Self-aware Memory Management for Emerging Energy-Efficient Architectures

Name: Biswadip (Deep) Maity

Chair: Nikil Dutt

Date: July 6, 2023

Time: 9:00 AM

Location: DBH 3011

Committee: Fadi Kurdahi, Nalini Venkatasubramanian

Title: Self-aware Memory Management for Emerging Energy-Efficient Architectures

Abstract:

The ever-increasing demands of data-intensive applications have presented significant challenges in memory performance and energy efficiency. Traditional approaches to memory management, relying on workload-specific optimizations and static memory configurations, have become inadequate in addressing the dynamic nature of modern computing systems. Inspired by principles from psychology and neuroscience, computational self-awareness aims to develop intelligent systems capable of learning from past experiences, reasoning about their current state, and making informed decisions at runtime. My research investigates the application of computational self-awareness in memory management, with a specific focus on understanding the impact of different degrees of self-awareness within the memory subsystem. By applying self-awareness principles to memory management, we show that it is possible to create an energy-efficient memory subsystem that adapts to changing workloads and optimizes memory usage across various architectures and layers of the system abstraction.

Short Bio:

Biswadip Maity is a Ph.D. candidate at the Computer Science department of University of California, Irvine. Previously, he was a Software Engineer at Microsoft. His research is in the area of runtime resource allocation and power management for data-centric architectures with a special focus on self-aware embedded systems, memory systems, system software and architecture for heterogeneous systems. He is a member of the Dutt Research Group working with Prof. Nikil Dutt. His work spans across resource-constrained microcontrollers, IoT systems and Cloud computing performance (in collaboration with Facebook). Biswadip Maity is a Charpak Scholar, a DAAD Scholar, and a recipient of a Facebook Research Grant.