www.flexsoc.org CHALMERS
Main Research Project People Contact
Research Publications External links

FlexSoC: A Flexible System-on-chip Platform for Embedded Systems

The FlexSoC program, launched during 2003, aims to develop new architectural techniques for the complex processors necessary for high-performance embedded systems. One of our driving examples is the high-performance cellular phone. Many present-day cellular phones have features (such as games, 3d graphics, video playback, and wireless Internet connectivity) which until recently were only available in high-performance desktop computers. Moreover, in the cell-phone case, these features must be delivered in a package small enough to fit a purse or a pocket, and at fractions of the price and power dissipation of a desktop system.

The power dissipation is an especially vexing problem for a cellular phone. A high dissipation has many undesired consequences: batteries drain too quickly, or must otherwise be large and heavy; elaborate and costly heat-removal arrangements must be employed; and reliability is also adversely affected. Finally, the environmental load of consumer-electronics products is dominated by the power used in the field, so lower dissipation makes a product more environmentally friendly.

Although a cellular phone is used as an example here, the benefits of low power dissipation and ease of programming would be welcome in most embedded-processor applications. Software development costs are large for all complex embedded systems; power requirements directly affects the manufacturing cost for power supplies and enclosures.

The goals of the FlexSoC program requires a radical remodeling of the processor architecture. Instead of the present-day solutions, where a general-purpose microcontroller orchestrates a collection of special-purpose "accelerators", we envision a network of simpler heterogeneous datapath elements, controlled through reconfigurable instruction-stream decoding. Innovations in circuit design, in computer architecture, and in compiler techniques will all be necessary to bring this vision closer to reality.