Samsung+GloFo=BFF

[Jonathan]

http://www.bloomberg.com/news/2014-04-1 ... -tsmc.html

14nm is hard, expensive.

I believe that makes it only 3 processes on that node: Samsung+GloFo, TSMC, Intel. Or did I miss one?

The list of real men grows thin!

[Jonathan]

GIS real men have fabs

Pick result from first row, go to page.

http://www.monolithic3d.com/2/post/2011 ... y-now.html

iSuppli says I am correct.

A worrying note: Note that none of the DRAM guys are on the list. Whither main memory?

[quantus]

A worrying note: Note that none of the DRAM guys are on the list. Whither main memory?

DRAM's are all manufactured at older nodes because you can't scale the buried capacitor nearly as much as transistor sizes. The aspect ratio is too high to cut their size.

[Jonathan]

http://arstechnica.com/information-tech ... -business/

IBM dumps their fabs, confirms what we were all thinking. Now Common Platform really is just GloFo+Samsung.

Note to self: find some volume someplace.

[Jonathan]

GloFo gives up!

https://www.anandtech.com/show/13277/gl ... evelopment

Every new node requires billions of dollars in investments. Those costs are eventually amortized over each chip the company makes, so to keep increasing R&D costs from driving up chip prices, foundries need to produce more chips. To make more chips, they need to either run multiple fabs that use the same process technology (these are going to cost $10+ billion in the EUV era), or build giant fabs that process a gargantuan number of wafers (these are going to cost $20+ billion in the EUV era). Meanwhile, GlobalFoundries has only one leading-edge fab featuring capacity of 60,000 wafer starts per month.

ouch

[quantus]

GloFo seems to be going the TI route, but without the massive library of legacy designs necessary to keep it afloat :-\ IBM is likely to continue to develop technology, but it's gotta find another partner to continue the development. Where's Trump to keep our fabs open and not running away to China and Korea? I say this because I think Intel will be the next fatality since TSMC's volumes are climbing continuously and Samsung has already surpassed Intel's volumes. Samsung's been eating some of TSMC's lunch and keeping Intel from breaking into the contract fab market significantly. At some point, even Intel's volumes just won't support owning a fab on their own.

[Jonathan]

Where's Trump to keep our fabs open and not running away to China and Korea?

Not sure GloFo counts as "our". The money from the spin-off came from UAE. They kept it running in a way that a public company depending on large institutional investors like PERS would not.

Semiconductors is barely a market nowadays. SoftBank is another example of this. Or consider Samsung's ties to the Korean government. It's not entirely comfortable.

[Jonathan]

At some point, even Intel's volumes just won't support owning a fab on their own.

Note to self: find some volume someplace.

[quantus]

Or tell the fab guys to stop trying to charge so much of a premium when the product is obviously not enough better to warrant it. Really, Intel should be offering new-customer/preferred-customer discounts to attract people to their platform so they might get hooked on it.

Also, Intel needs to get much better about making their chips low power. Maybe make a super low power MIPS chip to get in at the low end and scale it up like ARM's been doing. At some point, ARM will move out of pure mobile and more directly compete with x86. Virtualization and other server features are working their way into the ISA and have been the main blockers to using them in power-conscious data centers that run a lot of virtualized workloads.

Ray tracing hardware might be a source of volume? At least for a while.

[Jonathan]

Discrete GPUs are only a few million units per year.

[quantus]

True, but at least they're big, so worth at least 3-5x of most alternatives... Make a decent line cell phones like Samsung or Apple?