BEGIN:VCALENDAR
CALSCALE:GREGORIAN
VERSION:2.0
METHOD:PUBLISH
PRODID:-//Drupal iCal API//EN
X-WR-TIMEZONE:America/New_York
BEGIN:VTIMEZONE
TZID:America/New_York
BEGIN:DAYLIGHT
TZOFFSETFROM:-0500
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
DTSTART:20070311T020000
TZNAME:EDT
TZOFFSETTO:-0400
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0400
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
DTSTART:20071104T020000
TZNAME:EST
TZOFFSETTO:-0500
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
SEQUENCE:1
X-APPLE-TRAVEL-ADVISORY-BEHAVIOR:AUTOMATIC
UID:252006
DTSTAMP:20261116T091345Z
DTSTART;TZID=America/New_York:20261123T110000
DTEND;TZID=America/New_York:20261123T130000
URL;TYPE=URI:https://www.wpi.edu/news/calendar/events/ece-phd-dissertation-
 defense-dev-mehta
SUMMARY:ECE PhD Dissertation Defense by: Dev Mehta
DESCRIPTION:Title:\nPhysical Security in AI Accelerators\n\nAbstract:\nWith
  the increasing deployment of artificial intelligence (AI) on physical dev
 ices\, the security of its intellectual property (IP) has become increasin
 gly important. Although the security paradigms employed to protect these c
 ircuits have been shown to be robust in simulation\, they frequently exhib
 it weaknesses once they are implemented in hardware. These weaknesses are 
 rooted in the inherent physics of CMOS-based circuits\, which renders it c
 onsiderably more difficult for designers to create circuits that withstand
  a wide variety of attack vectors and to avoid design choices that inadver
 tently weaken their designs.\nThis dissertation investigates a range of si
 de-channel analysis techniques in order to expose the vulnerabilities of s
 uch designs\, and it makes two primary contributions. First\, we demonstra
 te physical attacks that break several protection schemes that can be empl
 oyed to protect neural networks (NNs). Second\, we study the impact of phy
 sical design choices on security from both the attack and the protection p
 erspectives\, with particular emphasis on timing analysis. FPGA implementa
 tions\, in combination with pre-silicon analysis\, reveal the influence of
  timing on glitches and show how an elevated operating temperature can ren
 der a masked NN implementation vulnerable to a first-order attack. Beyond 
 the power side channel\, we further demonstrate that optical side channels
  can be exploited to attack AI accelerators at different levels of protect
 ion. Overall\, our results show that the security guarantees of masked and
  cryptographically protected AI hardware\, which hold in abstract security
  models\, such as the probing model for masking and the cryptographic secu
 rity models of garbled and universal circuits\, do not carry over to the p
 hysical implementation\, since timing differences below the clock-cycle ab
 straction and backside optical access expose sensitive values to an advers
 ary with physical access.\n\nResearch Advisor:\nProf. Fatemeh (Saba) Ganji
 \nECE Department\, WPI\n\nCommittee Members:\nProf. Patrick Schaumont\nECE
  Department\, WPI\nProf. Domenic Forte\nECE Department\, University of Flo
 rida
END:VEVENT
END:VCALENDAR
