{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,8,5]],"date-time":"2026-08-05T06:46:30Z","timestamp":1785912390567,"version":"3.56.0"},"reference-count":81,"publisher":"IEEE","license":[{"start":{"date-parts":[[2026,6,1]],"date-time":"2026-06-01T00:00:00Z","timestamp":1780272000000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/ieeexplore.ieee.org\/Xplorehelp\/downloads\/license-information\/IEEE.html"},{"start":{"date-parts":[[2026,6,1]],"date-time":"2026-06-01T00:00:00Z","timestamp":1780272000000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-029"},{"start":{"date-parts":[[2026,6,1]],"date-time":"2026-06-01T00:00:00Z","timestamp":1780272000000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-037"}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":[],"published-print":{"date-parts":[[2026,6]]},"DOI":"10.1109\/isca66397.2026.00054","type":"proceedings-article","created":{"date-parts":[[2026,8,4]],"date-time":"2026-08-04T19:13:44Z","timestamp":1785870824000},"page":"588-603","source":"Crossref","is-referenced-by-count":0,"title":["HybridSpec: Exploiting Hybrid-Bonding Memory to Accelerate LLM Serving Through Heterogeneous Architecture and Speculative Decoding"],"prefix":"10.1109","author":[{"given":"Zongle","family":"Huang","sequence":"first","affiliation":[{"name":"Tsinghua University","department":["Department of Electronic Engineering"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Wenbin","family":"Jia","sequence":"additional","affiliation":[{"name":"Tsinghua University","department":["Department of Electronic Engineering"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Yaolei","family":"Li","sequence":"additional","affiliation":[{"name":"Tsinghua University","department":["Department of Electronic Engineering"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Xinyuan","family":"Lin","sequence":"additional","affiliation":[{"name":"Tsinghua University","department":["Department of Electronic Engineering"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Shupei","family":"Fan","sequence":"additional","affiliation":[{"name":"Tsinghua University","department":["Department of Electronic Engineering"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Chen","family":"Tang","sequence":"additional","affiliation":[{"name":"Tsinghua University","department":["Department of Electronic Engineering"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Shuwen","family":"Deng","sequence":"additional","affiliation":[{"name":"Tsinghua University","department":["Department of Electronic Engineering"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Yongpan","family":"Liu","sequence":"additional","affiliation":[{"name":"Tsinghua University","department":["Department of Electronic Engineering"]}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"263","reference":[{"key":"ref1","article-title":"Sarathi: Efficient 1lm inference by piggybacking decodes with chunked prefills","author":"Agrawal","year":"2023","journal-title":"arXiv preprint"},{"key":"ref2","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2023.emnlp-main.298"},{"key":"ref3","article-title":"Medusa: Simple llm inference acceleration framework with multiple decoding heads","author":"Cai","year":"2024","journal-title":"arXiv preprint"},{"key":"ref4","volume-title":"Internlm2 technical report","author":"Cai","year":"2024"},{"key":"ref5","volume-title":"Transformer inference arithmetic","author":"Chen","year":"2022"},{"key":"ref6","article-title":"Accelerating large language model decoding with speculative sampling","author":"Chen","year":"2023","journal-title":"arXiv preprint"},{"key":"ref7","doi-asserted-by":"publisher","DOI":"10.1109\/MICRO.2014.58"},{"key":"ref8","doi-asserted-by":"publisher","DOI":"10.1109\/EPTC.2015.7412372"},{"key":"ref9","doi-asserted-by":"publisher","DOI":"10.1109\/EPTC.2009.5416582"},{"key":"ref10","volume-title":"Nvidia a100 datasheet","year":"2020"},{"key":"ref11","volume-title":"Nvidia h100 tensor core gpu","year":"2024"},{"key":"ref12","volume-title":"Deepseek-v3 technical report","author":"DeepSeek-AI","year":"2025"},{"key":"ref13","volume-title":"A