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 do not think it would be second source. Apple would have to go all in with Intel like they did with TSMC at 20nm. The whole Apple working with Intel on 14A is just me talking to myself, no facts involved. It is one way Intel could stay in the foundry business. If not Apple it could be QCOM. I do not think Nvidia or AMD would consider it. Broadcom is working with Rapidus.
I do not think it would be second source. Apple would have to go all in with Intel like they did with TSMC at 20nm. The whole Apple working with Intel on 14A is just me talking to myself, no facts involved. It is one way Intel could stay in the foundry business. If not Apple it could be QCOM. I do not think Nvidia or AMD would consider it. Broadcom is working with Rapidus.
Dual sourcing for the same product line is very difficult and costly in the semiconductor industry. The risks and costs are simply too high for most companies to manage, even for companies as large as Apple, Intel, Nvidia, Qualcomm, Google, or Microsoft.
I think Qualcomm is the only company that has dared to try dual sourcing for the same product line multiple times. The results were not encouraging.
Dual sourcing for the same product line is very difficult and costly in the semiconductor industry. The risks and costs are simply too high for most companies to manage, even for companies as large as Apple, Intel, Nvidia, Qualcomm, Google, or Microsoft.
I think Qualcomm is the only company that has dared to try dual sourcing for the same product line multiple times. The results were not encouraging.
This is very true. You also have to consider NDAs. If your team is designing to TSMC N2 it is not allowed to design to 18A or SF2. So, you will need two teams for the implementation phase of IC design.
QCOM used to have their own design rules which allowed them to multi source. I was in on a 40nm QCOM design that went to TSMC first, then UMC, SMIC, and Chartered. No way can you do that now. The new processes are not "T-Like" anymore, that ended with FinFETs.
This is very true. You also have to consider NDAs. If your team is designing to TSMC N2 it is not allowed to design to 18A or SF2. So, you will need two teams for the implementation phase of IC design.
QCOM used to have their own design rules which allowed them to multi source. I was in on a 40nm QCOM design that went to TSMC first, then UMC, SMIC, and Chartered. No way can you do that now. The new processes are not "T-Like" anymore, that ended with FinFETs.
Intel designed Arrow lake to 20A and N3. Nova Lake is supposedly designed to N2 and 18A.
Intel has already stated they are going to be second source or backup for several years.
Intel designed Arrow lake to 20A and N3. Nova Lake is supposedly designed to N2 and 18A.
Intel has already stated they are going to be second source or backup for several years.
Are these chiplet designs? Much different than full chips. I had heard that Intel is using design teams in India for TSMC designs? Either way I can assure you they are separate teams in "data clean" environments so there is no overlap.
Intel designed Arrow lake to 20A and N3. Nova Lake is supposedly designed to N2 and 18A.
Intel has already stated they are going to be second source or backup for several years.
Are these chiplet designs? Much different than full chips. I had heard that Intel is using design teams in India for TSMC designs? Either way I can assure you they are separate teams in "data clean" environments so there is no overlap.
Chiplet or not, higher cost is one of many reasons why Intel has about the same number of employees as TSMC and AMD combined, yet generates significantly less revenue and profit — if any.
I think sooner or later, Intel must fundamentally address this cost and efficiency problem. Letting go of more people is necessary, but it’s not a magic solution that will fix everything.
Chiplet or not, higher cost is one of many reasons why Intel has about the same number of employees as TSMC and AMD combined, yet generates significantly less revenue and profit — if any.
I think sooner or later, Intel must fundamentally address this cost and efficiency problem. Letting go of more people is necessary, but it’s not a magic solution that will fix everything.
Help me out. So Apples SOC for phones has what chips/chiplets in it and who does the packaging. I havent looked at the chip teardown in years. is it still DRAM on LOGIC chip?
INFO is pretty vanilla as far a packaging tech goes.
All SoC packaging is currently handled by TSMC using InFO (Integrated Fan-Out) technology, like you said, which integrates DRAM and logic chips. InFO is considered a major milestone in TSMC’s advanced packaging efforts. It was originally proposed by Chiang Shang-yi, who convinced Morris Chang that advanced packaging represented the future. Morris allocated significant resources to the initiative, but customer adoption was slow due to cost concerns. InFO marked the turning point, enabling advanced packaging to scale into HVM.
Moreover, there’s speculation that Apple may shift to WMCM (Wafer-Level Multi-Chip Module) packaging next year. Packaging capabilities will certainly be a critical factor for Intel if they aim to win Apple’s business from TSMC.
Only way I could see a deal working between Apple and Intel would be if they offered frighteningly competitive pricing per wafer. Apple does love squeezing their suppliers. To a level which would leave Intel with scant profit margins. But given the prospect of 14A not happening without an external partner, perhaps simply being able to cover capital costs would be good enough.
wccftech.com
