When I began my engineering courses five years ago, it was still unclear to me where the More Moore vs. More than Moore debate would go. At the time, and with as little experience and knowledge I had gained at that point in the ways of processes and device physics, I was convinced that somehow the silicon industry would continue to find ways to shrink and shrink the smallest feature size until we could somehow place each individual molecule exactly where we wanted it. Imagine the possibilities! If we could place every molecule individually in mass production, the kind of repeatability in mechanical and electrical design that we could achieve would be astounding!
Don’t get me wrong, I still have hopes that someday we could do this en masse, but with my more educated view from here I don’t think that’s happening any time in the next 10 years. (Please feel free to prove me wrong!)
It wasn’t until I took a process engineering elective class that I learned the difficulties that we are already having at a mere 16nm feature size. When you get down that small, you already have to deal with wave/particle theory. Mask sets need to have funky cutouts to make corners form correctly. Noise becomes a problem the lower your Vt voltage drops with these smaller feature sizes. Given this was just an undergraduate class, I didn’t learn as much about FinFETs and other creative process technologies being developed to work around these rising problems, but I got enough to understand that these are, for the most part, relegated to research laboratories for the time being.
Moore’s Law has shifted from a historical pattern to a self-fulfilling prophecy. As it gets harder and harder to cram more transistors into the same space, we’ve gotten creative with how we continue to fulfill the law of the industry. But the bubble must burst eventually. Feature size and transistors as they are defined today can only get so small before you have to break the laws of physics.
This is why I think this time is so exciting. Some companies are charging towards the smallest possible feature size. Others are venturing into other options. There has been a slow shift from “how many transistors can we cram on this thing?” to “how many features can we cram into this package?” It makes the possibilities almost limitless, and the approaches from companies across the industry are just as diverse.
For even more evidence of this, we need only look at the industries our devices go into. Systems are becoming more complex each day, being asked to take on more and more. We now demand our smartphones to do just about everything from check in on Facebook to sending a snap to your best friend on SnapChat, look up any vast number of obscure facts, monitor health metrics, play music, turn on the AC, monitor your in-home security camera, check what groceries are in the fridge while you’re at the grocery store… The number of things you can do with your smartphone is growing. Much of this growth of abilities is thanks to the growing size in the ecosystem in which we use our phones, the Internet of Things. I think more companies are recognizing that in order to get ahead, you have to do more than just build a better product. You have to be strategic in how you make it better.
You can't just sell one component in a system
You can’t just sell one component in a system either. You have to have knowledge of the entire ecosystem that you’ll be functioning in. In an industry where more and more start-ups are creating the visions of tomorrow, full solutions are going to be more valuable than just a power IC. Take a look at the Apple Watch for example. Apple took the System-in-Package approach just to fit as much processing and sensing power as they could into one compact elegant block of technology. The Apple Watch probably has more processing power than cell phones did just 10 years ago, yet it is less than half the size. And it has more features than most did.
But is Moore’s Law really collapsing per se, or is it simply coming to its natural end? Gordon Moore himself cited the physical limitations and that eventually this trend will stall. Current “accepted” predictions put the end of the trend less than 10 years away in 2025. Even International Technology Roadmap for Semiconductors (ITRS 2.0) met for the last time ever, opting to transition the industry’s focus to other means of increasing computing power. So when Moore’s Law finally ceases to exist, what will drive growth?
I happened to dive into the Internet Rabbit Hole chasing the answer and stumbled across Bell’s Law of Computer Classes. Rather than predict the complexity of new processors and transistor counts, Bell’s Law keeps things a little more open to interpretation. It describes the creation, evolution, and destruction of different “classes” of computer systems. This includes everything from mainframes to personal computers to wireless sensor networks. It describes the evolution of technology, how one innovation leads to the next. Bell estimates that a new industry for a new computer class emerges roughly every decade.
This isn’t to say that these industries last only one decade, but that at the very least a new one emerges and blossoms. It’s a law that I think more accurately describes the trend of technology growth in a way that doesn’t limit growth to one specific aspect. As engineers we are tasked with finding new ways to solve challenging problems with the tools we already have in our arsenal. This is how the IoT emerged, and how the concept of the smartphone became mainstream.
What new application will bloom in 2020? Who knows! But we can try to dream it!