Hacker News Reader

Comment by LatteLazy | Hacker News Reader

7nm is about 35 times the diameter of a silicon atom.

7mm is so called 'commercial name' nothing physical.

Million transistors per square millimeter (MTr/mm²) is better comparison metric than the commercial name for the process. Here is handy chart I copied from https://www.techcenturion.com/7nm-10nm-14nm-fabrication:

    Tech Node name  (MTr/mm²)

    Intel 7nm       (2??) 
    TSMC 5nm EUV    171.3
    TSMC 7nm+ EUV   115.8
    Intel 10nm      100.8
    TSMC 7nm Mobile 96.5
    Samsung 7nm EUV 95.3
    TSMC 7nm HPC    66.7
    Samsung 8nm     61.2
    TSMC 10nm       60.3
    Samsung 10nm    51.8
    Intel 14nm      43.5
    GF 12nm         36.7
    TSMC 12nm       33.8
    Samsung/GF 14nm 32.5
    TSMC 16nm       28.2

It's worth pointing out that the value for Intel's 10nm is on its high-density libraries, not its high-performance libraries, which are less dense.

"Libraries"?

There's a lot of detail in a really in-depth (but very interesting) AnandTech article about Intel's 10nm process [0]. The basic idea is Intel have multiple distinct sets of logical unit designs at different densities which have different characteristics. It's like having multiple software libraries that do the same thing but have different performance etc. except it's a library of silicon building blocks. You trade-off speed with density (highest performance library uses the least dense circuits). Each chip's design can be made up of bits of silicon from these different libraries. Different densities also have different power consumption, so in cases it makes sense to build low power chips that have small high performance, power sucking parts that can be used sparingly.

[0]: https://www.anandtech.com/show/13405/intel-10nm-cannon-lake-...

Libraries of standard cells.

https://en.wikipedia.org/wiki/Standard_cell

Are there any current products one can buy, regardless of expense, using Intel's 10nm?

I know there was one product, the Core i3-8121U, a very middling performance chip which saw an extremely limited release.

But it was discontinued.

10nm Ice Lake laptop chips are currently shipping in systems from Dell, HP, Lenovo, etc.

Well shiver me timbers, they've been out for weeks!

That processor solely exists as PR stunt to satisfy investors. It has a faulty GPU, lower clock speed, higher temp, and should be considered as a lab sample instead. You can take a look at Charlie's article about it[1]

[1]:https://semiaccurate.com/2018/05/29/is-intels-upcoming-10nm-...

AnandTech deep dive and review on the chip: https://www.anandtech.com/show/13405/intel-10nm-cannon-lake-...

It sure is, but 7nm in cpu lithography doesn't mean anything, it's pure marketing.

https://en.wikichip.org/wiki/7_nm_lithography_process

> The term "7 nm" is simply a commercial name for a generation of a certain size and its technology and does not represent any geometry of a transistor.

Yeah, since about 14nm the nanometer number has become kind of like CPU speed (mhz/ghz) became in the early-mid 1990s and for the same reason.

Long ago you could say e.g. a 600mhz CPU was faster than a 500mhz CPU. Then things like instruction level parallelism and other significant optimizations started and you had lower-core-speed CPUs that would in practice be quite a bit faster than higher-core-speed CPUs. Today a 1.6ghz mobile chip is far faster than a 2.4ghz earlier generation Pentium-4 for example, at least for most workloads.

My understanding is that chip fab processes have developed similar numbers of devils in details. It's not just about the smallest feature size but lots of other things like materials, transistor types, layout, power density, etc. I've heard that Intel's 14nm processes are comparable to other fabs' 10nm processes for instance.

Still this 7nm node is better than what Intel is currently shipping.

We need a new “PR rating”[1] system for feature size to make some sense of these numbers being bandied about.

[1]https://en.m.wikipedia.org/wiki/Performance_Rating

Of an individual silicon atom, apparently. I'd think the "grid size" of silicon crystals might be more relevant, and that's not necessarily the same number... Hmm.

0.235nm between atoms, 1/30th of 7nm. (Though, precisely what does the 7nm number measure?)

Nobody knows anymore, it’s kind of feature size, but everyone defines feature differently!

It seems like transistor density (N per unit area) would be a more useful measurement.

Isn't that the logic behind the "fake" nanometer measurements? Let's say everyone is on "32nm", but you got a new process[1] that packs twice the number of transistors than everyone else, but with the same feature size. Essentially, your 32nm feature size process is equivalent to everyone else's 22nm feature size process. Are you going to advertise your new process as "32nm" which is the same as everyone else, or "22nm"?

[1] eg. https://en.wikipedia.org/wiki/Multigate_device

I mean I'm no chip designer but surely they are using some crazy hacks like transistors that take up more space than the lines between them by being part of the substraight above and below the etched lines. And presumably they use interference/coherence to etch even finer details? There are probably about a million different techniques they use to pack in more transistors without making the process smaller.

Reply on news.ycombinator.com