There is a TV show called What Not to Wear that consists of some fashionistas giving advice to people on how they ought to dress. I once heard that the hipsters only wear black because they haven’t found anything darker yet, but that’s another story.
All the talk about analog integration in the “Integration Nation” section of this site got me thinking about what you should not try to integrate. Everything you add to a chip adds risk of a re-spin with the attendant cost, schedule, and other pressures. So let’s look at the analog functions you might consider integrating.
Preamps/signal conditioning: Not a bad choice, as long as there’s not too much digital stuff on the chip that might get into the small-signal, high-gain stages.
A/D and D/A converters: Sure... but make sure that the performance of this block is not the big selling point of the chip/SOC, since integrated converters often are a little behind the state-of-the-art in performance.
Phase-locked loops (PLLs): As long as you pay attention to keeping other signals away so the loop locks to the reference you want, OK. Multiple PLLs on a chip would be asking for trouble, though.
Oscillators: See comment about PLLs above. Two oscillators in proximity to one another will inevitably couple to each other and result in one of them being way off.
RF: This one is tough, since an RF amp is by nature pretty wideband and will gladly amplify any extraneous noise from switching circuits of any sort on the same chip. In addition, if the RF stage needs some inductors to tune it, we now live in an age where gain elements like transistors are small and cheap and inductors are big and expensive (note to younger engineers: it was the other way around for a hundred years or so). In some applications, a little extra noise can be tolerated since its effect will be diminished through the miracle of spread-spectrum and correlation.
Power management: “Yes, but...” A few tame linear LDOs with moderate output current are a harmless enough add-on. Put a few boost-mode switchers, or some fire-breathing multi-Ampere output LDOs, and watch the fun begin when the switching noise gets into everything else, or the linear regulator glows red-hot, warming up all the other circuits and making them drift all over the map.
Precision references: They’re small, and simple, but if there is too much other stuff on the chip creating thermal headaches, they’re not going to deliver the precision you expect.
Digital stuff: Well, the big trend in SOCs (System on a Chip) is to start with the digital part, then add the analog portion like I/O, power, whatever. Some types of converters, delta-sigma, autocalibrating, etc., depend on the digital stuff for their performance, so that’s an example of a good use of analog/digital integration. But some of the other analog functions are hard to do in the preferred digital CMOS processes. And you don’t want to hear a manager or customer saying something like “You’re telling me that I have to re-spin this multi-million transistor, 99.9% digital and working ASIC because your stupid little (op amp, reference, regulator, fill-in-the-blank) on the chip has a problem and ‘needs a tweak’?!?”
Biggest thing to watch out for in my view would be anything that dissipates power. Using the linear regulator thread, putting several together becomes that glowing device which shouldn't. The precision regulator requires some current to help in regulation, but thermal effects can start to be a challenge if things get too warm.
Doug, Thanks for the overview and highlighting the complexities involved in integrating modules...
A few more modules to make life more challenging is to add the high speed modules like the USB, SerDes its associated complex IO's, Noise cancellation modules with high dB of rejection and PSR constraints...
When it comes to this month’s ISSCC show, there are plenty of sections dedicated to solving issues with integrated analog technology. Here, our blogger looks at the sessions that offer new insight to difficult problems.
The International Solid State Circuits Conference kicks off February 9. While the mass media will focus on the biggest memory, faster processor, and smallest geometries reported, I prefer to look at the analog stuff.