","value":"501100004497"},{"scheme":"PIC","value":"999991334"},{"scheme":"OrgRef","value":"29975400"},{"scheme":"ISNI","value":"0000000106687884"},{"scheme":"RRID","value":"RRID:SCR_001099"},{"scheme":"Wikidata","value":"Q833670"}]},{"legalName":"Alma Mater Studiorum University of Bologna","acronym":"UNIBO","id":"openorgs____::b5ca9d4340e26454e367e2908ef3872f","pids":[{"scheme":"mag_id","value":"9360294"},{"scheme":"OrgReg","value":"IT0009"},{"scheme":"ROR","value":"https://ror.org/01111rn36"},{"scheme":"wikidata","value":"Q131262"},{"scheme":"fundref","value":"501100005969"},{"scheme":"OrgRef","value":"263329"},{"scheme":"PIC","value":"999993953"},{"scheme":"ISNI","value":"0000000417571758"},{"scheme":"FundRef","value":"501100005969"},{"scheme":"Wikidata","value":"Q131262"},{"scheme":"GRID","value":"grid.6292.f"},{"scheme":"RRID","value":"RRID:SCR_011610"}]},{"legalName":"ETH Zurich","acronym":"EPFZ","id":"openorgs____::fb1e14f93f04d43e1a10a9f17d12c669","pids":[{"scheme":"wikidata","value":"Q11942"},{"scheme":"FundRef","value":"501100003070"},{"scheme":"OrgRef","value":"210910"},{"scheme":"ROR","value":"https://ror.org/05a28rw58"},{"scheme":"fundref","value":"501100003006"},{"scheme":"FundRef","value":"501100003006"},{"scheme":"FundRef","value":"501100001710"},{"scheme":"fundref","value":"501100001710"},{"scheme":"GRID","value":"grid.5801.c"},{"scheme":"PIC","value":"999979015"},{"scheme":"Wikidata","value":"Q11942"},{"scheme":"fundref","value":"501100003070"},{"scheme":"mag_id","value":"35440088"},{"scheme":"OrgReg","value":"CH0012"},{"scheme":"RRID","value":"RRID:SCR_000962"},{"scheme":"RRID","value":"RRID:nlx_143698"},{"scheme":"ISNI","value":"0000000121562780"}]}],"communities":[{"code":"eosc","label":"EOSC","provenance":null},{"code":"enermaps","label":"Energy Research","provenance":null}],"collectedFrom":[{"key":"openaire____::9e3be59865b2c1c335d32dae2fe7b254","value":"Datacite"},{"key":"openaire____::081b82f96300b6a6e3d282bad31cb6e2","value":"Crossref"},{"key":"openaire____::d8b68cc0a53121f6f896883a7d60c1db","value":"DBLP"},{"key":"opendoar____::fe709c654eac84d5239d1a12a4f71877","value":"Lirias"},{"key":"opendoar____::6f4922f45568161a8cdf4ad2299f6d23","value":"arXiv.org e-Print Archive"},{"key":"opendoar____::64c53a52cb3bd1a01c03a64db985c0cc","value":"Archivio istituzionale della ricerca - Alma Mater Studiorum Università di Bologna"},{"key":"re3data_____::c4b2081b224be6b3e79d0e5e5556f631","value":"European Union Open Data Portal"}],"instances":[{"pids":[{"scheme":"doi","value":"10.1109/jssc.2025.3553765"}],"license":"IEEE Copyright","accessRight":{"code":"c_14cb","label":"CLOSED","scheme":"http://vocabularies.coar-repositories.org/documentation/access_rights/","openAccessRoute":null},"type":"Article","urls":["https://doi.org/10.1109/jssc.2025.3553765"],"publicationDate":"2025-06-01","refereed":"peerReviewed","hostedBy":{"key":"issn___print::0a5632e7375662893438064d67886ea3","value":"IEEE Journal of Solid-State Circuits"},"collectedFrom":{"key":"openaire____::081b82f96300b6a6e3d282bad31cb6e2","value":"Crossref"}},{"pids":[{"scheme":"doi","value":"10.48550/arxiv.2505.16346"}],"license":"CC BY","type":"Article","urls":["https://dx.doi.org/10.48550/arxiv.2505.16346"],"publicationDate":"2025-01-01","refereed":"nonPeerReviewed","hostedBy":{"key":"openaire____::55045bd2a65019fd8e6741a755395c8c","value":"Unknown Repository"},"collectedFrom":{"key":"openaire____::9e3be59865b2c1c335d32dae2fe7b254","value":"Datacite"}},{"pids":[{"scheme":"arXiv","value":"2505.16346"}],"alternateIdentifiers":[{"scheme":"doi","value":"10.1109/jssc.2025.3553765"}],"accessRight":{"code":"c_abf2","label":"OPEN","scheme":"http://vocabularies.coar-repositories.org/documentation/access_rights/","openAccessRoute":null},"type":"Preprint","urls":["http://arxiv.org/abs/2505.16346"],"publicationDate":"2025-05-22","refereed":"nonPeerReviewed","hostedBy":{"key":"opendoar____::6f4922f45568161a8cdf4ad2299f6d23","value":"arXiv.org e-Print Archive"},"collectedFrom":{"key":"opendoar____::6f4922f45568161a8cdf4ad2299f6d23","value":"arXiv.org e-Print