|
Peter Cochrane's Hard Drive 2000 Powering down to quantum level IT is now more than 35 years since Gordon Moore, the co-founder of Intel, predicted that the number of transistors on a memory chip would double every 18 months or less. This observation has since become known as Moore's Law, and despite regular predictions that it is all about to come to an end because of the limits of physics, it has been sustained since 1960. It has also been complemented by a lesser-known law put forward by Arthur Rock, stating that the cost of the capital equipment necessary to fabricate integrated circuits will double every four years. So here we are with more than 25 million transistors on a chip, and clock rates in excess of 500MHz commonplace; 100 million transistors per chip and 1,000Mz are in view. Meanwhile, production plants now cost more than $4 billion. So where is the limit? Today, we use around 100,000 electrons to store a single bit on individual transistor devices with feature sizes around 0.2 microns. In principle, we could easily get down to about 100 electrons/bit before we start to see significant quantum effects, and state (1:0) uncertainties giving the occasional bit errors that warrant active correction. A commercial chip realisation close to 10 electrons/bit is already seen as likely, providing we can afford the fabrication plant. So it appears that we are assured of at least another 10 years of Moore's Law by merely continuing on our present trajectory, which means almost another 1,000x increase in computing power. Another simple move would be to increase chip size and the number of transistors/chip to more than 100 million. By using X-ray or electron-beam etching, we can expect transistor feature sizes well below 0.1 microns, and we may then realise 1000 million. Also, we have yet to embark on any serious 3D fabrication to increase the volumetric density of bit storage and processing. Doubling the dimensions of a chip sees the number of transistors quadruple, but go to 3D and it increases eightfold. Such a move could see overall gains of a further 1000x, and I think we can therefore assume that Moore's Law will continue beyond 2020. What does all this mean? In 1980 we were buying chips with a storage capacity of around 64Kb; by 1990 this had risen to 16Mb, and today to 256Mb. Do not be surprised to find that 2010 sees capacities around 64Gb, and 2020 up to 1,000Gb, all at a reasonably constant price. Such capacities will easily see off today's CDs and DVDs, in terms of physical size and cost per bit stored. Beyond this, we will have to process down at the one electron/bit level to see further advances, and that is a realm we have yet to explore. But I suspect there are still some massive gains to be had by good architectural, interface and software design. By good power and thermal management, powering up only those parts of a chip that are needed at any one time, and getting the heat out efficiently, we should also achieve significant performance enhancements beyond those we enjoy today. Not only might we expect a constant or falling price; we should also expect a constant or falling power demand. My guess is 100,000,000x more, at a constant price and power consumption, before we really have to get into the sub-atomic world of quantum electronics. Peter Cochrane holds the Collier Chair for the Public Understanding of Science & Technology at the University of Bristol. His home page is: |
Telegraph Group Limited endeavours to ensure that the information is correct but does not accept any liability for error or omission.
Users are permitted to copy some material for their personal use, but may not republish any substantial part of the data either on another website or as part of any commercial service without the prior written permission of Telegraph Group Limited.