Will Intel, which squeezes out toothpaste, have a day when it goes “crazy”? – Fast Technology – Technology changes the future

Two months ago, when a truck carrying parts for the world's first 0.55NA EUV lithography machine appeared at Intel's factory in Oregon, the semiconductor industry began to discuss the possibility of Intel catching up with TSMC in chip manufacturing processes in the future.

However, Intel has revealed to the world that its ambitions go far beyond that.

In the early morning of February 22, Beijing time, Intel held its first Intel Foundry Direct Connect conference. Unlike previous events that focused on product display and technology sharing, this conference has only one theme – chip foundry.

Image source: Intel

In the semiconductor industry chain, chip foundry is an out-and-out “important and boring” link, but any one of the highlights displayed at this event is enough to surprise:

The Clearwater Forest Xeon processor using Intel 18A process (1.8 nanometer level) has completed tape-out, and the 1.4 nanometer Intel 14A process has made its debut.

The first system-level foundry model for AI chips is launched, providing full-stack optimization from factory network to software.

It has successfully won a chip order from Microsoft and will use Intel 18A for foundry.

IP and EDA partners for foundry services, including Synopsys, Cadence, Siemens, Ansys, Lorentz, and Keysight, indicate that the tools and IP are ready.

It is worth mentioning that Intel also announced a cooperation with ARM at the event. In the future, it will support the provision of foundry services for SoC (system-on-chip) based on the ARM architecture. This strong sense of dissonance made even those who came to the platform ARM CEO Rene Hass said jokingly:

“It feels as weird as running iTunes on Windows.”

In the past week, when the world was discussing Sora's revolutionary progress, Intel quietly came up with a complete solution to the explosion in AI computing power demand, once again proving the status of the former chip industry overlord.

IDM 2.0, a desperate situation?

When Intel is mentioned, people usually compare it with NVIDIA or AMD, but in fact, Intel's development path in the semiconductor industry is completely different from the latter two. One of the most essential differences is that Intel is one of the few adopters in the industry. IDM mode chip manufacturer.

The so-called IDM refers to everything from design, manufacturing, packaging and testing to selling private brand ICs.

Correspondingly, there are Fabless manufacturers such as Nvidia and AMD. They only do chip IC design, and the chip manufacturing process is completely completed by foundries. The other type is Foundry manufacturers represented by TSMC and GlobalFoundries, which are responsible for the foundry and packaging of chips.

The advantage of this division of labor model is that Fabless manufacturers do not need to bear the multi-billion dollar production line construction costs, and Foundry manufacturers do not need to maintain a large product development team and can focus on improving the process.

This clear division of labor and efficient upstream and downstream relationships in the industrial chain are the prerequisites for Moore's Law to continue to exist today. Therefore, chip manufacturers that are still using the IDM model today are actually “non-mainstream” to a certain extent.

Intel's business model is even more “evil”.

After taking office in 2021, current CEO Pat Kissinger proposed the “IDM 2.0 model.” That is to say, while insisting on producing its own chips, it also provides foundry services to third-party chip design companies, and at the same time hands over some process chips to other foundries to supplement its own processes.

This idea sounds wonderful, but even people who don’t understand the chip industry at all can detect a clue: third-party chip design companies, their products may already compete with Intel, so how can they safely manufacture chips? What about leaving the link to the latter?

Will Intel, which squeezes out toothpaste, have a day when it goes

The world's first 0.55NA EUV lithography machine received by Intel. Image source: Intel

To use an analogy, this is like Apple saying to Microsoft, “We are upgrading the MacOS system, and if necessary, we can also help you develop the Windows system.”

Another issue that cannot be ignored is that the chip foundry industry is a typical “asset-heavy, long-cycle” industry. If the construction can take over the production capacity of third-party chip design companies, it will inevitably increase the initial cost significantly. This was last year's This has been reflected in multiple financial reports. For example, in the second quarter of 2023, when Intel's overall operating profit margin was 33%, the profit margin of its IFS department responsible for the chip foundry business was -28%, which seriously dragged down the group. Performance.

Therefore, after the IDM 2.0 concept was proposed, doubts from the outside world have never stopped.

In June last year, after realizing that the IFS department had become a heavy burden, Intel chose to spin off the department. It will operate independently in the future and separately account for profits and losses in its financial reports.

At the same time, the semiconductor industry also ushered in a huge change.

As the demand for computing power surges globally, TSMC, the main OEM responsible for AI chip foundry, has obvious production bottlenecks. For example, the monthly production capacity of CoWoS advanced packaging has remained at 13,000-15,000 pieces (wafers), while Part of this production capacity will be taken away by chips such as Apple's M1 Ultra. Therefore, in the second half of 2023, the delivery cycle of Nvidia's high-performance computing cards will generally be extended to 12-16 months.

