15.4. The Future Direction of SOC Design

Engineers have been designing and building bridges for thousands of years. During those centuries, they have developed these several basic bridge types and many combinations of the basic types. Several different bridge types can often solve a specific bridging problem, so there’s no push-button method for selecting one bridge type over another except in situations where there’s one obvious solution to a bridging problem. Instead, bridge engineers have developed ways of generating different alternatives by combining the basic bridge types and then comparing the alternatives to pick the “best” one, where “best” includes a wide range of criteria including quantitative (span length, carrying capacity, cost, construction time, etc.) and qualitative (bridge aesthetics, era-specific and location-specific materials limitations, building-site construction restrictions, etc.).

This book advocates a similar approach to SOC architecture and design. SOC designers need rigorous ways of evaluating alternative architectures and rigorous methods for directly implementing the selected architecture without the need to manually translate and implement complex, human-readable specifications.

If you accept the ITRS:2005 prediction that future SOC designs will incorporate tens and then hundreds of processing engines, then you must ponder the following questions:

• How will all of those processing engines be designed?

• Will the processing engines be programmable or fixed-function?

• How will the processing engines be controlled?

• How will all of those processing engines communicate with each other?

The current state of the art in system design cannot answer these questions because only a relatively few people have designed and worked with such complex heterogeneous processing systems. Automated design tools do not yet exist to deal with such complexity. However, people have thought about such systems and have developed some highly parallel-processing systems. The art of developing such systems still needs to be reduced to an engineering regimen. The following sections briefly discuss some of the most interesting experiments in highly parallel computing that might benefit from SOC implementations.