Arguments — mostly partisan ones — about the relative merits of analog, digital, and mixed-signal integration have echoed since each technology launched, which in the case of analog integration is over a half century ago. No matter the bravado with which various proponents claim the superiority of one domain or the other, when generalities give way to specifics, the answer is always application dependent.
Such an outcome is intuitively satisfying if for no other reason than, were it not so, one technology or another would have fallen into disuse long since. The fact that, more than 50 years on, analog integration is still a critical piece of the electronics industry’s collective toolkit speaks to the technology’s vitality and the degree of innovation that its practitioners continue to bring to bear.
The trajectory of semiconductor processes, particularly over the last 15 years, has dramatically benefited digital integration. Process shrinks have allowed digital designers to pack well over 3 billion transistors on a single die and substantially reduce per-function costs.
But transistor size and count are not the only measures of IC-process value. Though shrinking process nodes get headlines (largely, I suspect, because they refer to a parameter that even those in the popular press can appreciate), they by no means represent the full extent of fabrication-process innovations. Analog circuits still perform many functions that digital circuits either cannot implement or can’t perform as well or as efficiently. Amongst these is a host of system-interface functions that allow systems — including largely digital ones — to connect to our stubbornly analog planet… and selves.
For many devices, IC designers can achieve adequate performance with analog and digital functions cohabiting on a single die, and familiar mixed-signal ICs result. There are other situations, however, for which target system performance requires dedicated analog ICs. These include ultra low-noise, wide bandwidth, high-voltage, and high-current applications.
Even where it is technically feasible to implement an analog function on an IC-fabrication platform otherwise optimized for digital circuits, it may not be economically sensible to do so. Analog functions, which generally do not benefit from device shrinks, may perform perfectly well on 250, 180, or 130nm processes, and they do so at considerably lower cost than if pushed onto 45, 32, or 22nm nodes simply for the sake of increased integration.
Finally, there’s the matter of sourcing flexibility: Analog design has always been a specialized craft and, in the modern day, one that is increasingly so in ways that are, again, often application specific. Highly dense digital ICs also often derive from development programs that depend on specialized expertise. Companies that invest in and grow in-house proficiency in one area often find it challenging to do so in the other in ways that are commercially efficient. There are exceptions of course, but this may be one reason one rarely finds, say, advanced switch-fabric ICs and physical-layer ICs from the same vendor.