|
|
MIT Electrical Engineering and Computer Science
EECS Event |
Thursday, April 26, 2001
3:30 PM (refreshments 3:15)
Edgerton Hall, Room 34-101
LCS Distinguished Lecture
Abstract
Life After Silicon: An Oxymoron?
Silicon has come to dominant our view of computing, since virtually all computing machines since the TX-1 have been based on it. As the dominant and sole infrastructure for computing, there is great interest (and concern) about what will happen when scaling of silicon technology stops. Recently there has been many papers written about the problems in scaling VLSI wires, and how the wiring problem will limit our ability to build faster computers. While people first started talking about the limitations of CMOS scaling over twenty years ago, recently the concerns about silicon limitations have increased. This concern has given rise to an active discussion about what is the next computing technology after silicon CMOS.
I will start by looking at VLSI scaling, and how we have leveraged these capabilities into increased processor performance. This analysis will lead to a number of challenges for future computer architects, even if scaling could be maintained. Given these challenges along with the technical problems of scaling silicon, I will briefly look at some heir apparent technologies like molecular and quantum computing. Unfortunately, in my view these technologies are unlikely to even compete with silicon for computing let alone surpass it.
The main issue is money -- we have an enormous sunk investment in silicon, and it is unlikely that we will make such a large investment in a new technology, unless it is already making money. Thus the question for new technology is not how great they might be, but how can they be useful now to warrant the development dollars needed to compete with silicon. Given this focus on economics, I will present Sam Wood's data on the economics of silicon scaling, and show that silicon scaling depends on demand for silicon increasing as the price decreases. Sometimes in the next 10-20 years, the cost of electronics will be a significant fraction of the world's GDP, and demand is likely to stop growing exponentially. When this happens, investment will slow, and so will scaling. At this point silicon will become like steel is today: an essential, but relatively stable building material.