Let’s set the record straight. Moore’s Law is not, nor has it ever been, defined as a doubling of transistors on a chip every 18 months. So, let’s wipe the slate clean and start from scratch.
In 1965, Gordon Moore, then director of Fairchild Semiconductor’s R&D Labs, penned his infamous article Cramming more components onto integrated circuits for Electronics Magazine. In that article, he stated that “the complexity for minimum component costs has increased at a rate of roughly a factor of two per year... Certainly over the short term this rate can be expected to continue, if not to increase.”
Now, this couple of sentences alone is not enough to garner Moore’s Law straight away – it needs a little translation. By ‘complexity’ Moore means the number of transistors, or ‘components’, on a single integrated circuit. Easy enough. But the key comes from the term ‘minimum component costs’. Moore noted that the more components you crammed onto a chip, the lower the cost per component. However, there were significant technical challenges to cramming huge numbers (i.e. numbers in the 1000s) of components on a single chip. That meant costs increased rapidly once you started reaching the limits of the manufacturing capabilities.
These two forces worked against each other and determined that there was an optimal level of complexity that gave the highest number of components on a chip for the lowest cost per component. Moore’s observation was that the number of components per chip at this optimal complexity level was doubling every 12 months, and that there was “no reason to believe it will not remain nearly constant for at least 10 years”.
THE OTHER LAWS
However, it only took until the early 1970s before Moore was forced to revise his prediction. As more components were crammed on to the chips, the design of the chips became increasingly more complex, which made manufacturing even more difficult and expensive.
This slowed things down to a 24-month cycle, which became the official formulation of the law in 1975 – the same formulation that remains with us today. In fact, if you chart the number of transistors in all of Intel’s mainstream processors since 1971 you’ll find they fit with uncanny precision to Moore’s 24-month formulation.
If the law was set to an 18-month timescale then we’d already have chips with over 10 billion transistors – something we won’t see until at least 2010. So that’s Moore’s official line. Now let’s look at all the other ‘Moore’s Laws’.
The most common formulation of the law, which arose in the 1980s, is an abstract doubling of ‘computing power’ every 18 months. Given it’s difficult to get a single objective measure of ‘performance’, it’s hard to pin down the accuracy of this prediction. However, it’s not far wrong.
Processor performance measure in millions of instructions per second (MIPS) has doubled roughly every 20 months since 1971. At least that falls somewhere between 18 and 24 months, despite the fact that MIPS aren’t the best way to measure performance. Moore’s Law has also often been interpreted as being a doubling of frequency every 18 months. Sadly, frequency has proven a poor match for transistor density, and we’ve been falling behind even the 24-month curve ever since the 386. So next time you hear the 18-month ‘Law’ quoted, you’ll know what to say: “Do you mean a doubling ever 18 months in complexity of minimum cost components, or doubling of performance as measured in MIPS, or perhaps you meant doubling in frequency?