At each new technological step, the hash rate per dollar spent increased by an order of magnitude. FPGAs don't count as a technological leap, because they only incrementally improved the hash rate per dollar of TCO.
Once ASICs become common, I expect the difficulty growth to level off to a steady rate, until the next major technological leap which increases hash rate by 10x per dollar (or more). ASICs 2.0 will probably not be that leap.
I don't agree that you only call GPU and ASIC a technological step.
All new forms of mining are a technological step.
ASICs are the end station in IC design at the moment. Nothing faster than ASICs.
I think that now that we are on ASICS the growth will start alligning with the usual moore's curves of current tech.
Which means about 2x better every 2 years or so.
We will probably see some more spikes when the ASICs go to smaller processes but nothing shocking after that. Untill quantum computers destroy criytography, that is.