peek inside apple\u2019s a4 processor","author":"Djuric","year":"2010"},{"key":"ref14","volume-title":"Hybrid bonding process flow","author":"Patel","year":"2024"},{"key":"ref15","volume-title":"opt-2.7b","year":"2020"},{"key":"ref16","volume-title":"opt-66b","year":"2020"},{"key":"ref17","doi-asserted-by":"publisher","DOI":"10.1109\/IEDM13553.2020.9372039"},{"key":"ref18","doi-asserted-by":"publisher","DOI":"10.1109\/ISSCC19947.2020.9062957"},{"key":"ref19","doi-asserted-by":"publisher","DOI":"10.1109\/ISSCC.2017.7870268"},{"key":"ref20","doi-asserted-by":"publisher","DOI":"10.1109\/ICCAD66269.2025.11240889"},{"key":"ref21","doi-asserted-by":"publisher","DOI":"10.52202\/079017-2181"},{"key":"ref22","doi-asserted-by":"publisher","DOI":"10.1145\/3676641.3716009"},{"key":"ref23","doi-asserted-by":"publisher","DOI":"10.1145\/3620666.3651380"},{"key":"ref24","volume-title":"Moesd: Unveil speculative decoding\u2019s potential for accelerating sparse moe","author":"Huang","year":"2025"},{"key":"ref25","doi-asserted-by":"publisher","DOI":"10.1109\/MSSC.2015.2474235"},{"key":"ref26","doi-asserted-by":"publisher","DOI":"10.3390\/mi15050557"},{"key":"ref27","doi-asserted-by":"publisher","DOI":"10.1109\/LSSC.2022.3171862"},{"key":"ref28","volume-title":"Fiddler: Cpu-gpu orchestration for fast inference of mixture-of-experts models","author":"Kamahori","year":"2025"},{"key":"ref29","volume-title":"ktransformers: A flexible framework for experiencing cutting-edge 1lm inference optimizations","year":"2024"},{"key":"ref30","doi-asserted-by":"publisher","DOI":"10.1145\/3600006.3613165"},{"key":"ref31","doi-asserted-by":"publisher","DOI":"10.1145\/3600006.3613165"},{"key":"ref32","doi-asserted-by":"publisher","DOI":"10.1109\/ISSCC49657.2024.10454440"},{"key":"ref33","first-page":"19274","article-title":"Fast inference from transformers via speculative decoding","volume-title":"International Conference on Machine Learning. PMLR","author":"Leviathan","year":"2023"},{"key":"ref34","doi-asserted-by":"publisher","DOI":"10.1109\/tpds.2019.2928289"},{"key":"ref35","doi-asserted-by":"publisher","DOI":"10.1145\/3695053.3731008"},{"key":"ref36","doi-asserted-by":"publisher","DOI":"10.1145\/3620666.3651352"},{"key":"ref37","volume-title":"Eagle-2: Faster inference of language models with dynamic draft trees","author":"Li","year":"2024"},{"key":"ref38","article-title":"Eagle: Speculative sampling requires rethinking feature uncertainty","author":"Li","year":"2024","journal-title":"arXiv preprint"},{"key":"ref39","article-title":"Eagle-3: Scaling up inference acceleration of large language models via training-time test","author":"Li","year":"2025","journal-title":"arXiv preprint"},{"key":"ref40","volume-title":"Enhancing memory efficiency in large language model training through chronos-aware pipeline parallelism","author":"Lin","year":"2025"},{"key":"ref41","article-title":"Optimizing speculative decoding for serving large language models using goodput","author":"Liu","year":"2024","journal-title":"arXiv preprint"},{"key":"ref42","doi-asserted-by":"publisher","DOI":"10.23919\/ICEP55381.2022.9795476"},{"key":"ref43","article-title":"Duodecoding: Hardware-aware heterogeneous speculative decoding with dynamic multi-sequence drafting","author":"Lv","year":"2025","journal-title":"arXiv preprint"},{"key":"ref44","doi-asserted-by":"publisher","DOI":"10.1109\/ECTC.2016.201"},{"key":"ref45","volume-title":"Dynamic speculation lookahead accelerates speculative decoding of large language models","author":"Mamou","year":"2024"},{"key":"ref46","doi-asserted-by":"publisher","DOI":"10.1145\/3669940.3707215"},{"key":"ref47","volume-title":"Llama-2\u201313b-chat-hf","year":"2023"},{"key":"ref48","doi-asserted-by":"publisher","DOI":"10.1145\/3620666.3651335"},{"key":"ref49","volume-title":"He who can pay top dollar for hbm memory controls ai