Archive"}},{"alternateIdentifiers":[{"scheme":"doi","value":"10.48550/arxiv.2505.16346"}],"type":"Preprint","urls":["https://doi.org/10.48550/arXiv.2505.16346"],"publicationDate":"2025-01-01","refereed":"nonPeerReviewed","hostedBy":{"key":"openaire____::d8b68cc0a53121f6f896883a7d60c1db","value":"DBLP"},"collectedFrom":{"key":"openaire____::d8b68cc0a53121f6f896883a7d60c1db","value":"DBLP"}},{"alternateIdentifiers":[{"scheme":"doi","value":"10.1109/jssc.2025.3553765"}],"type":"Article","urls":["https://doi.org/10.1109/JSSC.2025.3553765"],"publicationDate":"2025-01-01","refereed":"nonPeerReviewed","hostedBy":{"key":"openaire____::d8b68cc0a53121f6f896883a7d60c1db","value":"DBLP"},"collectedFrom":{"key":"openaire____::d8b68cc0a53121f6f896883a7d60c1db","value":"DBLP"}},{"alternateIdentifiers":[{"scheme":"doi","value":"10.1109/jssc.2025.3553765"}],"accessRight":{"code":"c_abf2","label":"OPEN","scheme":"http://vocabularies.coar-repositories.org/documentation/access_rights/","openAccessRoute":null},"type":"Article","urls":["https://doi.org/10.1109/jssc.2025.3553765","https://lirias.kuleuven.be/handle/20.500.12942/765416"],"publicationDate":"2025-06-01","refereed":"nonPeerReviewed","hostedBy":{"key":"opendoar____::fe709c654eac84d5239d1a12a4f71877","value":"Lirias"},"collectedFrom":{"key":"opendoar____::fe709c654eac84d5239d1a12a4f71877","value":"Lirias"}},{"pids":[{"scheme":"handle","value":"11585/1039394"}],"alternateIdentifiers":[{"scheme":"urn","value":"2-s2.0-105003373516"},{"scheme":"doi","value":"10.1109/jssc.2025.3553765"},{"scheme":"urn","value":"W4409641128"}],"accessRight":{"code":"c_14cb","label":"CLOSED","scheme":"http://vocabularies.coar-repositories.org/documentation/access_rights/","openAccessRoute":null},"type":"Article","urls":["https://doi.org/10.1109/jssc.2025.3553765","https://hdl.handle.net/11585/1039394"],"publicationDate":"2025-01-01","refereed":"nonPeerReviewed","hostedBy":{"key":"opendoar____::64c53a52cb3bd1a01c03a64db985c0cc","value":"Archivio istituzionale della ricerca - Alma Mater Studiorum Università di Bologna"},"collectedFrom":{"key":"opendoar____::64c53a52cb3bd1a01c03a64db985c0cc","value":"Archivio istituzionale della ricerca - Alma Mater Studiorum Università di Bologna"}},{"alternateIdentifiers":[{"scheme":"doi","value":"10.1109/jssc.2025.3553765"}],"type":"Article","urls":["http://dx.doi.org/10.1109/JSSC.2025.3553765"],"publicationDate":"2025-01-01","refereed":"peerReviewed","hostedBy":{"key":"issn___print::0a5632e7375662893438064d67886ea3","value":"IEEE Journal of Solid-State Circuits"},"collectedFrom":{"key":"re3data_____::c4b2081b224be6b3e79d0e5e5556f631","value":"European Union Open Data Portal"}}],"links":[{"header":{"relationType":"resultProject","relationClass":"isProducedBy","relatedIdentifier":"corda_____he::e3602c1969e72fb08f6d6b1a25e881f9","relatedRecordType":"project","relationProvenance":"sysimport:crosswalk:repository","trust":"0.9"},"collectedfrom":[{"dsId":"openaire____::3f264f93cf3b0cfc4ede188a6300455c","dsName":"CORDA - COmmon Research DAta Warehouse - Horizon Europe"}],"projectTitle":"Outplaying the hardware lottery for embedded AI","code":"101088865","funding":{"funder":{"id":"ec__________::EC","shortname":"EC","name":"European Commission","jurisdiction":{"code":"EU","label":"European Union"},"pid":null},"level0":{"id":"ec__________::EC::HE","description":"Horizon Europe Framework Programme","name":"HE"},"level1":{"id":"ec__________::EC::HE::ERC","description":"European Research Council","name":"ERC"},"level2":{"id":"ec__________::EC::HE::ERC::HORIZON-ERC","description":"HORIZON ERC Grants","name":"HORIZON-ERC"}},"startDate":"2023-06-01","endDate":"2028-05-31"},{"header":{"relationType":"resultProject","relationClass":"isProducedBy","relatedIdentifier":"corda_____he::bbf48969e885458782e7964603eb742d","relatedRecordType":"project","relationProvenance":"sysimport:crosswalk:repository","trust":"0.9"},"collectedfrom":[{"dsId":"openaire____::3f264f93cf3b0cfc4ede188a6300455c","dsName":"CORDA - COmmon Research DAta Warehouse - Horizon Europe"}],"projectTitle":"Seamless design of smart edge processors","code":"101070374","funding":{"funder":{"id":"ec__________::EC","shortname":"EC","name":"European Commission","jurisdiction":{"code":"EU","label":"European Union"},"pid":null},"level0":{"id":"ec__________::EC::HE","description":"Horizon Europe Framework Programme","name":"HE"},"level1":{"id":"ec__________::EC::HE::HORIZON-RIA","description":"HORIZON  Research and Innovation Actions","name":"HORIZON-RIA"},"level2":{"id":null,"description":null,"name":null}},"startDate":"2022-11-01","endDate":"2026-04-30"}],"otherTitles":null,"inDiamondJournal":false,"green":true,"isGreen":true,"isInDiamondJournal":false}