When it comes to microchips — the miniature machines that run so much of our lives — innovation has driven us farther than we ever thought possible. The rate of semiconductor innovation is slowing, however, the way traffic on a freeway bottlenecks as lanes diminish.
When Gordon Moore, the cofounder of Intel, asserted nearly 50 years ago that the number of transistors on integrated circuits would double roughly every two years, he was charting a path for the company that would dominate the industry — and dictate the direction of technology — for decades. Lately, much ado has been made of the coming end of Moore's Law, although the law was never an inflexible law of physics but more like a rule, a likely way forward.
Intel's superior manufacturing prowess in the early years of computing technology allowed the company to move the goal posts for everyone: Every 18 to 24 months, Intel would double something — performance, storage — and the rest of the industry would scramble to keep up, locking in Intel's dominance. Essentially, Moore created the conditions for his law to hold true — until today, as the laws of physics limit engineers' ability to keep squeezing more circuits onto a chip.
There, however, is a possible "off-ramp" to Moore's Law that offers new hope — not just of continued improvement in semiconductors but of an open, competitive playing field unlike the Intel-dominated landscape of the past half century.
To understand where the industry is today and where innovation is headed, it's helpful to think of the microchip as a metropolitan area and its components as buildings.
Decades of innovation have made the components of a microchip smaller and smaller. Yet chips have grown larger as more and more components are packed onto them to meet increased computing needs — making the interconnections between each minuscule part more spread out.
That "sprawl" is like the suburbs around a city. The same problems that apply to a sprawling metropolitan area apply to the microchip: getting from point A to point B requires increasing time and energy just like driving a congested freeway from a suburban home to a job downtown does. Information travels across bigger microchips with less efficiency at slower speeds, while consuming more power.
We can't increase the surface area of microchips much more without running into those problems, and we're getting closer every day to the limits of how small we can make components. So what next?
The trend in urban development today is to build up, bringing people back into city centers and transforming suburbs into functioning city units where jobs, shopping, and homes are as closely connected as possible. That idea applies to microchips in the form of three-dimensional interconnect. It's the off-ramp to Moore's Law: In 3D interconnect, engineers stack wafers like the floors of a skyscraper in extreme miniature, with vertical connections (think elevators communicating between floors) in addition to traditional horizontal links.
What's especially compelling about this off-ramp, this technological detour, is that it provides an opportunity for companies besides the two usual suspects — Intel and AMD — to compete. Intel has lost its once singular ability to move the goal posts on innovation.
Although both companies are working towards 3D interconnect, neither has figured out how to volume-manufacture the technology. Two other players to watch closely are Samsung and TSMC. It's anybody's race to win.
There are still technical challenges to resolve, but commercial-scale manufacturing could come as soon as three to five years from now, speeding the pace of innovation again and enabling new ways for the technology industry to be creative.