training","author":"Morgan","year":"2024"},{"key":"ref50","doi-asserted-by":"publisher","DOI":"10.1109\/ISSCC42614.2022.9731694"},{"key":"ref51","doi-asserted-by":"publisher","DOI":"10.1109\/ISSCC19947.2020.9063110"},{"key":"ref52","doi-asserted-by":"publisher","DOI":"10.1109\/ISPASS.2019.00042"},{"key":"ref53","doi-asserted-by":"publisher","DOI":"10.1109\/JSSC.2022.3193354"},{"key":"ref54","doi-asserted-by":"publisher","DOI":"10.1109\/HPCA57654.2024.00078"},{"key":"ref55","doi-asserted-by":"publisher","DOI":"10.1109\/ISCA59077.2024.00019"},{"key":"ref56","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2024.findings-acl.489"},{"key":"ref57","article-title":"Mooncake: A kvcache-centric disaggregated architecture for 1lm serving","author":"Qin","year":"2024","journal-title":"arXiv preprint"},{"key":"ref58","volume-title":"Qwen3\u20131.7b","year":"2025"},{"key":"ref59","article-title":"Qwen3\u201332b","year":"2025","journal-title":"Qwen"},{"key":"ref60","doi-asserted-by":"publisher","DOI":"10.1145\/3620666.3651324"},{"key":"ref61","doi-asserted-by":"publisher","DOI":"10.1145\/3352460.3358302"},{"key":"ref62","article-title":"Megatron-Im: Training multi-billion parameter language models using model parallelism","author":"Shoeybi","year":"2019","journal-title":"arXiv preprint"},{"key":"ref63","doi-asserted-by":"publisher","DOI":"10.1109\/HPCA61900.2025.00128"},{"key":"ref64","doi-asserted-by":"publisher","DOI":"10.52202\/079017-0522"},{"key":"ref65","volume-title":"TinyLlama","year":"2024"},{"key":"ref66","doi-asserted-by":"publisher","DOI":"10.1109\/IEDM.2017.8268317"},{"key":"ref67","doi-asserted-by":"publisher","DOI":"10.48550\/ARXIV.1706.03762"},{"key":"ref68","volume-title":"benchmark_serving.py","year":"2023"},{"key":"ref69","doi-asserted-by":"publisher","DOI":"10.23919\/VLSITechnologyandCir57934.2023.10185427"},{"key":"ref70","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2024.findings-acl.456"},{"key":"ref71","volume-title":"Per-sequence speculative decoding","year":"2025"},{"key":"ref72","doi-asserted-by":"publisher","DOI":"10.1109\/ECTC.2006.1645753"},{"key":"ref73","first-page":"521","article-title":"Orca: A distributed serving system for Transformer-Based generative models","volume-title":"16th USENIX Symposium on Operating Systems Design and Implementation (OSDI 22)","author":"Yu","year":"2022"},{"key":"ref74","volume-title":"Llm inference unveiled: Survey and roofline model insights","author":"Yuan","year":"2024"},{"key":"ref75","volume-title":"Llm inference unveiled: Survey and roofline model insights","author":"Yuan","year":"2024"},{"key":"ref76","doi-asserted-by":"publisher","DOI":"10.1109\/ISCA59077.2024.00037"},{"key":"ref77","article-title":"Fastdraft: How to train your draft","author":"Zafrir","year":"2024","journal-title":"arXiv preprint"},{"key":"ref78","doi-asserted-by":"publisher","DOI":"10.1109\/DAC63849.2025.11132883"},{"key":"ref79","first-page":"193","article-title":"\\{DistServe\\}: Disaggregating prefill and decoding for goodput-optimized large language model serving","volume-title":"18th USENIX Symposium on Operating Systems Design and Implementation (OSDI 24)","author":"Zhong","year":"2024"},{"key":"ref80","volume-title":"Nanoflow: Towards optimal large language model serving throughput","author":"Zhu","year":"2025"},{"key":"ref81","first-page":"749","article-title":"NanoFlow: Towards optimal large language model serving throughput","volume-title":"19th USENIX Symposium on Operating Systems Design and Implementation (OSDI 25)","author":"Zhu"}],"event":{"name":"2026 ACM\/IEEE 53rd Annual International Symposium on Computer Architecture (ISCA)","location":"Raleigh, NC, USA","start":{"date-parts":[[2026,6,27]]},"end":{"date-parts":[[2026,7,1]]}},"container-title":["2026 ACM\/IEEE 53rd Annual International Symposium on Computer Architecture (ISCA)"],"original-title":[],"link":[{"URL":"http:\/\/xplorestaging.ieee.org\/ielx8\/11617393\/11617394\/11617519.pdf?arnumber=11617519","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,8,5]],"date-time":"2026-08-05T05:59:50Z","timestamp":1785909590000},"score":1,"resource":{"primary":{"URL":"https:\/\/ieeexplore.ieee.org\/document\/11617519\/"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,6]]},"references-count":81,"URL":"https:\/\/doi.org\/10.1109\/isca66397.2026.00054","relation":{},"subject":[],"published":{"date-parts":[[2026,6]]}}}