Some manufacturing mistakes made by Intel in the past few years have brought great advantages to competitors such as AMD. Now, AMD is in an ambitious five-year plan to regain the magic of its chip manufacturing.
This week, at the 2022 IEEE VLSI Technology and Circuit Seminar, which started on Monday, Intel explained in detail how to make faster, cheaper and more reliable chips from the manufacturing perspective in the near future.
Details revolve around "Intel 4", which is a manufacturing node, formerly known as the chip manufacturer's 7-nanometer process. Intel plans to use this node for products that will enter the market next year, including the meteor lake CPU for PC and the computing module of Granite Rapids server chip.
"Intel 4" is the transition process of Intel. It is the first Intel process node to use EUV (extreme ultraviolet, that is, soft X-ray) lithography technology instead of deep ultraviolet immersion lithography technology. The process progress of "Intel 4" maintains Moore's Law by doubling the achievable transistor density relative to the process node of "Intel 7" (formerly known as 10nm enhanced super fin (10ESF)).
Intel previously promised that the performance per watt of "Intel 4" was 20% higher than that of "Intel 7". "Intel 4" was previously called the company's 10nm enhanced SuperFin node, which powered the Alder Lake client CPU and the recently delayed Sapphire Rapids server chip.
In the briefing, Ben Sell, an executive in charge of Intel 4 development, said that the node is progressing smoothly, and his team has been able to improve the performance of Intel 4 by about 265,438+0.5% with the same power. On the contrary, "Intel 4" can provide the same frequency level as "Intel 7" and reduce power consumption by 40%.
This means that future chips like Meteor Lake will not only have better performance, which we always hope to achieve with new chips, but also be more efficient. Improving efficiency can have a great impact on reducing the power consumption of PC or server or prolonging the service life of laptop battery.
A progress made by Sell's team in improving the frequency of "Intel 4" is to increase the capacitance of metal insulator metal capacitor by 2 times, which has been used by Intel as a building block for chips since the 14nm process of Broadwell CPU was introduced in 20 14.
According to Sell, the increased capacitance leads to a smaller large voltage swing, which in turn increases the available voltage of the CPU and allows it to run at a higher frequency.
"What we see on the product, generally speaking, means that you can run the product at a higher frequency," he said.
Although improving performance is very important for new manufacturing nodes, it is also important to reduce costs and make the chip manufacturing process more reliable. In these respects, Searle said that his team has made good progress.
According to Sell, EUV allows Intel to simplify the lithography process compared with the immersion process that Intel used for nodes before. In practice, this means that Intel can reduce the number of layers required for etching chip designs on silicon wafers from five to one.
"Now, everything can be printed on a single layer, giving you the same structure," he said.
Searle told us that the use of EUV can also improve the manufacturing yield, which means that when new chips are put into production, the number of defective wafers will decrease.
Another advantage of EUV is that it will reduce the chip manufacturing cost of Intel products using "Intel 4", despite the high cost of using EUV. Sell believes that this is because EUV reduces the steps and tools needed to make chips.
"Besides the lithography tools in our factory, there are many other tools. Once you combine everything into one step, many of them are unnecessary," he said.
He believes that this simplified process will enable Intel to increase production capacity.
"This means that you need less clean space. So, in general, either you need to build fewer fabs or you can get more output from each fab, "Searle said.
These and other process improvements represent a more modular approach adopted by Intel in developing new nodes. Compared with the more radical methods used by chip manufacturers to develop nodes, this is a huge change, which has caused Intel to suffer major errors and delays on 10nm and 7nm nodes in the past few years.
"The main thing we are doing now is that we are adopting a more modular development method, which means that you don't have to have a huge step, but there are some smaller steps and several modules in this process, which you can develop independently. This makes it easier to develop each module in time without having to solve all other problems to understand the complexity of this module, "Sell said.
In addition, Intel has introduced new methods and improvements to the basic circuit, which will become an important part of future solutions. One of the circuit innovations uses computational near memory (CNM) technology to improve the performance of eight nuclear RISC-V processor, which will be exhibited in the spotlight demonstration of the seminar. These and future innovations developed by Intel will not only support Intel's product portfolio, but also aim to benefit customers of Intel OEM Services (IFS), Intel's new OEM business.
Intel will also directly demonstrate the MOCVD of 2D materials on the 300mm silicon platform, including the first P-type WSe2 used in BEOL and FEOL applications. MoS2 nFET shows variability with increasing geometric ratio.