Intel’s replacement unit “Ami” will use 18A in 2025.
1 Å = 10^–10 meters = 0.1 nanometers.
In February of this year, Intel’s new CEO, Pat Gelsinger, officially took office. This veteran who has worked at Intel for 30 years outlined his vision of ” IDM 2.0 .”
In the new roadmap, in addition to the use of external chip foundry, the development of process technology is an important aspect, all of which revealed the mystery at the Intel Accelerated conference on July 27.
This morning, Intel CEO Pat Gelsinger and Intel’s senior vice president and technology development manager, Dr. Ann Kelleher, etc. introduced to us Intel’s detailed development route from this year to 2025 and beyond.
“From chip design, production testing to integration of the entire chain, this is a capability that only Intel can achieve. Intel’s problem is that it needs to continue to accelerate,” Pat Gelsinger said.
Intel’s next node is not 10nm but Intel 7, and then Intel 4, Intel 3, and Intel 20A after entering the GAA era.
Pat Gelsinger said: “Intel is reconsidering how to release and brand its innovations in chips.”
Intel’s latest announcement released Intel’s processor roadmap for the next five years, new chips and packaging technologies, and made the promise of “annual cadence of innovation”. The ultimate goal is to allow Intel to regain its leadership in the processor field by 2025.
Intel’s future products will no longer use nano-based node nomenclature, but instead use a brand-new naming method introduced by Intel itself. Intel said that the new naming method will more accurately describe the process nodes of the entire chip industry. This change will begin with Intel’s 12th Gen Alder Lake, which will be released later this year.
Intel’s third-generation 10nm chips will be called “Intel 7”, and will no longer be named based on nanometers like last year’s 10nm SuperFin chips.
This sounds like a marketing strategy aimed at making Intel’s upcoming 10nm chips more competitive with AMD’s products. At present, AMD’s products have launched 7nm process chips through TSMC , and Apple has also launched 5nm process M1 chips.
Although these technologies seem to be leading, they are not exactly the case. In the naming of the chip, due to the advancement of 3D packaging technology and the physical properties of semiconductor design, the node name does not actually refer to the size of the transistor on the chip.
From a technical point of view, Intel’s 10-nanometer chips with TSMC or Samsung and other competitors “7 nano” brand hardware roughly the same, with Intel using similar production technology, and provides match the United States transistor density. The same is true in the field of commercial hardware. For example, Intel’s 10nm chip can compete with AMD’s 7nm Ryzen chip.
Therefore, Intel’s “rebranding” is an important change in the way the company’s technology nodes are introduced.
Intel’s new architecture roadmap.
In the industry, people originally used the size of the gate to reflect the process technology. At the current stage, the development of chip capabilities has largely depended on other technologies.
“The decline in numbers will continue to indicate the evolution of technology, but people need to understand that future process evolution and numbers are no longer directly related.” said Song Jiqiang, director of Intel China Research Institute.
Intel’s architecture route and new technology
Let’s take a look at Intel’s new roadmap and its practical significance.
Intel 7 is the new name of Intel’s third-generation 10 nm technology and the successor to Intel’s 10-nanometer SuperFin (also known as Intel’s second-generation 10 nm process chip, the most notable of which is the 11th-generation Tiger Lake). Intel said that compared to the previous generation, the new Intel 7 hardware will provide approximately 10% to 15% performance increase per watt, or if hardware manufacturers want to keep performance the same, Intel 7 will increase Power efficiency and battery life.
The first Intel 7-based products will be launched as early as this year. It is expected to release Alder Lake chips for consumer products at the end of 2021, and Sapphire Rapids chips for data centers will be released in 2022.
Intel 4 is the official architecture of the original Intel 7nm process. Last summer, Intel was forced to postpone it to 2023 due to manufacturing problems. Initially planned to be launched in 2021, this is Intel’s next major technological leap, using extreme ultraviolet (EUV) technology. In contrast, Samsung and TSMC’s 5-nanometer node products have already used this technology. It will still use the wide FinFET transistor architecture that Intel has been using since 2011.
Due to technological improvements, Intel 4 transistor density is expected to reach about 2-250 million transistors per square millimeter, while TSMC’s current 5-nanometer node has a transistor density of approximately 171.3 million transistors per square millimeter.
Intel stated that Intel 4 will increase the performance per watt by about 20% while reducing the overall area. It is expected to be put into production in the second half of 2022, and plans to launch the first batch of Intel 4 products in 2023 (Meteor Lake is used for consumer products, Granite Rapids is used in data centers).
