IBM World's Smallest Processor 7nm SiGe chips
How did IBM get a7 nanometer processor when Intel's having problems making anything smaller than 10 nano meters?
Creating a working 7nm chip required moving past pure silicon, IBM revealed. IBM—working with GlobalFoundries, Samsung, SUNY Polytechnic Institute, and others—carved the transistor channels out of silicon-germanium (SiGe) alloy in order to improve electron mobility at such a small scale. Intel has also said 10nm will be last gasp for pure silicon chips
Intel’s 14-nanometer Broadwell chips suffered from lengthy delays, stuttering Intel’s vaunted tick-tock manufacturing schedule. TSMC, the company that manufactures graphics processors for AMD and Nvidia, has been stuck at the 28nm node for years now. Intel plans to push into 10nm in 2017, but IBM’s looking beyond that, and just revealed the world’s first working 7nm processor—but it took some pretty exotic manufacturing to get there.
But that’s not even the most impressive number. IBM says that when the industry embraces 7nm manufacturing techniques, processors will be able to be stuffed with an incredible 20 billion transistors. By comparison, Intel’s new Broadwell-U processors pack “only” 1.9 billion transistors.
IBM and co. had to turn to cutting-edge lithography techniques to etch features onto the chip. The companies utilized extreme ultraviolet lithography, which Intel has also been investing heavily in for years now. The details behind EUV get complicated, but essentially, it’s a beam of light with a far narrower wavelength (read: width) than current lithography tools. The benefits of moving to a smaller feature-etching tool when working on a chip with 7nm components is obvious
EUV lithography is an more interesting innovation. Basically, as chip features get smaller, you need a narrower beam of light to etch those features accurately, or you need to use multiple patterning (which we won't go into here). The current state of the art for lithography is a 193nm ArF (argon fluoride) laser; that is, the wavelength is 193nm wide. Complex optics and multiple painstaking steps are required to etch 14nm features using a 193nm light source. EUV has a wavelength of just 13.5nm, which will handily take us down into the sub-10nm realm, but so far it has proven very difficult and expensive to deploy commercially (it has been just around the corner for quite a few years now).
The 7nm SiGe chips are nowhere near production-ready, but when they’re cleared for commercial use around 2017, IBM says they’ll result in “at least a 50 percent power/performance improvement for next generation… systems” on account of all those process improvements.
source: Ars Technica
Creating a working 7nm chip required moving past pure silicon, IBM revealed. IBM—working with GlobalFoundries, Samsung, SUNY Polytechnic Institute, and others—carved the transistor channels out of silicon-germanium (SiGe) alloy in order to improve electron mobility at such a small scale. Intel has also said 10nm will be last gasp for pure silicon chips
Intel’s 14-nanometer Broadwell chips suffered from lengthy delays, stuttering Intel’s vaunted tick-tock manufacturing schedule. TSMC, the company that manufactures graphics processors for AMD and Nvidia, has been stuck at the 28nm node for years now. Intel plans to push into 10nm in 2017, but IBM’s looking beyond that, and just revealed the world’s first working 7nm processor—but it took some pretty exotic manufacturing to get there.
But that’s not even the most impressive number. IBM says that when the industry embraces 7nm manufacturing techniques, processors will be able to be stuffed with an incredible 20 billion transistors. By comparison, Intel’s new Broadwell-U processors pack “only” 1.9 billion transistors.
IBM and co. had to turn to cutting-edge lithography techniques to etch features onto the chip. The companies utilized extreme ultraviolet lithography, which Intel has also been investing heavily in for years now. The details behind EUV get complicated, but essentially, it’s a beam of light with a far narrower wavelength (read: width) than current lithography tools. The benefits of moving to a smaller feature-etching tool when working on a chip with 7nm components is obvious
EUV lithography is an more interesting innovation. Basically, as chip features get smaller, you need a narrower beam of light to etch those features accurately, or you need to use multiple patterning (which we won't go into here). The current state of the art for lithography is a 193nm ArF (argon fluoride) laser; that is, the wavelength is 193nm wide. Complex optics and multiple painstaking steps are required to etch 14nm features using a 193nm light source. EUV has a wavelength of just 13.5nm, which will handily take us down into the sub-10nm realm, but so far it has proven very difficult and expensive to deploy commercially (it has been just around the corner for quite a few years now).
The 7nm SiGe chips are nowhere near production-ready, but when they’re cleared for commercial use around 2017, IBM says they’ll result in “at least a 50 percent power/performance improvement for next generation… systems” on account of all those process improvements.
source: Ars Technica
No comments
Post a Comment