Recently a friend of mine was lamenting the fact that so many new processors continue to be designed, and so few achieve any commercial success. “Why do people persist in designing new processors?” he asked. “Maybe because universities keep training processor designers,” I speculated. On closer reflection, though, I believe there’s more to the story.
Historically, processor designers created highly specialized DSP-oriented architectures with the goal of achieving real-time signal processing performance with a programmable device, at a low cost. For some applications (such as high-quality video encoding), achieving sufficient speed at reasonable cost is still a primary motivator for architectural innovation. But for many others, current processors are sufficiently fast. In fact, many signal processing applications can now be run on general-purpose processors (GPPs).
With clock speed increases afforded by modern IC fabrication processes, even a simple, inexpensive 32-bit GPP with few (or no) DSP-oriented features is capable of handling many real-time audio tasks, like MP3 decoding. These GPPs, with their simple architectures and strong development tools, are in many respects easier to work with than DSPs—or even DSP-enhanced GPPs. Given the ease-of-use advantage of simple processors and their adequate performance, why continue to design (and use) specialized, DSP-oriented processors?
The answer, I believe, is battery life. Many embedded signal processing applications are highly constrained in terms of energy consumption. By using specialized architectures, processors can meet the real-time computation requirements of the application while operating at lower clock frequencies (and, as a result, lower voltages) than general-purpose architectures, thus achieving much lower energy consumption for the same application. That's why system designers still often put a specialized DSP chip or core next to a GPP—even when the GPP could handle the DSP tasks by itself. For many energy-sensitive applications, the added hardware cost and programming complexity is justified.
So for processor designers, there’s good news: Innovation in processor architecture is still very valuable. But in many cases, the goal of that innovation has shifted—from maximum performance to minimum energy consumption.
Add new comment