Since the 1960’s CPU transistor sizes have been steadily shrinking over time.
Since the 1960’s the number of transistors on a Central Processing Unit have steadily increased from 2,300 on Intel’s first microprocessor the 4004, to more than 5 billion on Intel’s 62-Core Xeon Phi, released in 2012.

Moore’s Law
According to Moore’s Law, the number of transistors on an integrated circuit doubles every 18-24 months. Although Moore’s Law isn’t a law of the physical sciences, it is an observation by Gordon Moore made in the 1960’s. Transistor count is the most used method of measure for the complexity of integrated circuits. Intel’s Quad-Core Itanium Tukwila released in 2010 had 2 billion transistors on it’s die. As impressive as this is, Moore’s Law cannot continue indefinitely without modification. The reason for this is the laws of physics, that as the size of transistors shrinks and the number of transistors on a CPU die increases, transistors will eventually reach the limits of atomic sizes. At this size, silicon becomes unsuitable as a material to build integrated circuits out of, due to quantum tunneling and other factors. CPU die sizes will either have to get larger to fit more transistors (abet smaller transistors) and or increase the number of layers for each die.
Moore’s Law is a model of exponential growth, and as such exponential growth is the fastest model of growth.
Whether 100 billion or even 1 trillion transistors may fit on a single chip in the future, will most likely depend on making integrated circuits out of other materials than silicon and an ever increasing die sizes to facilitate such a number of transistors. If that doesn’t happen then Moore’s Law will certainly end.

Sources

Instruments , National. "Impact of Latest Multicore Processors on PXI-Based Test and Measurement Applications - Developer Zone - National Instruments." National Instruments Netherlands. 06 Jul 2012. Web. 24 Mar 2013. .