RAM price increase (in DRAM bits/$) follows an exponential trend, like Moore's Law for integrated circuits, and has been fairly stable the last decade.
source
There's an interesting relationship with feature size
source
I don't know much about RAM works, but I know a fair bit about semiconductors. Current MOSFET technology involves a source, a gate and a drain and current gate lengths (about 29 nm) are getting very close to a fundamental limit where quantum tunnelling will overtake the effects of thermal fluctuation and signal-to-noise ratio will increase as feature size is decreased (this limit is at about 22nm, according to my calculations. I can post this work if anyone is interested, it didn't make the final edit of my report). Eventually, the noise from quantum tunnelling will be equal to the actual signal of the electrons travelling from the source to the drain (a signal-to-noise ratio of 0.5) and this CMOS device will be rendered utterly useless.
This limit will be reached very soon (possibly by 2020), and is inevitable for silicon-based semiconductor technology and will certainly affect the improvement of RAM. As a result, RAM will continue to get more expensive but will not increase in speed or information density at the same exponential rate - until a new paradigm is reached. Fortunately, there are a variety of
spintronic technologies that may end up saving the day.