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:20260730T152637Z
LOCATION:
DTSTART;TZID=America/Los_Angeles:20260728T123300
DTEND;TZID=America/Los_Angeles:20260728T123300
UID:dac_DAC 2026_sess315_RESEARCH1139@linklings.com
SUMMARY:From Characterization to Microarchitecture: Designing an Elegant a
 nd Reliable BFP-Based NPU
DESCRIPTION:Jie Zhang (Southeast University), Jiapeng Guan and Hao Zhou (D
 alian University of Technology), Xiaomeng Han and Tinglue Wang (Southeast 
 University), Ran Wei (Lancaster University), and Zhe Jiang (South East Uni
 versity)\n\nBFP is emerging as an attractive data format for edge NPUs, co
 mbining wide dynamic range with high hardware efficiency. However, its beh
 avior under hardware faults and its suitability for safety-critical deploy
 ments remain largely underexplored. Here, we present the first in-depth em
 pirical reliability study of BFP-based NPUs. Using RTL-level fault injecti
 on on NPUs, our bit- and path-level analysis reveals pronounced heterogene
 ous vulnerabilities and shows that the conventional end-to-end check becom
 es largely ineffective under nonlinear block scaling. Guided by these insi
 ghts, we design a fault-tolerant BFP-based NPU microarchitecture that alig
 ns the BFP computational semantics with reliability constraints. The desig
 n uses a row/column-wise blocking strategy to decouple the fixed-point man
 tissa computations from the scalar exponent path, and introduces ultra-lig
 htweight protection mechanisms for each. Experimental results demonstrate 
 that our design achieves near–dual modular redundancy reliability with onl
 y 3.55% geometric mean performance overhead and less than 2% hardware cost
 .\n\nTopics: Quantum\n\nTrack: DES6. Quantum Computing, Systems\n\nSession
  Chairs: Hokeun Kim (Arizona State University); BaekGyu Kim (Daegu Gyeongb
 uk Institute of Science and Technology); Xiang Chen (Peking University); S
 angyeob Kim (Yonsei University); Robert M. Radway (University of Pennsylva
 nia); Prashanth Mohan (University of Pittsburgh); Xiaoxuan Yang (Universit
 y of Virginia, Stanford University); Ramtin Zand (University of South Caro
 lina); Hadi Kamali (University of Central Florida); Jiafeng Xie (Villanova
  University); Prabuddha Chakraborty (University of Maine); Junichiro Kadom
 oto (University of Tokyo); and Yosuke Ueno (RIKEN)\n\n
END:VEVENT
END:VCALENDAR