Intel 3 is scheduled to be produced in the second half of 2023 and is the new name for the second-generation 7-nanometer products under Intel’s previous naming scheme. Like Intel 4, it is still a FinFET product, although Intel said it will provide additional optimizations and use EUV, compared with Intel 4, the performance per watt increased by about 18%. The release date or product name of the Intel 3 chip has not yet been announced, but it is speculated that they will not be available until 2024.
Then the chip manufacturing process will enter a new era: Intel 20A, the previous 5nm manufacturing process, will be launched in 2024. A stands for Ångström (abbreviated as Angstrom, symbol Å, 1 Å = 10^–10m = 0.1nm). This is an important sign of Intel’s transition from FinFETs to Gate-All-Around (GAA) transistor RibbonFETs.
Intel has also introduced a new PowerVia technology, which can greatly increase the density of transistors in the transmission circuit on the back of the wafer. Intel will be the first company in the industry to apply this technology.
In addition, what is not listed in the Intel architecture roadmap is Intel 18A, which is expected to be launched in 2025. Intel 18A will use ASML’s latest EUV lithography machine High-NA EUV machine, which it first obtained, which can perform more accurate lithography.
According to the AnandTech website confirmed from Intel, Intel will provide Intel 3 and Intel 20A to foundry customers.
The complete timeline is shown below:
According to the information obtained so far, a more intuitive comparison between the new and old names of Intel’s process nodes is as follows:
In addition to the new architecture roadmap, Intel announced two major updates to the Foveros chip stack packaging technology, and the second-generation technology will debut in the Intel 4 Meteor Lake processor in 2023.
The Foveros chip stack can integrate several hardware elements into a single chip. For example, the Intel Lakefield chip stacks 5 CPU cores, 1 integrated GPU and DRAM in a compact stack. Compared with the traditional design, this can save internal space and integrate a variety of different manufacturing processes and even small chips of different brands into the same large chip to achieve unprecedented performance. The following picture shows the specific updates: Foveros Omni and Foveros Direct.
Foveros Omni will make it easier to mix and match tiles, regardless of their size, so that the area of the base tile in the stack is smaller than that of the top tile, allowing more diversity of stacked chips. Foveros Direct will achieve direct copper-to-copper bonding between components, thereby reducing resistance and reducing bump pitch to less than 10 microns , increasing the interconnect density of 3D transistor stacks by an order of magnitude, allowing multi-chip packaging The boundary between and single chip is gradually blurred.
The two new Foveros technologies are scheduled to be put into production in 2023-24.
The integration of multiple small chips faces the challenge of yield rate. Intel said that it has strong chip testing capabilities that can increase yield rate and reduce the possibility of scrap.
Major Intel OEM customers: AWS and Qualcomm
At present, Intel’s shipments of 10nm chips have surpassed its 14nm chips. Intel’s new architecture naming may help the company more accurately position current and future products to cope with competition, but it still does not change the status quo of Intel’s chip manufacturing technology.
Even if it is admitted that Intel 7 is equivalent to 7nm products from other foundries, 7nm chips and 5nm hardware from foundries such as TSMC and Samsung have already been shipped. This means that companies that rely on these external foundries-such as Apple, AMD, Nvidia , Qualcomm, and almost all other major technology companies-can still get more advanced chips than Intel’s best products.
For example, Apple’s most advanced M1 Mac already uses TSMC’s 5nm chip-and easily surpasses similar products from Intel. There are rumors that AMD will also develop a 5nm Zen 4 processor as early as 2022, which may provide similar competition for Intel to cope with its gradually eroding competitors.
Even if its roadmap has an ambitious annual rhythm, Intel is still lagging behind. In the roadmap announced today, it is expected to catch up with the industry’s leading level after the launch of Intel’s 20A process around 2024. And it expects to regain its leadership in the semiconductor business with Intel 18A by 2025.
Intel will not comment on Intel’s acquisition of GlobalFoundries. In addition, Pat Gelsinger said that the technologies announced today were developed and manufactured in the United States.
Intel’s IDM 2.0 plan also includes opening Intel’s own wafer factories to the outside world. At the event, Intel also announced the latest developments in its foundry services-Intel’s first new customer will be AWS, which will receive Intel Packaging solutions. At the intel 20A process node, the important customer of Intel’s foundry will be Qualcomm, and the Snapdragon mobile phone chip in 2024-25 may use Intel’s foundry.
Finally, Intel stated that there will be more important news on Intel Innovation on October 27th.
Posted by:CoinYuppie，Reprinted with attribution to:https://coinyuppie.com/intel-20a-1nm-is-not-enough-intel-has-increased-the-process-node-by-one-unit-of-measurement/ Coinyuppie is an open information publishing platform, all information provided is not related to the views and positions of coinyuppie, and does not constitute any investment and financial advice. Users are expected to carefully screen and prevent risks.