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:20260402T024533Z
LOCATION:3000\, Level 3
DTSTART;TZID=America/Los_Angeles:20250625T110000
DTEND;TZID=America/Los_Angeles:20250625T111500
UID:dac_DAC 2025_sess106_RESEARCH021@linklings.com
SUMMARY:MOSS: Multi-Modal Representation Learning on Sequential Circuits
DESCRIPTION:Mingjun Wang and Bin Sun (State Key Lab of Processors, Institu
 te of Computing Technology, Chinese Academy of Sciences); Jianan Mu (Insti
 tute of Computing Technology, Chinese Academy of Sciences); Feng Gu (State
  Key Lab of Processors, Institute of Computing Technology, Chinese Academy
  of Sciences); Boyu Han (Stanford University); Tianmeng Yang (Peking Unive
 rsity); Xinyu Zhang and Silin Liu (Institute of Computing Technology, Chin
 ese Academy of Sciences); Yihan Wen (Beijing University of Technology); Hu
 i Wang (CASTEST, Beijing); Gao Jun (Institute of Computing Technology, Chi
 nese Academy of Sciences); Zhiteng Chao (State Key Lab of Processors, Inst
 itute of Computing Technology, Chinese Academy of Sciences); Husheng Han a
 nd Zizhen Liu (Institute of Computing Technology, Chinese Academy of Scien
 ces); Shengwen Liang (State Key Lab of Processors, Institute of Computing 
 Technology, Chinese Academy of Sciences); Jing Ye (State Key Laboratory of
  Computer Architecture, Institute of Computing Technology, Chinese Academy
  of Sciences); Bei Yu (The Chinese University of Hong Kong); and Xiaowei L
 i and Huawei Li (Institute of Computing Technology, Chinese Academy of Sci
 ences)\n\nDeep learning has significantly advanced Electronic Design Autom
 ation (EDA), with circuit representation learning emerging as a key area f
 or modeling the relationship between a circuit's structure and functionali
 ty. Existing methods primarily use either Large Language Models (LLMs) for
  Register Transfer Level (RTL) code analysis or Graph Neural Networks (GNN
 s) for netlist modeling. While LLMs excel at high-level functional underst
 anding, they struggle with detailed netlist behavior. GNNs, however, face 
 challenges when scaling to larger sequential circuits due to long-range in
 formation dependencies and insufficient functional supervision, leading to
  decreased accuracy and limited generalization.\nTo address these challeng
 es, we propose MOSS, a multimodal framework that integrates GNNs with LLMs
  for sequential circuit modeling. By enhancing D-type Flip-Flop (DFF) node
  features with embeddings from fine-tuned LLMs on RTL code, we focus the G
 NN on critical anchor points, reducing reliance on long-range dependencies
 . The LLM also provides global circuit embeddings, offering efficient supe
 rvision for functionality-related tasks. Additionally, MOSS introduces an 
 adaptive aggregation method and a two-phase propagation mechanism in the G
 NN to better model signal propagation and sequential feedback within the c
 ircuit.\nExperimental results demonstrate that MOSS significantly improves
  the accuracy of functionality and performance predictions for sequential 
 circuits compared to existing methods, particularly in larger circuits whe
 re previous models struggle. Specifically, MOSS achieves a 95.2% accuracy 
 in arrival time prediction.\n\nTopics: AI\n\nTracks: AI2: AI/ML Applicatio
 n and Infrastructure\n\nSession Chairs: Jun Shiomi (The University of Osak
 a) and Subhajit Dutta Chowdhury (Advanced Micro Devices (AMD))\n\n
END:VEVENT
END:VCALENDAR