Over the last 5-6 years Moore’s Law has slowed down to a trickle, as a result of which the majority of industry analysts have announced the death or the end of the said law. Intel is the only company that still believe in it though, regardless of whether it has helped or not. In a new post published on Medium, titled “A Bright Future for Moore’s Law”, Robert Chau, the Director, Components Research at Intel writes that Moore’s Law has evolved rather than ended and will continue to be a “guiding principle” for the semi-conductor industry.
Moore’s Law demands the doubling of transistors on a chip with each process generation. Make a note of the “doubling of transistors on a chip” part, we’ll explain the reason later. Although this law has held more or less for the last several decades none of the last few processes from TSMC has demonstrated this kind of increase in density. Intel…well, they haven’t produced any new node in the last 5 years. The 10nm process which offers roughly 2-3x as much density compared to the older 14nm was launched more than half a decade after the latter, or in other words four generations later than it was originally expected. That means over four generations the density of Intel’s microprocessors remained the same while the fifth one increased it by a magnitude of 2.4-2.7x.
Not exactly in line with Moore’s Law is it? Now, Intel has conceded that the miniaturization or the increase in chip density will be much slower over the next decade, so it believes that the law will be succeeded by another technology: The MCM or chiplet design, the same thing that Intel called out AMD for, describing its Epyc CPUs as glued together dies.
The most important paragraph from Intel’s post is the following:
I believe they can collectively be categorized into two broad areas: monolithic scaling and system scaling. Monolithic scaling might be referred to as “classic” Moore’s Law scaling, with a focus on reducing transistor feature sizes and operating voltages while increasing transistor performance. System scaling improvements are the gains that help us incorporate new types of heterogeneous processors via advances in chiplets, packaging, and high-bandwidth chip-to-chip interconnect technologies.
What that basically means is that the density of the processors will keep increasing, thanks to the use of a chiplet design, the same one the company called “glued-together desktop dies” a few years earlier. Now let’s rewind back to the definition of Moore’s Law.
It’s the doubling of transistors on a chip with each process generation. The doubling of transistors on a chip, not on a massive processor built using multiple dies. Now, while I do agree with Intel’s core message: It’ll be the chiplet and heterogeneous packaging technologies that will drive the semi-conductor industry in the years to come, but I don’t think that has anything to do with Moore’s Law which is effectively dead.
It’s rather ironic seeing Intel adapt (and embrace) the same technology that it ridiculed a few years back, and that too not for a minor side-project or a single-processor but their entire microprocessor portfolio.