Question 1: How to view the backwardness of Intel's 7nm chip technology?
First of all, we must clarify what the 7nm process is behind, but from this title, it is actually misleading. Based on current data, Intel's 7nm lag refers to the mass production time relative to TSMC's 5nm, rather than its own performance.
semiconductor nodes have been dividing process generations according to density. The estimated density range of Intel 7nm is 200 or 240. The former is a conservative value and the latter is an aggressive value. At present, it is impossible to determine which one it is? However, if calculated based on 200, it is slightly higher than TSMC N5, which can be regarded as a contemporary generation, while if it is 240, it is significantly higher than TSMC's N5, which is half a generation ahead, between N3 and N5. However, no matter from that perspective, Intel lost here. TSMC mass-produced N3 in the second half of 2022, and even encountered 7nm. Intel's 7nm will be at most barely kept up (for 240).
However, at least most of Intel's CPUs don't have to worry about the density backwardness for the time being. TSMC's process has little advantage over cores, because semiconductor processes are not just density, but also performance (which can be understood as frequency) and power consumption. But because these are not easy to consider, the details will be ignored for the time being. But what you need to know is that as far as TSMC's current roadmap is concerned, the performance power consumption improvement of N5 and N3 is minimal, and it is even less than Intel's plus-sharpening process. Everyone has seen 14++. The same is true for
to 10nm. Compared with the 10+ Tiger Lake 1185G7, the 10+ Tiger Lake 1185G7, which is 28W, has increased the frequency by 30% (2.3/3.0), and the limit has also been raised by 17%. Although it has the advantage of a low starting point, isn’t it also a highlight to complete 14 to 14++ in one polishing? Then, look at the performance polishing level of N5 N3?
Of course, there is another problem. Although Intel does the process itself, Intel does not outsource it. Now Intel is not going to directly play TSMC, but to use TSMC's AMD, Nvidia and other . Intel 7nm is expected to be released from 2022 to 2023, and the corresponding AMD will not be introduced in 5nm in mid-to-second half of 2022, and large-scale imports will be introduced in 2023. For Nvidia, even the consumer level has not even reached 7nm this year. Do you think it will reach 5nm soon? It can be said that at least at this node, although 7nm seems to be very weak, the problem is actually not that serious. (If the current plan goes well).
Question 2: Is Intel's dominance no longer guaranteed?
First of all, it depends on what status is. If we talk about comprehensive semiconductor production technology, Intel's dominance has long disappeared. If it is the dominance of the product, it is hard to say in the long run, but it is still maintained in the medium and short term. Intel's core business is CPU. One of the major requirements for surpassing Intel is at least shipped to exceed Intel, right? But look at the number of films next year (AMD's main force is still 7nm), and the AMD CPU + GPU has 200,000 pieces, which is basically just the same as the 180,000 pieces outsourced by Intel. These 180,000 pieces are just Intel's new independent graphics and some PCHs. Remember that data analysis shows that 180,000 pieces are about Intel's normal use for one month, so it is still a long way from Intel's loss of dominance. Intel's wealth is very rich. So it won’t be in the medium and short term, but it’s hard to say in the long term.
Conclusion: What is the future of Intel?
In the short term, Intel's most secure approach is to parallelize its own technology and outsourcing. Although Intel's technology is relatively vague now, the CPU architecture is really fine. If you find TSMC to manufacture it, it can at least temporarily solve some of the problems that are dragged down by its own technology, which will help curb AMD's CPU attack and expand new businesses (GPUs, etc.). TSMC's technology is suitable for this.
However, if Intel wants to maintain its advantages for a long time, then its own factory cannot be lost. Why? Although the process of TSMC is stable, Intel will inevitably lose its own process-level features after using TSMC. TSMC's process that is focused on stability and high density inevitably falls into the same way as AMD and ARM systems to compete in low-frequency multi-core. How many selling points does the X86 have to sell after lacking a single core?
For Intel's long-term judgment, please look at 5nm.Intel processes actually follow the Tick-Tock strategy. New technologies (HKMG, FinFet, Co Internet, COAG, etc.) are introduced at 45nm 22nm 10nm (HKMG, FinFet, Co Internet, COAG, etc.), while 32nm 14nm 7nm is usually just a microscope before, and the technical context follows the previous process in a large number. The 10nm crash will inevitably lead to problems with such 7nm. In addition, if you switch to EUV for 7nm, you will also face insufficient experience in the first EUV. 5nm GAA is relatively stable in EUV and is technically less related to the previous 10/7nm. If 5nm has no big problems, then Intel still has a chance. If there are big problems, it may really be out of the way.
