At last month’s Microprocessor Forum, Sun Microsystems CTO Greg Papadopoulos predicted that microprocessors as we know them will disappear by 2010. In his view, microprocessors will continue to absorb surrounding chips until the entire computer is contained in a single chip. This prediction reminded me of claims by Texas Instruments that cell phones will soon contain nothing more than a single chip and a handful of passive components.
There is no denying the trend towards higher integration in everything from PCs to cell phones. The advantages of increased integration—particularly the potential for reduced system size, cost, and power consumption—are so compelling that it is tempting to extrapolate this trend line to single-chip utopia. However, the benefits of integration are not always relevant, and they are sometimes coupled with serious drawbacks.
Consider the advantage of reduced size. While some applications are pressed for space, others have room to spare. For example, who needs a smaller cell phone? They are barely big enough to hold now. Even if you did shrink the chips, you wouldn’t shrink the phone much: today’s cell phones are mostly battery, display, and keypad.
The benefits of lower power consumption are also dicey. All other things being equal, you’d expect that fewer chips would equate to lower power consumption. But all other things are never equal, particularly if you are building a truly single-chip solution. Today, most “system on a chip” designs require external memory, power electronics, and other supporting chips. In many cases, each chip uses a different fabrication process. Going to a single-chip solution may require fabrication process compromises that cause some components to consume more power. Depending on the mix of components and inter-component activity, this could lead to higher overall power consumption.
Likewise, cost is an oft-cited argument for integration. Again, this argument is strongest when all other things are equal, such as when all components are digital logic. If you have to use special, expensive fabrication techniques to integrate dissimilar components onto a single chip, the cost argument may break down. And this is before you consider the extra engineering effort, yield problems, and other factors that crop up as chips grow more complex.
For most applications, the benefits of increased integration still outweigh these drawbacks. But like all good things, these benefits eventually must end. Despite the hype, few applications will see a day when “system on a chip” can be taken literally.
Add new comment