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:20240626T180033Z
LOCATION:3010\, 3rd Floor
DTSTART;TZID=America/Los_Angeles:20240626T103000
DTEND;TZID=America/Los_Angeles:20240626T104500
UID:dac_DAC 2024_sess142_RESEARCH1614@linklings.com
SUMMARY:MASC: A Memory-Efficient Adjoint Sensitivity Analysis through Comp
 ression Using Novel Spatiotemporal Prediction
DESCRIPTION:Research Manuscript\n\nChenxi Li (Super Scientific Software La
 boratory, China University of Petroleum-Beijing); Boyuan Zhang (Indiana Un
 iversity, Bloomington); Yongqiang Duan and Yang Li (Super Scientific Softw
 are Laboratory, China University of Petroleum-Beijing); Zuochang Ye (Tsing
 hua University); Weifeng Liu (Super Scientific Software Laboratory, China 
 University of Petroleum-Beijing); Dingwen Tao (Indiana University, Bloomin
 gton); and Zhou Jin (Super Scientific Software Laboratory, China Universit
 y of Petroleum-Beijing)\n\nAdjoint sensitivity analysis is critical in mod
 ern integrated circuit design and verification, but its computational inte
 nsity grows significantly with the size of the circuit, the number of obje
 ctive functions, and the accumulation of time points. This growth can impe
 de its wider application. The intimate link between the forward integratio
 n in transient analysis and the reverse integration in adjoint sensitivity
  analysis allows for the retention of Jacobian matrices from transient ana
 lysis, thereby speeding up sensitivity analysis. However, Jacobian matrice
 s across multiple timesteps are often so large that they cannot be stored 
 in memory during the forward integration process, necessitating disk stora
 ge and incurring significant I/O overhead.\nTo address this, we develop a 
 memory-efficient sensitivity analysis method that utilizes data compressio
 n to minimize memory overhead during simulation and enhance analysis effic
 iency. Our compression method can efficiently compress the sparse tensor t
 hat contains the Jacobian matrices over time by exploiting the spatiotempo
 ral characteristics of the data and circuit attributes. It also introduces
  a shared-indices technique, a cutting-edge spatiotemporal prediction mode
 l, and robust residual encoding.\nWe evaluate our compression method on 7 
 datasets from real-world simulations and demonstrate that it can reduce th
 e memory requirements for storing Jacobian matrices by more than 16x on av
 erage, which is significantly more efficient than other state-of-the-art c
 ompression techniques.\n\nTopic: EDA\n\nKeyword: Analog CAD, Simulation, V
 erification and Test\n\nSession Chairs: Weidong Cao (Washington University
 , St. Louis) and Ahmet Budak (Analog Devices, Inc. (ADI))
END:VEVENT
END:VCALENDAR