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:3006\, Level 3
DTSTART;TZID=America/Los_Angeles:20250623T113000
DTEND;TZID=America/Los_Angeles:20250623T114500
UID:dac_DAC 2025_sess139_RESEARCH1656@linklings.com
SUMMARY:A Novel Image-Graph Heterogeneous Fusion Framework for Static IR D
 rop Prediction
DESCRIPTION:Dan Niu, Dekang Zhang, and Yichao Cao (Southeast University); 
 Zhou Jin (China University of Petroleum, Beijing); Chao Wang and Yichao Do
 ng (Southeast University); and Changyin Sun (Anhui University)\n\nIR drop 
 analysis is crucial for ensuring the reliability and performance of integr
 ated circuits (ICs) but poses computational challenges as the IC designs g
 row larger, especially for ultra deep-submicron VLSI designs. Deep learnin
 gs (DL) as the efficiency-promising solutions, mainly employ various CNN-b
 ased networks to achieve image-to-image IR drop predictions. However, they
  neglect and lose the power delivery network (PDN) global spatial features
  and cell instance topological information. This paper proposes a novel im
 age-graph heterogeneous fusion framework (IGHF), which integrates the effe
 ctiveness and complementarity of dual branches (CNN and GNN) for higher pr
 ediction performance. In the CNN-based Power ScaleFusion Unet branch, the 
 proposed long-range and local-detail encoder (LLE) integrates seamlessly w
 ith the hierarchical and adjacent compensation group (HACG) module. This d
 esign facilitates effective multi-scale global-to-local spatial power feat
 ure extraction within the PDN and enables adaptive high-to-low-level featu
 re fusion and compensation in the decoder.\nMoreover, a cell voltage aware
  (CVA) module in the GNN branch is designed to adaptively aggregate PDN to
 pological features of heterogeneous neighbors of different orders. Compara
 tive experiments demonstrate that the proposed IGHF achieves significant a
 ccuracy improvements, outperforming the state-of-the-art MAUnet and widely
 -used IREDGe methods by considerable margins of 24.6\% and 55.0\% reductio
 n in prediction error, while the prediction maps possess higher structural
  fidelity. Transfer experiments indicate that IGHF with transfer learning 
 can improve the accuracy in real circuits with the few-shot real circuit t
 est cases.\n\nTopics: EDA\n\nTracks: EDA6: Analog CAD, Simulation, Verific
 ation and Test\n\nSession Chairs: Arindam Basu (School of Computer Science
  and Engineering, Nanyang Technological University) and Markus Olbrich (Le
 ibniz University Hannover)\n\n
END:VEVENT
END:VCALENDAR