BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/Los_Angeles
X-LIC-LOCATION:America/Los_Angeles
BEGIN:DAYLIGHT
TZOFFSETFROM:-0800
TZOFFSETTO:-0700
TZNAME:PDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0700
TZOFFSETTO:-0800
TZNAME:PST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260730T152638Z
LOCATION:
DTSTART;TZID=America/Los_Angeles:20260728T125400
DTEND;TZID=America/Los_Angeles:20260728T125400
UID:dac_DAC 2026_sess315_RESEARCH1570@linklings.com
SUMMARY:NVLLM: A 3D NAND-Centric Architecture Enabling Edge on-Device LLM 
 Inference
DESCRIPTION:Mingbo Hao, Changwei Yan, Haoyu Cui, Zhihao Yan, Yizhi Ding, Z
 hangrui Qian, and Weiwei Shan (Southeast University)\n\nThe rapid growth o
 f LLMs demands high-throughput, memory-capacity–intensive inference on res
 ource-constrained edge devices, where single-batch decoding remains fundam
 entally memory-bound. Existing out-of-core GPUs and SSD-like accelerators 
 are limited by DRAM-bound weight movement and inefficient storage-access g
 ranularity. We present NVLLM, a 3D NAND-centric inference architecture tha
 t offloads FFN computation into the Flash array while executing attention 
 on lightweight CMOS logic with external DRAM. Through wafer-to-wafer stack
 ing, NVLLM tightly integrates multi-plane 3D NAND with compute pipelines, 
 ECC units, and buffers, enabling page-level FFN weight access without DRAM
  traversal. All GEMM/GEMV operations are decomposed into dot-product primi
 tives executed by out-of-order PE lanes, operating directly on raw NAND re
 ads with integrated ECC. Attention weights remain in DRAM, and a KV-cache–
 aware scheduler maintains throughput as context grows. Evaluated on OPT an
 d LLaMA models up to 30B parameters, NVLLM achieves 16.7× speedup over A80
 0 out-of-core inference and 1.1×–4.7× improvement over SSD-like designs, w
 ith only 2.7\% CMOS area overhead.\n\nTopics: Quantum\n\nTrack: DES6. Quan
 tum Computing, Systems\n\nSession Chairs: Hokeun Kim (Arizona State Univer
 sity); BaekGyu Kim (Daegu Gyeongbuk Institute of Science and Technology); 
 Xiang Chen (Peking University); Sangyeob Kim (Yonsei University); Robert M
 . Radway (University of Pennsylvania); Prashanth Mohan (University of Pitt
 sburgh); Xiaoxuan Yang (University of Virginia, Stanford University); Ramt
 in Zand (University of South Carolina); Hadi Kamali (University of Central
  Florida); Jiafeng Xie (Villanova University); Prabuddha Chakraborty (Univ
 ersity of Maine); Junichiro Kadomoto (University of Tokyo); and Yosuke Uen
 o (RIKEN)\n\n
END:VEVENT
END:VCALENDAR