Earlier this year Intel officially Confirm some of its 2023 CPU Will be manufactured by TSMC using unspecified manufacturing technology. New rumors from Taiwan indicate that Intel is preparing to use four products manufactured by TSMC’s N3 (3nm) node, and the lineup looks strange (perhaps impractical). In addition, production may be much earlier than expected, which is a very unlikely event. Since this is a rumor, it is best to read it carefully.
Four major products
Apple and Intel are It is expected to be one of the first companies to obtain TSMC’s N3 The manufacturing process sometime next year. Although we can make logical guesses about what Apple might do with N3, Intel’s plan remains largely a mystery.According to unofficial information from Taipei Economic DailyIntel is preparing a GPU and three server processors, which will be manufactured using TSMC’s N3 technology, which is a product lineup that looks strange (if not strange). Since the information may come from only one source, please treat it with caution.
Intel comfirmed Last year, it has been committed to “Atom and Xeon-based SoC based on Intel and TSMC processes.” In fact, the company has an Atom-marked SoC (Snow Ridge, Elkhart Lake) series for 5G base stations, edge computing and IoT applications, some of which can be attributed to server-level solutions. Using TSMC process technology to produce next-generation Atom brand (or Xeon brand) SoCs for these applications usually makes sense because they must be cheap and low power consumption (this is where TSMC’s N3 may come into play). However, we are not sure about Intel’s plan.
For many reasons, it is surprising to see GPU products in Intel’s first rumored N3 chips. Graphics processors do not use cutting-edge nodes because they require GPU-specific IP blocks, such as high-speed internal and external interfaces or output interfaces (DisplayPort, HDMI, eDP, etc.). These IP blocks cannot benefit from high transistor density because they can hardly Extension. At the same time, computing GPUs for AI and HPC (TSMC uses the term HPC to refer to accelerator ASICs, CPUs, GPUs, FPGAs, network controllers, SSD processors, and certain types of chips) applications have different requirements and may adopt Almost any node density and sufficient performance that provide the required transistors.
At this point, apart from Xe-HPG (aka DG2), Xe-HP and Xe-HPC (aka Ponte Vecchio), we know nothing about Intel’s GPU plans. Suppose, Intel can use TSMC’s N3 to make a small “classic” GPU and use the chip as a pipe cleaner for the process (although this pipe cleaning has no meaning for the CPU or SoC). Or it can use TSMC’s N3 for computing GPU blocks, and then use it for data centers or HPC-constrained products (but whether Intel needs a new DC or HPC GPU in early 2023 is a question).
Ramp will start in July 22?
There is also a controversy in the report. It is said that mass production of at least one of these chips will begin as early as July 2022, which seems a bit early for N3 to enter the HVM (mass manufacturing) stage. N3 will start to increase in the second half of 2022, but usually when the company declares the second half of the year, they refer to the back of the year, not the middle of the year. Officially, the climb of N3 was three to four months later than that of N5 (started in April 19).
“N3 is delayed by about three to four months[ed] Compared to 5 nanometers,” Say CC Wei, TSMC’s chief executive officer, was on a recent earnings call with analysts and investors. “3nm technology is very complicated, whether it is processing technology or customer product design. So, we cooperated with a customer, and finally, we decided to mass-produce in the second half of next year. […] We decide together with our customers [to] Best suited to their needs. ”
It depends on how you count, July Can The first month of becoming the N3 slope, especially if you consider Digital age A report earlier this week stated that TSMC is expected to begin mass production of Apple’s SoC for use in “anyone IPhone or Mac computer in the second half of 2022. The use of these words clearly indicates that there is no reliable information on what products Apple will use N3 to produce next year. In addition, if Apple is to launch a new iPhone in September and start mass sales from October to November, the chip bound to the iPhone must enter HVM in April.
At the same time, CC Wei hinted on the earnings conference call that in terms of quantity, smartphone SoC will become the largest adopter of N3 technology in its first year. Therefore, whether non-smart phone N3 chips are put into production in 2022 or 2023, the number will be lower than the number of smart phone components manufactured using this node.
“N3’s first year is accelerating, but smartphones still play the biggest role,” Say Wei. “HPC* applications are also important, and increasingly important. […] Therefore, in the N3 node, in addition to smartphones, we do expect HPC applications to become important. ”
TSMC has more than 480 customers, and companies such as Qualcomm or MediaTek may adopt N3 in their smartphone SoCs before Apple. Despite this, Apple may still be ahead of its competitor’s personal computer N3 SoC.
Intel’s official plans for 2023 are vague and include processor products made by TSMC, although we don’t even know their code names. These plans also did not mention any GPUs. For this reason, please consider any leaks related to Intel and N3 with extreme caution, as they may be only partially correct or completely incorrect.
The biggest problem with Intel and Apple’s N3 plan is not when their chips will enter mass production, but which chips will use 3 nanometers first. TSMC itself expects smartphone SoCs to lead all other N3 chips in terms of number in the first year, so don’t expect Intel to use N3 for its products for high-volume PCs, at least not in the first year of the rise of N3.
Taking all factors into consideration, although we can expect that some of Intel’s 2023 N3 products will sometimes go into production within 2 hours of 2022 (for example, in the fourth quarter), we will not make any guesses about the nature of alpha products. In fact, considering that N5 will only be available for HVM in the second half of the year, it will be interesting to see which product will adopt this node first.