The wheel keeps turning. The same day that Intel announced the departure of Pat Gelsinger from the company, we had the opportunity to participate in a technical session in which several engineers from the subsidiary specialized in the manufacture of integrated circuits explained to us what are they working on. And, leaving aside the difficult situation that Intel is going through, the truth is that what they have at hand in this area is objectively very interesting.
Before moving forward we are interested in taking note of one of the promises that Pat Gelsinger made shortly after taking the reins of this company. One of the pillars of his management sought to consolidate Intel as a manufacturer of advanced chips for third parties with the purpose of competing with TSMC, and, to a lesser extent, with Samsung, on a face-to-face basis. And to achieve this, Gelsinger set out to have the most advanced integration technology and the best transistors in the chip industry by 2025.
Intel plans to have integrated circuits with a trillion transistors by 2030
Exactly two years ago, at the beginning of December 2022, Intel announced that by 2030 its photolithography will have advanced so much that it will be able to produce integrated circuits that will bring together no less than a million million transistors. To achieve something like this before the end of this decade, it has to greatly refine its transistors and integration technology. And those responsible for doing so are essentially the same engineers who have participated in such relevant innovations as FinFET transistors or stressed silicon.
Intel already has the necessary technology to transfer more than 15,000 ‘chiplets’ in parallel to a wafer
There is no doubt about this: Intel is putting all its effort into increasing the competitiveness of its integration technologies. You have no choice if you really want to. fight face to face with TSMC and Samsung. Be that as it may, the interesting thing is that its engineers are working on several innovations that, on paper, look really good. In the next slide we can see one of them, and one of the data that is worth not overlooking is that Intel already has the necessary technology to transfer more than 15,000 in parallel to a wafer. chiplets.
A priori it is difficult to believe it if we keep in mind that the standard diameter of wafers currently produced is 300 mm (12 inches). It seems impossible that more than 15,000 integrated circuits can be transferred to one of them simultaneously, but it all makes sense if we observe that each of these chips has an area of 1 mm². Incredible. Furthermore, according to Intel this process takes a few minutes, while current integration technologies require several hours, or even more than a day to achieve the same result.
When this innovation leaves the laboratory and reaches Intel’s semiconductor plants, all the integrated circuits it produces will benefit from it, but there is one type of chip in particular that, in theory, will emerge especially strengthened: GPUs for artificial intelligence (AI). ). Intel wants to consolidate itself in this market with the purpose not only of increasing the competitiveness of its own AI chips; It also wants to attract new customers and manufacture data center GPUs designed by third parties. Today TSMC is the company that leads this sector of the semiconductor industry.
Intel says it has refined its GAA transistors enough to adopt a 6nm gate length with a channel thickness of just 1.7nm
The second slide that you have a little higher also contains several interesting pieces of information, but there is one that is worth noticing: Intel claims that it has refined its GAA transistors (Gate-All-Around) necessary to adopt a gate length of 6 nm with a channel thickness of only 1.7 nm. These figures are quite impressive and clearly reflect that the integration technology that Intel describes is much more advanced than what it currently has in production.
A note before moving forward: GAA transistors are going to beat at the heart of the next batches of cutting-edge integrated circuits. TSMC, Intel and Samsung, the three largest semiconductor manufacturers on the planet, they are working on themalthough at the moment only Samsung is using them. In fact, it started GAA chip production on its first-generation 3nm node in June 2022.
The innovations we have just delved into are not the only ones Intel engineers are working on. They also claim to have managed to drastically optimize the interconnections within each integrated circuit, as well as to have reduced defects during the manufacturing process that can significantly degrade the performance of the chips. Intel Foundry has other innovations on its hands, but the ones we just explored allow us to get a pretty good idea of what they are preparing. 2025 promises us strong emotions.
Images | Intel
In Xataka | Intel will not spin off its chip manufacturing business freely. You have acquired a commitment that you cannot avoid
Add Comment