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:20260402T024534Z
LOCATION:3002\, Level 3
DTSTART;TZID=America/Los_Angeles:20250623T110000
DTEND;TZID=America/Los_Angeles:20250623T111500
UID:dac_DAC 2025_sess120_RESEARCH2079@linklings.com
SUMMARY:A PulseWidth-IN-PulseWidth-Out Universal Nonlinear Processing Elem
 ent for Time-Domain In-Memory Computing Systems
DESCRIPTION:Yihao Chen, Pengcheng Feng, Zhigang Li, Gang Chen, Xianghua Lu
 , Rongxuan Shen, and Xiaoxin Xu (Institute of Semiconductors, Chinese Acad
 emy of Sciences)\n\nTime-Domain In-Memory Computing (TD-IMC) has emerged a
 s a promising analog computing architecture for edge AI applications. Howe
 ver, the lack of developed hardware operators, especially general nonlinea
 r operators, necessitates frequent cross-domain data transmission in pract
 ical TD-IMC systems, significantly reducing energy efficiency.\nIn this wo
 rk, we propose a PulseWidth-IN-PulseWidth-OUT Universal Nonlinear Processi
 ng Element (PIPO-UNPE) to address the challenges of nonlinear processing i
 n analog computing. By implementing an RRAM-based two-layer ReLU network, 
 the PIPO-UNPE performs universal nonlinear operations entirely in the time
  domain. Algorithmically, we introduce Dynamic Loss-Responsive Subset Enha
 ncement (DLRSE) to boost the performance of this low-cost network in funct
 ion approximation tasks. From a hardware perspective, we design an RRAM-ba
 sed pulse-driven programmable current source and a low-latency dispersion 
 comparator-based voltage-to-time converter (VTC) to enhance both the energ
 y efficiency and precision of the PIPO-UNPE. Hybrid simulations reveal tha
 t the PIPO-UNPE consumes 912 uW of power while delivering a throughput of 
 10M NOPS (Nonlinear Operations Per Second). Incorporating the PIPO-UNPE in
 to the TD-IMC accelerator can increase energy efficiency by a factor of 9.
 5 to 25, keeping the accuracy loss below 0.1%.\n\nTopics: Design\n\nTracks
 : DES2A: In-memory and Near-memory Computing Circuits\n\nSession Chairs: Y
 u Cao (University of Minnesota) and Sumitha George (North Dakota State Uni
 versity)\n\n
END:VEVENT
END:VCALENDAR