| Manuel Calavera |
19.07.2017. 20:18 |
Malo o nadolazećim proizvodnim procesima
Citiraj:
...
Node lead aside, Intel claims that similar node names don’t mean similar performance anymore. They claim their process to be superior as compared to others, based on the density of the logic transistors. Intel does have a superior node as compared to competitors but here’s the catch. AMD is planning to go to GlobalFoundries’ 7nm. GF’s process density at 7nm is higher as compared to Intel’s density at 10nm. Details follow:
Intel has the smallest CPP and MPP and is the most dense 14nm process.
In other words, Intel’s 14nm is superior to its counterparts as it can pack more transistors on a given area. According to one estimate, Intel provides more than 1.5x logic density than other processes on the same node. Transistor density comparison is as follows:
MTr/mm 2
| GlobalFoundries 14nm
| Intel 14nm
|
|
|
| Estimated Transistor Density,
|
| 32.5
| 37.5*
| Source: semiwiki
Intel's superior 14nm process is under attack
It’s worth emphasizing that AMD is giving Intel a hard time with Ryzen and EYPC despite the node superiority of Intel. EPYC outclasses Broadwell based Intel server processors. We have yet to see how Skylake-SP compares with EPYC. Ryzen also is bringing the multi-thread challenge for Intel. The point is,
AMD is giving Intel a run for their money based on an inferior 14nm process compared to Intel’s process, featuring best-in-class process density.
This indicates AMD’s design efficiency. Once AMD has a process lead, which is expected during the second half of 2018, the performance dynamic will change drastically. And it’s quite possible that we will see Intel playing catch up in terms of performance and power.
Global Foundries 7nm is denser as compared to Intel’s 10nm.
Intel 10 nm process features a CPP and MPP of 54nm and 36nm respectively. Cell height is around 272nm. The initial 10nm process has 25% better performance or 0.56x better active power than the initial 14nm process. It is worth mentioning that Intel’s first 10nm will have lower performance as compared to its 14++ process.
MTr/mm 2
| Global Foundries 7nm
| Intel 10nm
|
|
|
| Estimated Transistor Density
| 127.3
| 100.8*
| Source: semiwiki
It can be seen from the table that Global Foundries' 7nm is denser as compared to Intel’s 10nm - 7nm should be denser. Transistor density conforms that Global Foundries’ process will be able to outperform Intel’s 10nm at a lower power rating.
Global Foundries process is expected to have 1.23x the density of Intel’s 10nm process. Overall, AMD will have a performance lead over Intel by the second half of 2018. And this is going to persist until Intel updates its process technology.
The Dual Effect
The new process features 17M gates/mm 2 leading to performance gains up to 40% and 60% reduction in power consumption over the 14nm generation. This is unprecedented as performance increases and power reduction is being promised simultaneously. Previously, the industry always reported performance increases for a given power draw or a reduction in power draw for a given performance.
To review, Intel is expected to lose its node lead by the second half of 2018. As mass production of Global Foundries 7nm LP is expected during the second half of 2018, this puts AMD just six months behind Intel’s 10nm launch. However, Intel already has delayed the launch of 10nm parts several times. Therefore, it’s difficult to say how much time lead Intel will have with its 10nm process.
Bears often put the argument that AMD can’t compete in terms of R&D, and sooner or later, AMD will be behind Intel in terms of process technology or the design. However, they forget that AMD is fabless, which allows AMD to focus on design. Foundry partners of AMD take care of its manufacturing R&D. In effect, Intel is up against Samsung, Global Foundries and IBM when it comes to process technology. Samsung was ranked fourth in semiconductor R&D spending during 2016. IC Insights also gives Samsung a fourth spot in terms of semiconductor R&D spend. The point is that AMD has help in R&D through foundry partners. Consequently, comparing R&D expenditure of Intel and AMD to derive a bear thesis isn’t a good idea. Further, being a fabless company, AMD has the luxury to focus on design while Intel’s efforts are split between process and design.
10nm chip takes around $192 million for the total design cost while a 7nm chip costs around $271 million, according to Gartner. EUV can lead to cost savings. In theory, EUV can simplify the patterning process. As GlobalFoundries is planning to introduce EUV later, costs will come down for the manufacturing process. However, there’s another side to this story. According to GlobalFoundries,“ For most customers in most of the markets, 7nm appears to be a more favorable financial equation. It offers a much larger economic benefit, as well as performance and power advantages that in most cases balances the design cost a customer would have to spend to move to the next node.”
Semiengineering further notes that “based on PPASC metrics and the cost-per-transistor curve, 7nm looks like a better option.”
All in all, it’s not as straight forward in terms of cost advantage. Nevertheless, there are indications that overall costs are favorable in 7nm. Even if one argues that 7nm has a higher cost and will result in margin deterioration, this doesn’t make up for a strong bear argument. The reason is that as 7nm is superior, and a price premium can be charged on the products. Historically, Intel has charged a premium amid superior node technology. AMD can do the same if its node lead actually materializes.
Bottom Line
The thesis is simple. Node lead is the critical success factor. AMD is expected to have a node lead in the near future. Intel’s process, although superior to counterparts, is expected to lag behind GlobalFoundries’ 7nm amid transistor density. Intel’s R&D superiority is debatable as AMD can benefit from collaboration. As Intel dominated the market with its superior process in the past, AMD is set to take almost half the market share from Intel given a slightly better performing process technology. We also believe that AMD’s design is inherently better as Ryzen in creating problems for Intel using an inferior 14nm process. AMD is a strong buy as the company is set to take around half the market share from Intel. In the past, Intel captured almost all the market share with its superior node, there’s no reason to discard AMD of taking around 50% share amid slightly superior process technology
|
https://seekingalpha.com/article/408...&uprof=28&dr=1
You've got to go deep before you can be extreme, TSMC is moving to 7nm |
