Last month BDTI completed an analysis of the latest DSP cores from the three leading core licensors. Paging through the analysis, I noticed some striking similarities among the competing cores. All use flexible, multi-issue architectures; RISC-like instruction sets; and a mix of 16- and 32-bit instructions.
Interestingly, these basic architectural features are also found in many high-end embedded general-purpose processors. And the similarities between DSPs and GPPs are growing.
It almost seems as if processor designers have agreed on what the perfect embedded processor should look like and that all architectures are converging on this design. Indeed, the announcements I've seen recently suggest that the trend toward architectural similarity will continue.
A casual observer might think that because processors are becoming more alike, no processor will hold an advantage over its competitors for long. This may be true to a certain extent, but there will continue to be important differences among architectures. These differences often reflect differences in target markets. For example, some processors include video-oriented instructions but not communications-oriented instructions, while the reverse is true for other processors. Even two chips based on the same architecture can have quite different characteristics: One chip might be fast but power-hungry, another chip slower but more energy-efficient.
There are also huge differences among processors in such areas as tool quality, availability of application software and on-chip integration. Again, the differences often reflect differences in target markets. For example, some processors have peripherals and coprocessors designed for consumer electronics, while others have supporting blocks designed for telecom infrastructure.
One upshot of all this is that vendor focus will play an increasingly important role in distinguishing embedded processors from one another. As a result, vendors that focus on markets in which they are already strong will develop momentum that will be hard for competitors to overcome. Vendors that are not focused on a given application will find it increasingly difficult to remain relevant in that application. And vendors that are not already in a given market will find it increasingly difficult to enter it.
For processor vendors, my advice is to pick your battles carefully. While there is nothing wrong with providing general-purpose solutions, I think it will become increasingly necessary for vendors to concentrate their efforts on application-specific solutions. And even if your processors can theoretically compete in a wide range of applications, you can only provide focused solutions for a limited number of applications.
For system developers, my advice is to pay close attention to a vendor's target markets and market share when selecting a processor. Vendors that are marginal players in a market may not be able to sustain business in that market-possibly leaving you stranded when it comes time to build a second-generation product.
Jeff Bier, president of Berkeley Design Technology Inc. (www.BDTI.com), a consulting firm providing analysis and advice on DSP technology. Kenton Williston of BDTI contributed to this column.