One of the big questions at DAC this year was when would Moore's Law end. Yes, we have heard this prediction so many times and at each node, some new piece of technology comes along, a new process, a new material -- something happens to save the day. But this year there was almost desperation in the voice of many and the reasons given for its end are diversifying. I heard reasons that ranged from economic, to physics, to practical limitations associated with getting new devices (if they exist) ready in time for production when the crunch comes. I heard people saying that this is being brought up as a potential national security crisis. Others seem to still be happy saying that someone will fix the problems and we will all be fine.
Unfortunately there is no equivalent of Moore's Law for analog. I am not saying that analog densities have not been getting higher, but it doesn't appear as if there is a unifying set of issues and challenges that the entire industry is trying to solve, almost collectively. Maybe that is just because I am not in the in-crowd, but as I Google information related to analog, I find so many conflicting pieces of information. Even within the same company, there can be highly disparate viewpoints, but the companies are not important here and I am not picking on anyone.
Let me give one example of a disagreement. I will start with a keynote given by Joachim Kunkel, general manager of the Synopsys Solutions Group. He said that adoption of new technology nodes has been fairly consistent up to the 32/28nm nodes, but now there is a slowdown. Contrary to that, Wally Rhines, CEO of Mentor Graphics, said at DAC this year that "If you look at the last four or five process technology nodes, you will note that the pace of adoption has accelerated." So which is it?
Perhaps they are both right. Perhaps it depends on the types of design you are doing. While Rhines was looking at the primary SoC market, Kunkel was talking about mixed-signal designs and in particular the difficulties being presented for the analog portions as they migrate from 45 to 28 and from 28 down to 20nm.
So why do we have this separating of the ways? Kunkel believes that it is because of the rules that are being defined at the finer geometries, rules that are perhaps more difficult for analog circuits to adhere to than digital. For example, at 28nm double patterning becomes necessary and with it more stringent rules regarding poly and metal densities that must be kept uniform across the die.
For digital this may just mean adding dummies and a little bit of area goes to waste, but for analog, it is imperative that matched devices are contained within areas of the chip that have consistent densities -- otherwise their parameters will drift from each other. It appears that any separation between digital and analog fundamentally comes down to the way in which the rules can be created, understood, incorporated into designs, and automated. So perhaps this shows that more investment is needed in analog tools if those portions of the design are not going to be the thing holding us back from getting to the edge of Moore's Law and seeing what lies beyond.
Do you believe that adoption of new process technologies is slowing down for analog while speeding up for digital? What do we need to get analog back on track?