Does the Roadmap need more shortcuts to the future?
08/01/2003
 |
The Roadmap process has helped the semiconductor industry keep its entire manufacturing infrastructure moving forward at a rapid but synchronized pace, somewhat like rolling up a huge carpet without creating kinks and bulges. The industry needs teamwork, like the neatly aligned grounds crew at a ballpark smoothly removing the tarp from the infield after a deluge.
Since manufacturing capabilities set the pace for progress in chipmaking, the Roadmap also helps semiconductor companies develop their own internal roadmaps for sequencing steadily more capable, competitive circuits.
In recent years, the road-mapping process has been expanded to include factors such as yield management and environment, health, and safety issues. New materials are also increasingly being worked into potential solutions to help extend the basic CMOS process by pushing out some of the physical limits.
However, looking ahead, there may be a need to take an even broader perspective in future road-mapping exercises. As the industry approaches the billion-transistor chip, a question arises about how all those devices can be used. There are certainly logical reasons for them in memory chips, powerful processors, video engines, and the like. There is also the possibility of many system-on-a-chip solutions, especially for compact, portable devices. But combining extensive intellectual property on one chip, especially with wide variations in design tools used to synthesize individual circuit functions, can make overall verification too difficult.
Since each fab has its own process variations, IP optimized for different fabs also may not be compatible for use on the same chip. A fabless company may choose to use only the IP available from a particular foundry, but this may prove too limiting. As a result, some SOC projects become so cumbersome they are abandoned in favor of multichip solutions. Progress toward smoothing the integration of IP on a single chip could be indicated in future roadmaps.
There is also the problem of mixed signal devices needed for a steadily expanding world of applications. Some fabs specialize in combining CMOS with analog and rf, or mixing high-performance logic with DRAM and even SRAM on the same chip. But the needs are expanding even wider. Some chips could use some optical circuitry; others might incorporate microelectromechanical systems or MEMS devices. For very high frequencies, some compound semiconductor devices might be needed. Extending basic CMOS processing to incorporate these wide variations becomes too costly, however, so the industry is moving in that direction by adding more convenient variations, like silicon-on-insulator (SOI), silicon/germanium, and strained silicon. System-in-a-package (SiP) is a method for combining devices made with different processes into the same package, but at the sacrifice of the performance advantages of single-chip solutions.
There are efforts to economically combine devices made with different processes onto single wafers, perhaps using the wafer-cleaving techniques developed for making bonded SOI wafers. Motorola has made considerable progress toward overcoming the lattice mismatch problems of building compound semiconductor devices onto a CMOS wafer. More of the specialized device requirements might be incorporated into substrates, so fewer processing steps will be needed to achieve unique functions.
The core digital CMOS processes will remain as a central theme of roadmaps for many generations to come, but as the industry looks to broaden applications, these non-mainstream processes may become part of the infrastructure planning and coordination as well. It may not be too long before even nanotechnology devices like carbon nanotubes enter the mix. If efficient, cost-effective ways are found to combine all these specialized properties with mainstream CMOS processes, the applications universe may expand dramatically.
An expanded Roadmap could provide some very worthwhile shortcuts to the future, benefiting users as well as the industry.
Robert Haavind
Editor in Chief