| Title | The End of Traditional CMOS |
| Author | *Moderator: Raul Camposano (Xoomsys, United States), Panelists: Gul Agha (University of Illinois at Urbana-Champaign, United States), Yasuhiko Hagihara (Device Platforms Research Laboratories, NEC Corporation, Japan), Igor Markov (University of Michigan, United States), Chandu Visweswariah (IBM T. J. Watson Research Center, United States) |
| Page | p. 553 |
| Abstract | The rumors of CMOS’ death have been greatly exaggerated. After 2 decades as the workhorse of the electronics industry, device counts have scaled by ~104 and speeds by ~102. Transistor “performance” has consequently scaled by a ~106 and has arguably been the main driver of system performance. CMOS isn’t showing signs of ending its reign any time soon. Or is it?
The panel will discuss this question, in particular the following positions:
•Business as usual. “Simple” (Dennard) scaling has not been simple for decades, its just getting a bit harder but essentially nothing new. We have dealt with new processes, materials, devices, circuits for a long time. We can make it below 10nm.
•The problem is really economic. Scores of companies are exiting the fab business already. Scaling will most definitely end some day, and the end is coming slowly. Depending on the volume, different applications are getting stuck at different nodes. But CMOS will continue to be the principal game in town for a long time. So, it is more important to look at what we can do (design) with silicon than to further scale it.
•CMOS will be “hybridized” by add on technologies, for example to increase communication speed both on- and off-chip; or to produce small, very fast non-volatile memories.. Main candidates: Optical and nanoswitches.
•We ought to look at something new like Quantum Computing. Some niche applications like cryptography will benefit greatly and will drive the development of such new technologies. Power, reliability and material limits (among others) will prevent further progress in CMOS. |