PORTLAND, Ore. — Every engineer is worried whether Moore's Law (that the density of transistors will double every two years) can be extended forever. So far, merely scaling to smaller sizes has kept Moore's Law in play, but now that we are approaching the atomic scale, many see the handwriting on the wall: When you get down to one atom per memory cell, Moore's Law has to end — or has it?
Many other factors are involved besides size. Scaling is but one hurdle Moore's Law has to clear, according to University of Michigan professor Igor Markov, IEEE Fellow and author of VLSI Physical Design (Springer). In this month's Nature, Markov explores the fundamental limits to computation from manufacturing to energy consumption, physical sizes, algorithms, and design and verification efforts.
"In 1956, Richard Feynman famously pointed out that 'There's plenty of room at the bottom.' This is no longer the case, but there's plenty of energy at the bottom," Markov told EE Times. "A few years ago, the semiconductor industry realized that energy-efficiency can be improved quite a bit, but instead of traditional scaling one must rely on new, clever tricks, including micro-architectural restructuring."
Thus Moore's Law is no longer just about making transistors smaller, but about continuing to increase computational capacity in other ways that face new problems, some of which engineers have never faced before. Multicore parallel processors is not the answer either, because ultimately they face the same problems, according to Markov. The answer is exploiting the limits to Moore's Law that are not so limiting.
Moore's Law has many obstacles, some expressed by fundamental laws of nature (column 1, engineering), but it is also at the mercy of the other factors on the top row, some of which are not as limiting as some think.
(Source: Igor Markov)
"One can go much further, at least in theory," Markov told us. "For example, we don't know how to use nuclear energy for computation. Of course, this is difficult because of additional fundamental limits, such as quantum limits on heat transfer and entropy flow."
In the Nature paper, Markov reviews not only the limiting factors, but also the salient trends that could achieve the goals of faster computations for single-core or parallel processors in different ways than just scaling. In fact, Markov claims that several fundamental limits may not be hard stops at all, possibly indicating new opportunities for emerging technologies, prompting him to make this extensive study of the limits of Moore's Law's limits.