There are innovations which feed back on themselves, usually involving greater precision or density. Microprocessors are the cononical case, where reducing the size of transistors reduces distances, energy requirements, heat management, and increases (to the square) component density enabling more capacity in the same volume.
The picture becomes rapidly less impressive when we look at where that processing goes, in many cases much of it is consumed with processing that's not directly related to useful output, though one area that has benefitted greatly has been usability: a computer user used to be a very highly skilled and trained position, now infants and those with severe cognitive or other impairments can make productive use of computers --- a lot of processing power goes into simply expanding the potential user-base.
(This isn't a slight or dig at who computer users are at present, though I admit some frustrations with how this transforms mass-market platforms, devices, and software away from power-user capabilities and interests.)
Network-dynamics systems (which microprocessors arguably are) can also see similar types of dynamics, at least over a range of scale.
In many other domains though, the engineering possibility frontier is one of sharply diminishing returns, after even only a fairly trivially small amount of initial progress. Occasionally there are bursts of progress, often as two separte areas of technological develoopment are integrated together, but even that often provides only a brief period of high returns.
What I see lacking is even a rudimentary model of what domains within technology do and do not afford for accelerating returns to scale, or what the bounds and limits of the exceptions might be, or what the actual delivered net benefit of such accelerating returns might be.