Loading Events

Faculty Candidate Seminar

CSE Lecturer Candidate Seminar: Andrew Lukefahr

Andrew LukefahrPhD CandidateUniversity of Michigan
SHARE:

CSE Lecturer Candidate Seminar
Title: Building Energy-Efficient Mobile Systems

We demand a lot from our smartphones. We want our websites rendered in
milliseconds, always-on connectivity, and multi-day battery life. To meet these
goals, today's phones require high performance and energy efficiency in both
software and hardware. However, many programmers today have very limited
knowledge of the underlying computational hardware or how to interact with it.
My talk will begin with an introduction to the bitwise operators necessary to
enable programmers to begin interacting directly with hardware. We will use
these operators to transmit and receive over a simulated USB interface.

Then we will dive deeper into the hardware by examining smartphone processor
architectures, which utilize both high-performance and energy-efficient
processor cores. Together these cores have the potential to deliver high
performance when needed while still maximizing battery life. Unfortunately,
migrating applications between these cores is expensive, often requiring tens of
thousands of cycles. Due to this overhead, designers avoid migrating to
energy-efficient cores while the phone is in use, missing opportunities to
utilize these cores and reducing battery life. However, even high-performance
applications show fine-grained phases of low performance. This remainder of the
talk is focused on building a single "composite" core that attempts to push the
notion of heterogeneity into the core itself. This enables a Composite Core to
quickly transition between high-performance and energy-efficient modes without
impacting performance, and allows it to capture and exploit fine-grained phases
to maximize energy efficiency.
Andrew Lukefahr is a PhD Candidate in Computer Science and Engineering at the
University of Michigan. His current research focuses on improving energy
efficiency of mobile devices through next-generation heterogeneous core
architectures. While at Michigan, he has served as a primary instructor for
Introduction to Computer Organization. His research and teaching interests
include energy-efficient mobile computing, embedded systems, and neural-inspired
computation

Sponsored by

CSE