Lecture
Datapath design for high-performance microprocessors:
An energy-delay perspective
Speaker: |
George Dimitrakopoulos |
Date: |
Thursday, 22 November 2007 |
Time: |
12:00-14:00 |
Location: |
"Mediterranean Studies" Seminar Room, FORTH. Heraklion, Crete |
Host: |
| Abstract: |
Integrated circuit design
has seamlessly entered the power-limited scaling regime, where
the traditional goal of achieving the highest performance has
been displaced by optimization for both performance and power. Multi-GHz microprocessors are an example of power-constrained designs where efficient circuit topologies and techniques are required to achieve high operating frequencies while keeping power consumption to a resonable level. Therefore, the design of energy-delay efficient datapaths which are employed for the execution of the processor's integer and floating point instructions as well as the execution of media-oriented SIMD instructions is of high importance. In this talk, at first we will briefly present an optimization framework that allows the designer to explore the whole energy-delay space of each design and to quantify the benefits earned from new circuit techniques and logic-level optimizations. In the following, we will present a variety of new datapath circuits and design techniques that are based on an inherently simpler implementation and offer superior performance in the energy-delay sense. |
| Bio: |
Giorgos Dimitrakopoulos received
the Diploma in computer engineering and informatics in 2001 from
the Computer Engineering and Informatics Dept. of the University
of Patras. He received the M. Sc. degree in "Hardware/Software
Integrated Systems" and the PhD in computer engineering in 2003
and 2007, respectively from the same department. Since 2001, he is a member of the Technology and Computer Architecture Laboratory of the Computer Engineering and Informatics Department, University of Patras. His research interests include VLSI circuit design, energy-efficient microarchitectures and global energy-delay optimization using a mathematical programming approach. |
