Boy, this would be a great time...
Boy, this would be a great time to own a chip facto - Ah, nevermind.
IBM has reached a major milestone in its computer chip R&D efforts – by shrinking transistor gates down to 7nm, far below today's proportions. On Thursday, Big Blue announced that it has successfully produced test chips with functional transistors using a 7nm (nanometer) process technology, and claim an industry first. …
Apple actually do very little outside of marketing and making pretty cases. Board design is outsourced to India, the chips are fabbed by various 3rd parties, everything is put together by Chinese factories, most of the software is either out-sourced or just taken from the NetBSD project, and even most of the logistics are handled by third parties. The only things they really do are design the non-functional mechanical bits, run a couple shops, and market the hell out of everything.
Can't push out anything smaller than 28nm? You mean other than the ~150 million A8s they've made for Apple so far in 20nm? Everyone (except Intel, because they started at a higher price point than the foundries) is reporting that cost per transistor is slightly higher when they go below 28nm, so use of those nodes has been less than originally hoped for. But customers who have a little extra to spend for the improved power/performance like Apple, don't care about that.
Most likely the reason why Apple is rumored to be going back to Samsung for the A9 is that Samsung has their 14nm process ready which will further improve power/performance beyond 20nm. TSMC is skipping ahead to 10nm as their next main node because of the aforementioned cost issues, claiming that they've made some changes in that process to address those concerns and their 10nm node should result in lower $/transistor than previous nodes including 28nm. Of course, the $100+ million cost for a mask set at 10nm means that only pretty high volume or very cost insensitive chips will be produced in that node by anyone.
Rumor has it Apple will have the A10 made in 10nm, which probably indicates they're going back to TSMC, unless Samsung plans to have 10nm next year also or the long-fabled idea of Intel making Apple's SoCs finally happens.
>Can't push out anything smaller than 28nm? You mean other than the ~150 million A8s they've made for Apple so far in 20nm?
Good luck getting in front of Apple instead of praying for their sloppy seconds which is probably more the position the first posters company was in, ie the little guy.
The inflexibility of the foundries is a lot of the reason I have job working for a somewhat dated internal fab owned by my company. If you are a small player then its very hard to guarantee you will be able to always meet demand using foundries as demand tends to spike for everyone at once and size usually does matter who get serviced first.
The assumption seems to be that IBM is interested in this for making CPUs.
I would guess that the first application of very small (and therefore very fast) 7nm transistors will be in specialist datacomms devices (things like 100Gbit Ethernet). It's also probably true that in that field, billions of transistors aren't needed. Mere thousands might be useful, mere millions certainly would be.
There's a hint, in that they are talking about SiGe, not plain Si. SiGe is more difficult (but intrinsically faster).
Everyone in the semi Biz is aware that next gen conductors will require different processes and materials and they are continuing to develop these. There is so little gain in node drop since ~32 Nm that it's all pretty much just hype for the modest power reduction gains.
Perhaps just me but when I saw SiGe/Silicon-Germanium, I got the urge to put both my Fulltone '70 BC-108C Silicon Fuzz and Hartman Vintage Germanium Fuzz back on my board and see if I could get my amp to go into meltdown.
Maybe I'll add my Black Arts Pharaoh and LSTR to the mix and see if it starts raining frogs. But then my Blackout Effectors Musket would feel left out. I'll be a bit worried for my brother but I'm a middle child - I'll be alright.
Sorry - what was this article about again?
Biting the hand that feeds IT © 1998–2019