Computers

IBM's new 5nm architecture crams 30 billion transistors onto fingernail-sized chip

IBM's new 5nm architecture crams 30 billion transistors onto fingernail-sized chip
Wafers of the new 5 nm chips are tested in pods in a New York facility
Wafers of the new 5 nm chips are tested in pods in a New York facility
View 3 Images
Wafers of the new 5 nm chips are tested in pods in a New York facility
1/3
Wafers of the new 5 nm chips are tested in pods in a New York facility
A scan of IBM Research Alliance's 5 nm transistors, made of stacked silicon nanosheets
2/3
A scan of IBM Research Alliance's 5 nm transistors, made of stacked silicon nanosheets
IBM researcher Nicolas Loubet holds a wafer of the new 5nm chips
3/3
IBM researcher Nicolas Loubet holds a wafer of the new 5nm chips
View gallery - 3 images

The smallest and most advanced chips currently commercially available are made up of transistors with gates about 10 nm long, but IBM has now unveiled plans to cut them in half. To create 5 nm chips, the company is ditching the standard FinFET architecture in favor of a new structure built with a stack of four nanosheets, allowing some 30 billion transistors to be packed onto a chip the size of a fingernail and promising significant gains in power and efficiency.

First coined in the 1970s, Moore's Law was the observation that the number of transistors on a single chip would double every two years. The trend has held up pretty well ever since, but the time frame of the doubling has slowed down a little in recent years. In consumer electronics, 14 nm chips are still stock-standard, but advances from the likes of Intel and Samsung mean that 10 nm versions have started hitting the high-end market.

But the march of technology never stops, and in 2015 IBM unveiled a 7 nm test chip, developed in conjunction with GlobalFoundries and Samsung. This prototype crammed some 20 billion transistors onto a fingernail-sized chip, thanks to some new manufacturing tricks and materials, and they're expected to be rolled out on a commercial scale in about 2019.

Now, the same group of companies has unveiled the next step beyond that. With individual switches just 5 nm in diameter, an extra 10 billion of them can be squeezed onto a chip the same size. While current manufacturing techniques could potentially shrink down to the 5 nm scale, the team instead developed a brand new architecture.

Semiconductors have been made using the FinFET architecture since about 2011. As its name suggests, these transistors are fin-shaped, with three current-carrying channels surrounded by an insulating layer. But, as often happens with technology, this structure is starting to bump up against the limits of how small it can be scaled, and the IBM team says shrinking the fins any further won't do much to improve their performance.

A scan of IBM Research Alliance's 5 nm transistors, made of stacked silicon nanosheets
A scan of IBM Research Alliance's 5 nm transistors, made of stacked silicon nanosheets

Instead, the 5 nm chips are made using stacked silicon nanosheets, which can send signals through four gates at once, instead of FinFET's three. They're created using Extreme Ultraviolet (EUV) lithography, a process that writes patterns on a silicon wafer using a much higher energy wavelength of light than the current technique. That means finer details can be created on the chip, and unlike existing lithography processes, the chips' power and performance can be adjusted continuously during manufacturing.

Compared to the current 10 nm chips, the 5 nm prototypes are capable of improving performance by 40 percent at fixed power, or provide a power saving of 75 percent at matched performance. The development could lead to smaller, more powerful and more efficient devices – but with 10 nm chips only just hitting the market, and 7 nm due for commercial release in 2019, these 5 nm chips are probably still about four years away.

Source: IBM

View gallery - 3 images
6 comments
6 comments
yawood
It always amazes me that Moore's Law could be coined in the 70s and has basically held true all these years. Every time they seem to be hitting the wall another new idea emerges that pushes the wall further out. Incredible.
JanKowalski
yawood
I sometimes think that it's conveniet to manufacturers. Insted of producing a revolutionary processor they produce and sell many generations of processrors that are a little better then previous ones. That makes more MONEY. It's like with high jump champion named Bubka - he didn't jump as high as he could. He was improving previous achievments for 1 centimeter and getting gold medals each time :-))))
KevinMcLelland
Yawood, to add to JanKowalski's thoughts, it's quite possible that these manufacturers, many of whom are government contractors, or perhaps the governments themselves, have already researched and developed all the technology that we will see for the next hundred years, but they only introduce it to the public slowly, in order to main control and a firm grip on the markets. I also don't think that it is just a coincidence that they just happen to find a solution every time they supposedly 'hit a wall'.
Gx Marius
JanKowalski| it seems to be true that...))))
Michael Z. Williamson
Shortly we will achieve Skynet.
BlueOak
IBM still does hardware?
I thought they had long ago run from the competition in hardware...to consulting.
Have they abandoned dark suits and white shirts? ;-)