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:3003\, Level 3
DTSTART;TZID=America/Los_Angeles:20250623T113000
DTEND;TZID=America/Los_Angeles:20250623T114500
UID:dac_DAC 2025_sess152_RESEARCH1169@linklings.com
SUMMARY:Security of Approximate Neural Networks against Power Side-channel
  Attack
DESCRIPTION:Japa Aditya, Jack Miskelly, Maire O'Neill, and Chongyan Gu (Qu
 een's University Belfast)\n\nEmerging low-energy computing technologies, i
 n\nparticular approximate computing, are becoming increasingly\nrelevant i
 n key applications. A significant use case for these\ntechnologies is redu
 ced energy consumption in Artificial Neural\nNetworks (ANNs), an increasin
 gly pressing concern with the\nrapid growth of AI deployments. It is essen
 tial we understand the\nsecurity implications of approximate computing in 
 an ANN con-\ntext before this practice becomes commonplace. In this work, 
 we\nexamine the test case of approximate ANN processing elements\n(PE) in 
 terms of information leakage via the power side channel.\nWe perform a wei
 ght extraction Differential Power Analysis\n(DPA) attack under three appro
 ximation scenarios: overclocking,\nvoltage scaling, and circuit level bitw
 ise approximation. We\ndemonstrate that as the degree of approximation inc
 reases the\nSignal to Noise Ratio (SNR) of power traces rapidly degrades.\
 nWe show that the Measurement to Disclosure (MTD) increases\nfor all appro
 ximate techniques. An MTD of 48 under precise\ncomputing is increased to a
 t minimum 200 (bitwise approximate\ncircuit at 25% approximation), and und
 er some approximation\nscenarios >1024. i.e. an increase in attack difficu
 lty of at least\nx4 and potentially over x20. A relative Security-Power-De
 lay\n(SPD) analysis reveals that, in addition to the across the board\nimp
 rovement vs precise computing, voltage and clock scaling\nboth significant
 ly outperform approximate circuits with voltage\nscaling as the highest pe
 rforming technique.\n\nTopics: Security\n\nTracks: SEC3: Hardware Security
 : Attack & Defense\n\nSession Chairs: Michael Zuzak (Rochester Institute o
 f Technology) and Prabuddha Chakraborty (University of Maine)\n\n
END:VEVENT
END:VCALENDAR