In my experience, mixed-signal ICs with high levels of integration fall into two categories. The first is the chip designed for a very specific application, often just for one end product. The other category includes ICs for general use in many different end products with similar needs.
For the first category, consider the PMICs for smartphones that my current employer has been quite successful with. Cleverness is required in integrating many functions into a small number of chips in a very short period of time. I do not work in this part of the company, and frankly, was shocked to recently find out how many voltage regulators are integrated into these devices — thus the request at the beginning of this blog.
It takes no small amount of engineering cleverness to integrate so many regulators with other, often quite dissimilar functions, and to do so economically and quickly. Frankly, I am amazed that these guys pull this off at all. And they do so regularly.
A different kind of cleverness is needed in developing integrated chips for a range of applications and/or end products. In this case, there are multiple end targets with different requirements. As the level of integration increases, the breadth of end targets narrow but, qualitatively at least, the problem is the same; there are many target end products for the chip to be designed into… if the developer is clever enough.
Let's go back to the dark ages (the early 90s) when I was designing microcontroller-based consumer products at a startup. If we were a larger company, we could have had a customized version of a microcontroller made with the peripherals and amounts of RAM and ROM (no on-chip flash yet) that suited our needs.
Instead, I had to look at the microcontrollers available, with each family having parts with different levels of integration. Some had nearly everything we needed, but also had a lot of what we did not need and thus cost too much. I had to balance the cost of implementing functions external to the µC vs. the cost of having too much integrated. This is an interesting trade-off to have to make, especially when a few cents difference can make or break a product.
The developer of more general integrated products has to run this thinking in reverse while predicting the future. And, as the wise old man Yogi Berra once said, “It's hard to make predictions, especially about the future.” This developer (not Yogi) has to look at a range of yet-to-be developed products and determine the optimal combination of features. If he integrates too much or too little, the product will not be successful. Like the third bed in Goldilocks and the Three Bears , it has to be just right.
Success with this type of product requires a good understanding of the markets for the end products that such chips will go into as well as a good understanding of the available technologies at hand to implement these chips. Once again, integration calls on the chip developer to not only know his business but his customer's business as well.