COLLOQUIUM Department of Computer Science and Engineering University of South Carolina Architecture Generation for the Totem Project Katherine Compton Department of Electrical and Computer Engineering Northwestern University Date: October 7, 2002 (Monday) Time: 3:30-4:30PM Place: Swearingen 1A03 (Faculty Lounge) Abstract Reconfigurable hardware is ideal for use in systems-on-a-chip (SoCs) as it executes applications in hardware instead of software, and yet maintains a level of flexibility not available with more traditional custom circuitry. This flexibility allows for both hardware reuse and post-fabrication modification— important for alterations in the target applications, bug fixes, and reuse of systems across multiple similar deployments to amortize design costs. While FPGA designs could be included in an SoC, these standardized tiles tend to sacrifice performance and area in order to achieve a high degree of flexibility. Instead, if any characteristics of the target application set are known in advance, they can be used to customize the reconfigurable hardware—removing unneeded flexibility in order to reduce area and power requirements and increase performance. The Totem Project focuses on the automatic creation of customized reconfigurable architectures, including high-level design, VLSI layout, and associated custom place and route tools. In this talk I will provide a quick overview of the project, followed by a more detailed discussion of the high-level design phase, or "architecture generation," which determines the actual logic mix and routing structure of the architecture. This includes methods for two categories of reconfigurable architectures: near-ASIC designs with a minimum of reconfigurability to allow for hardware reuse across target circuits, and more flexible architectures with a one-dimensional segmented routing structure. Each of these design methods shows significant improvements through tailoring the architectures to the given application area. By providing a full automatic tool set for the generation and use of customized reconfigurable hardware, the Totem Project work will enable SoC designers to easily create efficient hardware implementations of key algorithms, while maintaining the flexibility benefits of FPGAs. Furthermore, because these tools are automatic and much faster than manual design, a wider design space can be explored without significantly impacting design costs. Katherine Compton is a PhD candidate at Northwestern University, in Evanston, Illinois, who expects to complete her degree in June 2003. She is studying under Dr. Scott Hauck of the University of Washington. She completed her bachelor's and master's degrees from Northwestern in 1997 and 1999, respectively. Her research interests include reconfigurable computing, reconfigurable architectures, automated architecture design, and benchmark synthesis.