IBM’s 7nm Chip Doubles Performance, Proofs Moore’s Law Through 2018

Moore’s Law is one of those wonders of modern life that we all take for granted, like grocery stores and anesthesia dentistry.

For 50 years now, computer processors have been doubling their performance. per dollar per square centimeter every 1-2 years. This exponential trend has taken us from 500 flops of ENIAC (floating point operations per second) to 54 petaflops for the most powerful supercomputer to date, Tianhe-2 . This is about ten trillion times more than in a century. It’s incredible for everyone.

This achievement happened so reliably, for so long, that it became a mundane truth in computing.

We take it for granted.

That’s why it’s so scary that things might stop in the near future. A number of fundamental physical limitations are converging to stop the development of traditional silicon computer chips. Although there are theoretical computer technologies which may solve some of these problems, the fact remains that progress is currently slowing down. The days of exponentially improving computers may be coming to an end.

But not quite yet.

A new breakthrough from IBM shows that Moore’s Law still has legs. A research team led by the company has demonstrated a prototype processor with transistor components just 7 nanometers wide. That’s half the size (and four times the performance) of current 14nm technology, pushing the demise of Moore’s Law until at least 2018.

So how was this breakthrough achieved? And when can we expect to see this technology in real devices?

Old Atoms, New Tricks

The new prototype is not a production chip, but it was made using commercially scalable technologies that could be on the market in the next few years (there is a rumor that IBM would like this chip to be introduced in 2017-2018. A prototype is a product IBM/SUNY, an IMB research lab that collaborated with the State University of New York A number of companies and research groups collaborated on this project, including SAMSUNG and Global Foundries, a company that IBM is paying an estimated $1.3 billion to take over from its unprofitable chip manufacturing wing.

In essence, the IBM research team did two key improvements, that made this possible: the development of a better material and the development of a better etching process. Each of them overcomes a serious barrier to the development of more dense processors. Let’s look at each of them in turn.

best material

One of the barriers to smaller transistors is simply the reduction in the number of atoms. The 7nm transistor has components that are only about 35 silicon atoms in size. For current to flow, electrons must physically jump from the orbit of one atom to the orbit of another. In a pure silicon wafer, as is traditionally used, it is difficult or impossible to obtain sufficient current to pass through such a small number of atoms.

To solve this problem, IBM had to abandon pure silicon in favor of using an alloy of silicon and germanium. This has a key advantage: it increases the so-called «electron mobility» — the ability of electrons to pass through the material. Silicon is starting to perform poorly at 10nm, which is one of the reasons 10nm processor efforts have stalled. The addition of germanium jumps over this barrier.

More accurate etching

There is also the question of how you actually form such tiny objects. How computer processors work are produced using extremely powerful lasers as well as various optics and stencils to carve tiny features. The limitation here is the wavelength of the light, which puts a limit on how well we can etch elements.

For a long time, chip production has stabilized with the use of argon-fluoride laser with a wavelength of 193 nm. You may notice that this is slightly larger than the 14nm features we engraved. Fortunately, wavelength is not a hard limit on resolution. You can use interference and other techniques to achieve greater accuracy. However, the chip makers have run out of their smart ideas and now a major change is needed.