Keeping Up With Moore's Law

Throughout my long career the scientists and technologists have been successfully battling to keep Moore's Law on track. It's been so reliable that it has proven perfectly acceptable that, if you need to find a way to increase the speed or capacity of a technology product, you only need to plot the trend to the planned release date. If the trend meets the requirement, you're done. It's been very handy for those of us in the business of planning future products; doing nothing to achieve one's goals is nice work if you can get it.

As the technological barriers are progressively hurdled there are also those higher barriers that are imposed by the fundamental laws of physics. These range from light speed - limiting the speed a signal can travel within and between circuits - to quantum mechanics - the soft boundary between traditional electronics laws, including Ohm's Law, and the weird regime of the quantum. Reading this recent article in the New York Times called to mind this latter limitation: in particular that the quantum world is largely discrete and so you run into difficulty when the number of discrete particles drops to a low integer value.

As you improve the precision and accuracy of micro-circuit construction, the circuit density increases at a faster rate. This is simply because a micro-chip is 2-dimensional, and so density increases with the square of the lineal improvement. Unfortunately as we enter the realm of small quantum units, the number of those units decreases at a similar rate. For example (assuming that all other things remain the same, which they rarely do), if you have 100 electrons within a square circuit area that are employed to perform some function, if you double the lineal density you then have only 25 electrons on hand. While this is grossly simplified it does demonstrate the nature of the challenge: as particle counts drop toward one, the electronic laws that are based on mass quantities of particles become fuzzy in the face of the probabilistic nature of each of those particles. Macro-laws smooth out the individual differences of unpredictable particle behaviour.

Going to 3D has been a dream for as long as I can remember. In the case presented in the referenced article it can provide a way to avoid the quantum scaling problem for a little longer. However, 3D circuitry has its challenges. I remember when, long ago now, there was a move to 3D circuit boards. These allowed more circuit density with the discrete components then in use by allowing easier interconnection among the huge number of pin-outs of the many ICs and other, more basic components. There was a limit to this since components still had only two surfaces to work with for the most part (wires/traces, not components, were on the interior layers). In time the interconnection problem was eased by better CAD software and chips that replaced dozens of discrete components, making 3D circuit boards less of a focal point for further progress.

Within a chip, going to 3D has the same interconnection challenge while also having other penalties. Perhaps the worst are yields and heat. With increasing density there is an initial drop in yield, which is the fraction of components that exit manufacturing without disabling flaws. This can only get worse with multiple layers where the quantity of deposition layers and masks grows linearly with the number of layers in the 3D stack. It could be even worse. The benefits have to be great to justify the battle over refining processes to achieve economically acceptable yields.

Heat is already a problem in 2D microchips. It is more acute with microprocessors than memory chips, but it still exists and would need to be dealt with. Every gate generates heat during active operation (such as reading and writing), and heat generated within a 3D stack may be difficult to conduct to the surface for removal. There are of course many avenues open to attack this problem since it is very well understood. The optical method outlines in the article would most likely have less heat to contend with, but with the cost of other difficulties.

Since I am far outside this micro-electronic world I can only watch and wonder how these various difficulties can be overcome. If economical, reliable and, perhaps above all, scalable 3D solutions are discovered we will be firmly back on the Moore's Law trend line. Within some constraints, for a given technology base circuit density could increase up to the cube of lineal density. The third dimension most likely will not scale as well as the other two, yet even a modest achievement would have impressive results.