Event

Ph.D. Research Proposal Exam: Sidra Gibeault

Friday, September 19, 2025
1:00 p.m.

Souad Nejjar
301 405 8135
snejjar@umd.edu

ANNOUNCEMENT: Ph.D. Research Proposal Exam

Name: Sidra Gibeault

Committee:

Professor Daniel Lathrop (Chair)

Professor Timothy Horiuchi

Professor Donald Yeung

Date/time: Friday, September 19, 2025 at 1:00 PM

Location: AVW 2328

Title: Programmable Analog Ising Computing Using Stochastic Magnetic Tunnel Junctions

Abstract: 

With computing performance improvements declining and modern applications being highly data-centric, new computing architectures and devices are required to continue making performance improvements. Traditional von Neumann computers cannot meet the demands of modern applications, including large optimization problems, artificial intelligence training, and high-bit encryption. Although quantum computers have shown promising results at medium scales, they are still error prone and require specialized lab equipment. Probabilistic computing is an alternative computing paradigm that has emerged in recent years, leveraging the random thermal fluctuations of nanodevices to explore complex solution spaces. The most commonly used nanodevices for probabilistic computing, superparamagnetic tunnel junctions (SMTJs), can be coupled together with programmable interaction strengths, allowing problems to be mapped onto the network and solved via simulated annealing. Probabilistic computing demonstrations in the literature are mostly limited to small-scale proof-of-concept demonstrations, with a microcontroller storing interaction strengths and calculating state updates. Scaling up to larger probabilistic computing systems requires direct coupling of SMTJs via CMOS circuitry, which provides a scaling pathway that takes advantage of in-memory crossbar architectures. In this proposal, we present the design of an Ising machine, a type of probabilistic computer, that uses programmable analog circuitry to implement direct coupling between SMTJs, eliminating the microcontroller from the loop and providing a clear pathway towards scalability.