(Semiwiki) Intel 14nm Delayed Again?

[Sweepr]

The fact that they are only bringing K models to LGA desktops kinda hints that desktop Broadwell will be a ''niche'' product. Skylake is the true Haswell successor on desktops IMHO

 

[AtenRa]

You are assuming that a delay in 14nm ramp up will negatively impact 10nm time frame. The R&D pipelines run in parallel, albeit in different stages, and a delay in one of them is not assured to impact the other.

Same was true for 22nm and it was delayed vs 32nm, same was true for 14nm and again it was delayed vs 22nm. Just because they run in parallel doesn't mean one is not affecting the next one or that 10nm will run without any problems from start to the end.

 

[Fjodor2001]

The fact that they are only bringing K models to LGA desktops kinda hints that desktop Broadwell will be a ''niche'' product. Skylake is the true Haswell successor on desktops IMHO

Ok, so Intel has effectively switched to a ~3 year release cycle for desktop CPUs. I guess it makes sense, given how poor performance improvements we're seeing on a yearly basis nowadays.

 

[witeken]

Cannonlake is realistically a 2017 product given that the 2 years time frame between two nodes is not the case anymore.

Not according to Intel.

 

[mikk]

You are assuming that a delay in 14nm ramp up will negatively impact 10nm time frame. The R&D pipelines run in parallel, albeit in different stages, and a delay in one of them is not assured to impact the other.

I didn't assume this. The point is process shrinking gets harder and harder, the old ideal 2 years time frame between two processes is unlikely to be repeated. The trend goes to longer time frames between a node which makes it unlikely that we see 2 years after 14nm a 10nm product in the market.

 

[witeken]

Again, Intel disagrees with you. Mark Bohr (2012):

"Right now I'm spending my time personally on 10nm pathfinding and it looks like we have a solution there."

 

[Fjodor2001]

Since node shrinks are getting ever more expensive, the sheer economics of it all should explain why Intel 10 nm production will not commence in 2015. They simply need longer sales windows to make up for the increasing costs.

 

[mrmt]

Since node shrinks are getting ever more expensive, the sheer economics of it all should explain why Intel 10 nm production will not commence in 2015. They simply need longer sales windows to make up for the increasing costs.

The R&D is a very small part of the development, the hard part is the CAPEX, and to counter CAPEX the foundries rely on volumes, not longer sales windows. This is why Intel is so keen to get into the mobile market, they *need* the volumes to keep in the foundry race.

 

[Fjodor2001]

The R&D is a very small part of the development, the hard part is the CAPEX, and to counter CAPEX the foundries rely on volumes, not longer sales windows. This is why Intel is so keen to get into the mobile market, they *need* the volumes to keep in the foundry race.

In the mobile market the volumes may be higher, but the margins are much less so it's no magic bullet. Also, Intel has said that they will be successful in the mobile market for years, but so far it has not come true. ARM based chips dominate that market.

In addition, longer sales window means larger life time volume per node. The sales volume per time unit is nowhere near enough to mandate 10 nm in 2015, no matter if Intel is successful in the mobile space or not. They need 14 nm devices on the market long enough to sell sufficient amount of chips to earn enough money to get a profit from the 14 nm investment first.

 

[krumme]

The R&D is a very small part of the development, the hard part is the CAPEX, and to counter CAPEX the foundries rely on volumes, not longer sales windows. This is why Intel is so keen to get into the mobile market, they *need* the volumes to keep in the foundry race.

The customers is not there to pay for it. Its already happening. Not only is nodes getting exponentially more expensive. The consumer benefit is in no way following that pattern. And its the last factor that matters. Who cares about cost if the customer is there for it.

Intel really needs to take care of capex and they have been ajusting - if that word is right for the major shift. My guess is we will see more of this. The same goes for tsmc although their depreciation and business model is better in those new situations.

The business needs some new innovation on the consumer side to take back some of the money. We need a new innovative fruntrunner.

 

[erunion]

The same goes for tsmc although their depreciation and business model is better in those new situations.
.

In what way is TSMC fairing better?

Nvidia has been pretty vocal about how they are unhappy with tsmcs node development.

 

[witeken]

In the mobile market the volumes may be higher, but the margins are much less so it's no magic bullet.

Intel can sell a chip from MediaTek or Qualcomm with a 30% margin at a 50% margin or so. Instead, until they have a high market share, they'll use that advantage to make a 30% margin chip that no one else can make.

Regarding your comment on process nodes, you should look at it from this perspective: the PC and server market gives them a big, solid amount of cash to invest in other markets and to continue following Moore's Law. Moore's Law is the bare minimum in the semiconductor industry, and if you don't follow it, you'll fall behind. Intel will never think of slowing down. It gives such a huge amount of benefits: lower cost per transistor, higher margins, lower power consumption, higher performance, the ability to make chips no one else can, which results in higher market share and income.

Like former CEO of Intel Craig Barrett said: the last thing you should do is stop investing in new process tech.

 

[Fjodor2001]

Intel can sell a chip from MediaTek or Qualcomm with a 30% margin at a 50% margin or so. Instead, until they have a high market share, they'll use that advantage to make a 30% margin chip that no one else can make.

Regarding your comment on process nodes, you should look at it from this perspective: the PC and server market gives them a big, solid amount of cash to invest in other markets and to continue following Moore's Law. Moore's Law is the bare minimum in the semiconductor industry, and if you don't follow it, you'll fall behind. Intel will never think of slowing down. It gives such a huge amount of benefits: lower cost per transistor, higher margins, lower power consumption, higher performance, the ability to make chips no one else can, which results in higher market share and income.

Like former CEO of Intel Craig Barrett said: the last thing you should do is stop investing in new process tech.

Doesn't matter, you still need sufficient life time sales volume per node to make a profit.

So what if you get "lower cost per transistor, higher margins, lower power consumption, higher performance, the ability to make chips no one else can". If you still need to make a very large investment for each node, it's no good keeping the node for too short time making you sell too few chips per node. You'll lose money doing that.

 

[mikk]

Again, Intel disagrees with you. Mark Bohr (2012):

He doesn't disagree to anything. They have a solution doesn't tell anything about a time frame and by the way he told the same about 14nm. Intel itself, to be exact William Holt, told that it gets harder and harder with smaller and smaller process technologies. Everyone knows it.