Considering that there are currently only three manufacturers with advanced processes: TSMC, Intel, and Samsung Electronics, and Samsung Electronics’ yield problem after the 5-nanometer process node is still a mystery, Intel has become one of the few OEMs for chip design manufacturers. Optional.

This is also the reason why Intel proposed “AI chip system-level foundry” at this time.

The wealth is pouring into the sky, can Intel catch it?

Although Intel seems to be in the right place at the moment, does the company now have the strength to flex its muscles in the foundry industry? If we look at market share alone, according to statistics from TrendForce, Intel accounts for only 1% of the global chip foundry industry.

However, in terms of technology, Intel is still the leader in the industry.

The first is that its process is currently in a state of steady advancement. In July 2021, Intel announced a “four-year, five-process node” plan, which is to use four years to promote five process nodes: Intel 7, Intel 4, Intel 3, Intel 20A and Intel 18A, with the goal of 2025. Regain the process leadership of global chip foundry.

Judging from the current public information, two Intel 3-based products, Sierra Forest and Granite Rapids, have completed trial production after tape-out and design certification respectively. The “Am-level” Intel 20A and Intel 18A are also expected to be put into production this year.

Will Intel, which squeezes out toothpaste, have a day when it goes

Intel process node planning diagram, picture source: Intel

It is worth mentioning that these two process nodes will be equipped with what may be Intel's two most important technologies in the next decade: RibbonFET gate full surround field effect transistor, and PowerVia back power supply network.

Simply put, the former can promote the further shrinkage of the transistor size on the chip; the latter transplants the chip's power and signal lines to the back of the wafer, thereby reducing power consumption.

These two underlying technologies jointly laid the technical foundation for Intel's foundry business and ensured that Moore's Law can still be continued.

Of course, there is little need for such sophisticated technology in the industry right now. After all, the main chip that AI practitioners rely on, the Nvidia A100, has a manufacturing process of only 7 nanometers.

In the short term, the AI ​​industry may need Intel's technology accumulation in the packaging process.

At this meeting, Intel announced that FCBGA 2D+ will be included in Intel’s foundry technology portfolio for advanced system packaging and testing. The portfolio will include FCBGA 2D, FCBGA 2D+, EMIB, Foveros and Foveros Direct technologies.

This series of obscure technical terms allows Intel to meet the heterogeneous integration needs of different foundry customers, and also allows “chips” from different suppliers and built with different process nodes to work better together.

It is necessary to explain the concept of “core chip” here: it can be simply understood as a small modular chip. Its significance to the AI ​​industry is that it can connect CPU, GPU, and NPU in the same system at high speed, greatly improving the transmission rate between heterogeneous cores, improving data access speed while reducing data processing power consumption.

Will Intel, which squeezes out toothpaste, have a day when it goes

Intel 2.5D packaging production line, picture source: Intel

In addition, Intel has also made full preparations for potential startup customers of AI chip startups, such as Groq and SiMa.AI, which have been in the spotlight recently. These companies generally believe that special-purpose chips are more efficient than general-purpose chips, so their product research and development also focus on building chips specifically for certain architectures. However, the problem is that these startups have a variety of chip layout designs.

Therefore, among the partners who came to stand for Intel at this conference, you can see that the “EDA Big Five” in the chip industry (Synopsys, Cadence, Siemens, Ansys, Lorentz) were all present to demonstrate Intel's contribution to various industries. Prepare for software and tool certification.

In general, whether it is process manufacturing, packaging technology, or development tools, Intel has initially possessed the capability of “system-level foundry of AI chips”, but can Intel really sit back and relax from now on?

Obviously not. One of the most direct hidden dangers is that although the chip foundry department will be independently accounted for in future financial reports, the early construction and debugging costs of the production line still need to be paid by Intel.

This will be an astronomical sum of money. Judging from the current project progress officially disclosed by Intel, the company will invest a total of US$62.6 billion in Poland, Israel and Germany in the future to build wafer foundry production lines. With funding from the chip subsidy bill delayed, these investments will almost entirely need to be borne by Intel.

Another point that must be mentioned is that Intel CEO Pat Kissinger has stated many times that the chip foundry department after independence will be completely separated from the chip design department. The latter will evaluate different solutions on the market when there is a need for foundry. , comprehensive cost selection of foundry, absolutely fair competition.

Kissinger did. Currently, the Computer Tile part of Intel's two processors, codenamed “Arrow Lake” and “Lunar Lake” internally, is handed over to TSMC.

But in the long run, if the core modules of Intel's own processors are all manufactured by TSMC, how can other Fabless manufacturers safely place orders with Intel? This is probably an issue that Intel's foundry department has to consider.