Link to CALCM Home  

New Additions to the CASH Compiler

Tuesday October 19, 2004
Hamerschlag Hall D-210
4:00 pm



Tiberiu Chelcea
Carnegie Mellon University

CASH is a CAD tool developed by the Phoenix group at Carnegie Mellon University. CASH compiles high-level programs written in C into asynchronous circuits. The particular microarchitecture targeted by CASH is "Application-Specific Hardware", which is an implementation of Spatial Computation. ASH requires no clocks, nor any global signals. The computation structures in ASH are never shared. The resulting circuits use only localized communication, use no global control and are self-synchronized. The performance of ASH circuits is comparable with that of synchronous microprocessors, but with an energy-delay product of up to three orders of magnitude better.

While the current results are extremely promising, we believe that better circuits can be synthesized. In this talk, several new approaches to improving the quality of circuits generated by CASH are presented. First, the main performance bottleneck in our circuits, the interface with the main memory, is addressed. Second, the integration of two new pipelining styles is presented, as well as their impact on performance, area, and power consumption. Finally, the performance of some functional units is addressed through some asynchronous-specific modifications. These ideas are at various stages of integration into CASH, and several preliminary results will be presented.



Tiberiu Chelcea received the B.S. and M.Sc. degrees in Electrical Engineering from the Politehnica University of Bucharest, Romania, in 1995 and 1996, respectively, and the Ph.D. Degree in Computer Science from Columbia University, New York, in 2004.

He is currently a Postdoctoral Fellow within the Phoenix research group at the Department of Computer Science, Carnegie Mellon University. His research interests include asynchronous circuits, design techniques for mixed-timing digital systems, and synthesis methods from high-level languages down to asynchronous systems.

 

Department of Electrical and Computer EngineeringCarnegie Mellon UniversitySchool of Computer Science