I often like to think about and question so-called “conventional wisdom,” since so much of it is built on statements or facts that are no longer true, or perhaps were never really true — but were repeated so often they just seemed to be accepted as true.
One of the many such chunks of conventional wisdom is that designing and fabricating an analog IC is in many ways very different than a digital IC. To cite a few examples:
- The tools just aren’t as good; lots of designer experience is needed to balance the tradeoffs;
- Processes are quite different and often specialized;
- Feature size and die size are not driving factors;
- It's not about how many active devices you can cram onto the die, since many analog ICs only need a few (some leading analog ICs have about a dozen active devices);
- Thermal issues (drift, compensation, balance), and matching of key structures (such as gain resistors) are major design concerns;
- Packaging is smaller, and pin (lead) count is usually on the lower side, compared to digital ICs;
- Test fixtures and issues are different.
Were these statements true? In general, yes. There's lots of well-documented material and experience to verify them, no doubt about that.
But are they still true, and if so, to what extent? What made me think about this was a recent interview in EETimes with Louis DiNardo, the new CEO at Exar. He maintains that much of analog design is no longer so unique, that foundries can do a lot of the fab (no need for in-house facilities), and the tools have improved to where the “black magic” aspects of analog IC design are less of a factor.(See: Exar CEO Talks Analog, China, IoT.)
Is he right? Or is he just trying to rationalize Exar's technology and business model?
My view is this: As in most technical discussions, the answer is not black and white. There are shades of grey (but not quite 50 of them, thank goodness!), which depend on what you are trying to do. For a basic, moderate-performance analog circuit, he may have valid points.
But when you get to power-related ICs and regulators (even at low-power levels), low-noise amplifiers, RF PAs, higher-resolution converters, higher-speed amplifiers and converters, medium/high speed drivers, or… well, the list goes on, there are lots of cases where the IC design has lots of tricky tradeoffs, quirks, and so-called secret sauce.
The reality is that when you add up the segments that these more-challenging analog ICs represent, they add up to a significant part of the market, if not a majority (it also depends if you count by unit volume versus revenue, too).
For so-called commodity parts (sometimes derided as “low-hanging fruit”), DiNardo may be right. But for the many specialized or higher-performance analog-centric ICs, which truly differentiate a system's overall performance and thus make it competitive or distinctive to customers, I don't think so.
What's your sense of the situation? Is analog-IC design still so different than digital-IC design, or is it mostly a myth that the analog companies want to sustain?