The End (?) of Moore’s Law
There has been much written about Moore’s Law, and how it’s demise is just around the corner.
Of course, for the past 20 years, it has failed to fail. Circuit density has indeed continued to double every 18 months or so. It has always been regarded that physics would lead to the ultimate demise of Moore’s Law, as quantum effects began to infect circuit designs. Indeed, as we approach the 65 nm feature size, quantum effects are being observed. Chip designers, however, think they’ve got the quantum effects licked.
I have an alternate theory for what’s going to stop Moore’s law. Economics.
You’re no doubt laughing right now. Go ahead, get it out of your system. I’ll wait.
OK, better now?
As I was saying: economics will be the undoing of Moore’s law. It now costs so much to build a new plant to do production at the next feature size (node), that it becomes more and more difficult for the plants to pay for themselves. We’ve got the twin bugbears of power consumption and heat dissipation as well, and the costs for dealing with them.
I suspect that research will continue, and there will be niche products at ever smaller nodes (so Moore’s law won’t actually be dead), but mass-produced products are going to have to find more innovative ways to squeeze more performance out of chips rather than relying on the “shrink and ramp up the clock” method that has been going on for 30 years.
We’ve already seen the beginning - dual-core CPUs from AMD and Intel, Sony’s Cell architecture, the multi-core chip for the XBox360.
I suspect that with technologies like PCI-Express and HyperTransport that distributing computing over multiple packages will come back in to vogue, allowing for the distribution of power consumption (the new PentiumD 840 draws up to 125A on it’s 1.45v main power line), and heat dissipation (that chip can dissipate well over 100W at full operating load).
Intel has already demoed a CPU built on the 65 nm node, and heat concentration is even more problematic than at the 90 nm node. They had to go to 90 because clock speeds couldn’t go higher at 130. But it seems that the point of diminishing returns has been hit.
I suspect that if there are major performance gains to be had, they lie in either new semiconductor technologies (Bipolar hit the wall a long time ago, maybe CMOS time is nearly up), new design strategies, or even new materials.