Moore's Law is the well-known empiracal "law", coined by Intel founder Gordon Moore, that the number of transistors that can be contained in a silicon chip will roughly double every eighteen months. It happens to be a special case of a more general law of technology. The generalization can be stated as follows:
Over time, any technology that enjoys a sufficient market will experience exponential improvement in its ratio of capability to cost.
The market threshold required for a given technology and its doubling time for improvement depend on the nature and complexity of the technology. However, the required market size can be succinctly summarized as "large enough to justify the effort needed to improve the technology".
Likewise, the doubling time is "whatever period of time it takes for the engineers to make it twice as good". The actual time, and the degree to which improvement is achieved by increasing capability vs. decreasing cost, depend on cost and complexity of tools required to improve the technology, and on how much scope for improved capability the laws of physics allow.
Both the doubling time and the threshold market size can (and do) change over time. What we commonly label as individual technologies--e.g., the technology for making integrated circuits or the technology for manufacturing cars--are really complex systems of interacting technologies. At different times in the evolution of a given technology, different constituent technologies will be more or less dominant in setting its capability to cost ratio. As one "tentpole" technology is improved, it eventually ceases to be a "tentpole", and the improvement rate comes to depend on other sub-technologies. Some areas are more resistant to improvement than others, and set different thresholds for the market size needed to justify further R&D. The surprising constancy of Moore's Law itself is now threatened, not by any hard limitations of physical laws, but by the increasing size of investment required to develop the next generation of IC process technology. It seems to be nearing the point that the IC market is simply not large enough to justify the cost of moving quickly to the next level.