Event

Ph.D. Research Proposal Exam: Temitayo Nicholas Adeyeye

Tuesday, September 16, 2025
11:00 a.m.

Souad Nejjar
301 405 8135
snejjar@umd.edu

ANNOUNCEMENT: Ph.D. Research Proposal Exam

Name: Temitayo Nicholas Adeyeye

Committee:

Professor Daniel Lathrop (Chair)

Professor Dinesh Manocha

Professor Timothy Horiuchi

Date/time: Tuesday, September 16, 2025 at 11:00 AM 

Title: P-computing without P-bits: Using the Time Domain to Improve the Efficiency of Probabilistic Computing

Abstract: 

Two major drivers of the digital computing era have been advances in memory and computing technologies. However, the end of Dennard scaling has halted the era of “free” performance gains, with transistor scaling occurring at a consistent density factor. This means that while more transistors can fit in the same area with the same power and cooling constraints, not all transistors can be switched simultaneously. This leads to a phenomenon called dark silicon, where only certain parts of the chip can be powered at a time. Another recent challenge for digital computing is the unequal progress of computing and memory technologies. With processors running faster and physically separated from memory, performance drops because the processor must wait for data from memory. This bottleneck, known as the “Von Neumann” bottleneck, is worsened by the rise of data-heavy applications like machine learning and high-performance computing for scientific and engineering simulations. This work aims to address these issues by combining a low activity factor computing paradigm with in-memory computing using novel devices. Our goal is to create a highly efficient, scalable framework for solving complex optimization problems.

In this project, we utilize a temporal information encoding framework to extract stochasticity from probabilistically switching magnetic tunnel junction devices. We develop MTJ-based circuitry called probabilistic delay cells (PDCs) to generate stochastic, time-delayed signals. We analyze the distribution and tunability of the random signals produced and integrate the PDC with other temporal computing primitives to develop a probabilistic computer that uses the Markov Chain Monte Carlo sampling method to solve complex optimization problems.