You are currently viewing SemiWiki as a guest which gives you limited access to the site. To view blog comments and experience other SemiWiki features you must be a registered member. Registration is fast, simple, and absolutely free so please, join our community today!
I just being reading an article on AnandTech, AnandTech | Intel at ISSCC 2015: Reaping the Benefits of 14nm and Going Beyond 10nm. In the article Intel say that the cost per transistor has progressively gone down for each of their nodes. However, the only compare the cost per transistor with their own nodes, not with other foundries. So, I was wondering how does the cost per transistor compare between foundries?
The reason I ask is because Intel are using TSMC for their low cost SoFIA SoC. Could it be that the reason why is because the manufacturing costs are cheaper on TSMC's 28nm?
I just being reading an article on AnandTech, AnandTech | Intel at ISSCC 2015: Reaping the Benefits of 14nm and Going Beyond 10nm. In the article Intel say that the cost per transistor has progressively gone down for each of their nodes. However, the only compare the cost per transistor with their own nodes, not with other foundries. So, I was wondering how does the cost per transistor compare between foundries?
The reason I ask is because Intel are using TSMC for their low cost SoFIA SoC. Could it be that the reason why is because the manufacturing costs are cheaper on TSMC's 28nm?
No, SoFIA came from the Infineon Wireless acquisition and Infineon is a long time TSMC customer. The 28nm SoFIA was already in process when the acquisition happened. But you should know that Intel has been a TSMC customer for many years so they are not strangers. Future SoFIAs will be on Intel processes for sure.
No, SoFIA came from the Infineon Wireless acquisition and Infineon is a long time TSMC customer. The 28nm SoFIA was already in process when the acquisition happened. But you should know that Intel has been a TSMC customer for many years so they are not strangers. Future SoFIAs will be on Intel processes for sure.
I don't think Intel have ever presented any direct gate cost comparisons with TSMC, only density comparisons at each node (to show that they're better now) and transistor cost between nodes for Intel (showing that their costs are dropping each node).
I'd be very surprised if Intel foundry could be cheaper than TSMC on an apples-to-apples basis -- meaning for the same volume of the same product at the same time -- because their processes have performance and density prioritised over cost, as they should have given their market.
But they might take the view that their bleeding-edge fab costs (e.g. for 14nm) are quickly paid for by CPU sales, then they can offer cheap wafers for SoC assuming that the customer can put up with the pain of using their process -- in other words, subsidising the cost of later SoC wafers using profits from earlier CPU wafers. This would lower their overall margins but extend the useful lifetime of the fabs and bring in more total revenue, which makes business sense.
I don't think Intel have ever presented any direct gate cost comparisons with TSMC, only density comparisons at each node (to show that they're better now) and transistor cost between nodes for Intel (showing that their costs are dropping each node).
I'd be very surprised if Intel foundry could be cheaper than TSMC on an apples-to-apples basis -- meaning for the same volume of the same product at the same time -- because their processes have performance and density prioritised over cost, as they should have given their market.
But they might take the view that their bleeding-edge fab costs (e.g. for 14nm) are quickly paid for by CPU sales, then they can offer cheap wafers for SoC assuming that the customer can put up with the pain of using their process -- in other words, subsidising the cost of later SoC wafers using profits from earlier CPU wafers. This would lower their overall margins but extend the useful lifetime of the fabs and bring in more total revenue, which makes business sense.
Intel has a 49% stake in IM Flash Technologies which is a thin margin, high volume business. They produce a WiFi chip that sells for $15-30 (figure the unit costs are $5-10). Intel produces PC chipset chips for a similar amount.
At one time, these sideline businesses were probably good diversification and provided a way to extra value out of legacy fabs. Today, with the way Intel has grown, and continues to grow, they need blockbuster products. A chip that can make AllWinner meaningful money isn't interesting to a behemoth like Intel.
Intel has a 49% stake in IM Flash Technologies which is a thin margin, high volume business. They produce a WiFi chip that sells for $15-30 (figure the unit costs are $5-10). Intel produces PC chipset chips for a similar amount.
At one time, these sideline businesses were probably good diversification and provided a way to extra value out of legacy fabs. Today, with the way Intel has grown, and continues to grow, they need blockbuster products. A chip that can make AllWinner meaningful money isn't interesting to a behemoth like Intel.
I think you need a vertically integrated model to profit from $5 chips. Apple makes money, and Samsung, by making the chip, and putting it in a finished good. So they aren't really selling a $5 chip, but a $500 phone.
Intel is on the cusp of being vertically integrated. It's probably a pretty interesting time in Intel Labs, designing reference platforms for laptops, tablets and smartphones. Some of it could probably compete with Apple and Samsung, and make sense financially. So far, they seem to want to go to market through partners, but as this forum well knows, their partners are few and far between, in the mobile space. So I think inevitably they are going to bring out some "own brand" stuff.
There is more than cost. Even if cost per transistor is higher, for sure Intel 14nm is faster and lower power than TSMC 28nm. Plus, if one day Intel wants to build integrated AP/modem chips on their own process (which presumably they do) then why would they do a redesign for a TSMC 16nm versus their own 14nm.
There is more than cost. Even if cost per transistor is higher, for sure Intel 14nm is faster and lower power than TSMC 28nm. Plus, if one day Intel wants to build integrated AP/modem chips on their own process (which presumably they do) then why would they do a redesign for a TSMC 16nm versus their own 14nm.
Anybody's 14/16nm is faster and lower power than TSMC 28nm. The reason that Intel might want to use a foundry process instead of their own 14nm is that their process is optimised for CPUs and foundry processes are optimised for mobile SoC, so they're likely to be cheaper and probably lower power *for that application* (Intel have the best CPU process). To see why, read the explanation on this article about AMD Carrizo which is an SoC with similar design priorities to a smartphone:
There is also probably a lack of available IP which is vital for a smartphone SoC.
So overall there would be significant advantages for Intel to use a foundry for this. On the other hand if they used their own 14nm process it will help fill up their fabs after their 14nm CPUs are phased out in favour of 10nm... ;-)
