Future billion-transistor processor chips will require innovations in architecture to overcome the fact that although gate delays scale with transistor sizes, wire delays do not. A SOFT ARCHITECTURE is a high-performance computer built by integrating dynamic compiler software with simple, fast processing hardware. The dynamic compiler replaces the large, complex, instruction-scheduling logic that is used successfully to extract instruction-level parallelism in contemporary superscalar processors but that is likely to limit the clock rates of future designs due to long wires. The research objective is to determine whether and how a soft architecture can exceed the performance of an architecture using dynamic instruction scheduling in hardware. While the dynamic compiler must make decisions less frequently than hardware, it can make potentially better decisions than hardware by exploiting additional information and by employing adaptivity. The research method is to construct a sequence of soft processor prototypes, dynamic compilers atop both conventional and custom (simulated) hardware, and to evaluate the performance of those prototypes by comparing to equivalent all-hardware solutions using standard benchmarks. The project provides a set of workbench tools for simulation and analysis that will also be used in architecture classes at the graduate and undergraduate level.

Agency
National Science Foundation (NSF)
Institute
Division of Computer and Communication Foundations (CCF)
Type
Standard Grant (Standard)
Application #
9876180
Program Officer
Pratibha Varma-Nelson
Project Start
Project End
Budget Start
1999-03-15
Budget End
2003-02-28
Support Year
Fiscal Year
1998
Total Cost
$200,038
Indirect Cost
Name
Georgia Tech Research Corporation
Department
Type
DUNS #
City
Atlanta
State
GA
Country
United States
Zip Code
30332