The end of Moore’s Law

Zilog Z80A processor
Z80A processor, the one I used first in my Sinclair ZX81

As the first post in this series about the future of the computer we look at the basic hardware component: the processor. The development of processors (and chips in general) follows about the most well-known law in computers: Moore’s law. Gordon Moore was co-founder of Intel when he noticed the number of transistors they put on a chip doubled every other year. Since he noticed this in 1965 and 50 years have passed since then, processors nowadays have 2 25 (33,5 million) times more transistors than back then.

Historically, this law is extremely accurate. It almost feels like it’s a self-fulfilling prophecy and that manufacturers aim to reach it. One very explicit example is Intel with its tick-tock development. Every other year, Intel starts using another production process. This works really well for them: when they started tick-tock in 2006, AMD was a tough competitor. A few ticks and tocks later, Intel was both the technological as well as the market leader.

Intel announces new processors yearly. The first step, in 2006, was a tick: the existing architecture was put on another process, which was 65nm back then. The next year, the process stays the same and Intel improves the processor architecture. This keeps a clean division between the process and architectural improvements. Intel is able to avoid typical problems in the production of new processors this way. When another manufacturer updates both its process as well as its architecture at the same time, it can be hard to figure out if a bug is due to the process or the architecture. For example, TI’s TMS320C62 DSP processors experienced serious delays because of this.

When Intel ticks, they shrink the process with a factor of around √2. So on the same surface, they exactly double the number of transistors every other year.Self-fulfilling prophecy!

FinFET vergroot het contactoppervlak van gates.
FinFET is één manier om het contactoppervlak van gates te vergroten bij het verkleinen van het productieproces.

The production of smaller components has its own challenges. For example, the contact surface of the transistor gates becomes smaller, which makes it harder for enough electrons to pass. To solve that, processors with 25-20nm processes or smaller have gates that have a “fin” standing upright. The gate is wrapped over it, giving a larger contact surface. This kind of technique is an impressive workaround, but physically there is a hard limit on how small we can build transistors. And that limit is easy to calculate.

The most recent Intel Broadwell processors use a 14nm process. Graphene, a sheet of carbon molecules exactly one atom thick, is 0,345 nm thick (and silicon is a larger atom than carbon). Even if we assume that Intel succeeds in shrinking every other year by a factor of √2, then after 10 ticks the process is smaller than one atom. We can safely state that the end of an era is near.

In the next posts, I will cover other aspects of Moore’s Law: heat, production costs, and more. Did you hear about a material that can replace silicon? Or do you know how processors can evolve after Moore’s Law? Let me know in the comments below!