Zero volume test chips? Arm does not sell chips.
Who is the Intel 18A Large Customer Prepay?
- Thread starter Daniel Nenni
- Start date
Zero volume test chips? Arm does not sell chips.
delong.height
Active member
I think the question for Arm's business model is if Arm will transition to become a fabless and IP company instead of IP only company.
There isn't much revenue appreciation given Apple design their own chips using Arm's instruction set, and Qualcomm intention to design its own as well. The revenue from IP licensing is destined to drop. The massive quantities of Arm chips manufactured is not enough to make Arm a high growth company.
Competing with customers is a slippery slope. I have seen it done successfully but more often unsuccessfully. In my opinion building up an ecosystem for vertical markets is a key growth area. AI should be a target for sure, lots to do there.
RISC-V is a big risk for Arm, pardon the pun. Hopefully Arm arm's themselves for a full on battle in the trenches. It will be all about the ecosystems.
If ARM does that folks would reformat the entity of the fluid portions of their roadmaps to be on RISC V within the hour. ARM already does enough to grind their customer's gears that would be the final nail. It also helps that ARM has zero SOC/hard design design experience or even an avenue to really sell them. In short, in a not so surprising twist the guy who wrote the article has no clue what he is talking about. Put another way I would be less surprised if TSMC started making DRAM/NAND than I would if ARM made HPC chips for sale to end customers.
Nvidia being the whale is looking even better:
The chairman of TSMC admitted that the ongoing short supply of compute GPUs for artificial intelligence (AI) and high-performance computing (HPC) applications is caused by constraints of its chip-on-wafer-on-substrate (CoWoS) packaging capacity. This shortage is expected to persist for around 18 months due to rising demand for generative AI applicationsand relatively slow expansion of CoWoS capacity at TSMC.
"It is not the shortage of AI chips, it is the shortage of our CoWoS capacity," said Mark Liu, the chairman of TSMC, in a conversation with Nikkei at Semicon Taiwan. "Currently, we cannot fulfill 100% of our customers' needs, but we try to support about 80%. We think this is a temporary phenomenon. After our expansion of [advanced chip packaging capacity], it should be alleviated in one and a half years."
asia.nikkei.com
The chairman of TSMC admitted that the ongoing short supply of compute GPUs for artificial intelligence (AI) and high-performance computing (HPC) applications is caused by constraints of its chip-on-wafer-on-substrate (CoWoS) packaging capacity. This shortage is expected to persist for around 18 months due to rising demand for generative AI applicationsand relatively slow expansion of CoWoS capacity at TSMC.
"It is not the shortage of AI chips, it is the shortage of our CoWoS capacity," said Mark Liu, the chairman of TSMC, in a conversation with Nikkei at Semicon Taiwan. "Currently, we cannot fulfill 100% of our customers' needs, but we try to support about 80%. We think this is a temporary phenomenon. After our expansion of [advanced chip packaging capacity], it should be alleviated in one and a half years."
TSMC sees AI chip output constraints lasting 1.5 years
Advanced chip packaging will be 'paradigm shift' for semiconductor industry
asia.nikkei.com
What does that have to do with 18A? If NVIDA goes to intel for packaging as you speculate, they are under no obligation to buy wafers.Nvidia being the whale is looking even better:
The chairman of TSMC admitted that the ongoing short supply of compute GPUs for artificial intelligence (AI) and high-performance computing (HPC) applications is caused by constraints of its chip-on-wafer-on-substrate (CoWoS) packaging capacity. This shortage is expected to persist for around 18 months due to rising demand for generative AI applicationsand relatively slow expansion of CoWoS capacity at TSMC.
"It is not the shortage of AI chips, it is the shortage of our CoWoS capacity," said Mark Liu, the chairman of TSMC, in a conversation with Nikkei at Semicon Taiwan. "Currently, we cannot fulfill 100% of our customers' needs, but we try to support about 80%. We think this is a temporary phenomenon. After our expansion of [advanced chip packaging capacity], it should be alleviated in one and a half years."
TSMC sees AI chip output constraints lasting 1.5 years
Advanced chip packaging will be 'paradigm shift' for semiconductor industry
I think Nvidia will need both wafers and packaging. I also think 18A will out perform N2 so nimble HPC people like Nvidia can straddle different foundries. Especially with chiplets, Nvidia is all over chiplets. Intel will package other foundry chiplets, last I heard TSMC will not, especially if they have a packaging crunch. The next TSMC Symposium is in two weeks, we may know more then. Someone has to know who the whale is!
MKWVentures
Moderator
After checking some sources, I think we need to see the details on this. .... maybe it will be in the 10Q. As CFO notes "Pat was excited [too excited?]" It might not be the commitment level you think it is
I wonder about that strategy from TSMC. On the one hand it makes TSMC process tech more sticky. But on the other hand I feel like this opens the door for everyone else. Given TSMC isn’t a leader in every category (for example SOI, SiC, GaN, etc) some folks are forced to go outside TSMC’s ecosystem or they chose to do so for *insert reason here*. For these customers TSMC then losses on potential packaging contracts (at least not currently a problem in the current market), or worse those customers might get comfortable with a competitor’s ecosystem and might move wafer orders to them now that the barrier of switching is much lower.
Time will tell if they are making a mistake on this, or if it was just the competive moat needed to fend off the little guys on the trailing edge and the IDMs on the leading edge.
delong.height
Active member
hey, how about Arm selling ready-to-package chiplet that are ready to be packaged with chiplet Arm's customers are going to design?
In doing so, it completely depends on Arm's capabilities of finding customer, but it won't compete with client at the end of the day.
I think there are multiple ways for Arm to do to enable a greater revenue growth without compromised. But this is a better way for Arm and Intel.
That would work. A step between IP and full chip. I see a big chiplet super store coming, absolutely.
Are their any embedded programmers or RISC-V architects out there in SemiWiki land who wants to help make a multicore CPU with intertwined SRAM (Compute-In-Memory) + MRAM. We created most of the ecosystem. We can supply OpenAccess based mostly fully automated EDA tools (analog, digital, drc, lvs, UVM), most of the analog IP, handle the MPWs and packaging all at no charge. Sorry, only tiny US based companies (5 man or less).
Let's see if we can help out the ARM short sellers (can you sell short on these ADRs?)