Most of the IC industry has prospered, in large part, by a combination of process shrinks and higher levels of functional integration per die. I say “most” because, once again, the “analog” niche marches to a different tune, as the saying goes.
My evidence is this: In the past few months, vendors have announced a large number of single-function, basic building-block ICs: current-sense amplifiers, op amps, in amps (instrumentation amplifiers), comparators, ADCs and DACs, references, switches, drivers, and more. In each case, the new product either pushes performance of one key parameter to a higher level, or has a carefully balanced set of values for the most-critical parameters.
What I especially like about building-block ICs that they don't define or straitjacket your overall system architecture. Instead, they enable you to achieve the performance you need in various functional areas, while you develop the set of function sand channels you need to match your application. You don't pay for internal functions you don't need–and, in many cases, the real “cost” of such excess functionality is not the dollar price itself, but the added footprint, support components, interface issues, initialization, unexpected interactions, or debug headaches.
These building-block ICs have one other virtue: they let the IC designers and vendors focus on doing the best they can in a given area, using the process and design tools they have. While every IC is a unavoidable compromise–the “perfect” part doesn't exist–at least here, the tradeoffs are more sharply apparent, and the IC can optimized for its primary task. The compromises are fewer, which allows the IC chooser/user to decide more easily which tradeoffs make the most sense during the part-selection stage of the design cycle.♦