Intel’s failure, can the semiconductor tick-tock continue in the United States?

Some analysts have pointed out that if the end of Moore's Law is a wall, the first person to hit it will be the person walking in front.

After being delayed for three consecutive years on the 10nm process, Intel's 7nm process suffered another setback. The launch of the first batch of 7nm processors was delayed by six months, and the first batch of 7nm client processors was delayed from the original fourth quarter of 2021 to the end of 2022 or early 2023. The giant that was once at the forefront of semiconductor technology has hit a wall, even though Moore's Law has not yet come to an end.

However, what shocked the industry even more was that Intel CEO Bob Swan said that after careful consideration, he would consider outsourcing some high-end chip manufacturing.

Tick-Tock is suspended?

Tick-Tock is a development strategy model proposed by Intel for developing the microprocessor chip design and manufacturing business. Intel pointed out that the timing of each processor microarchitecture update and each chip process update should be staggered to enable the microprocessor chip design and manufacturing business to develop more efficiently. The name "Tick-Tock" comes from the sound made when the second hand of a clock moves. The company said that each "Tick" represents the update of the processor chip process of a generation of microarchitecture, which is intended to reduce the chip area, energy consumption and heat generation while the processor performance is almost the same; and each "Tock" represents the update of the microprocessor architecture and the improvement of performance based on the chip process of the previous "Tick". Generally, the cycle of a "Tick-Tock" is two years, with "Tick" accounting for one year and "Tock" accounting for one year, that is, one year to update the process and one year to update the architecture.

This strategy is often jokingly called the "toothpaste squeezing strategy" by many computer players, because the performance gap between each new generation of processors and the previous generation is very short. After entering the 14nm process in 2016, Intel slowed down the Tick-Tock cycle to three years, adding optimization links, further slowing down the actual update speed, and making each cycle a process, architecture, and optimization (Tick-Tock-Optimization).

Intel, regarded as a benchmark in the global semiconductor industry, has always adhered to the "two-pronged" IDM model of chip design and production since its establishment, especially in the field of process technology. The industry leader is usually determined by who reaches the most advanced process node first. Intel's process technology was once far ahead of others, and TSMC, Samsung and others could only catch up from behind, until TSMC and even Samsung surpassed it in 10 nanometers.

An industry insider pointed out that there are three main reasons why Intel went from leading to keeping pace or even lagging behind in 10nm. "First, Intel did not follow Moore's Law to increase transistor density by 2 times at the 10nm node, but by 2.7 times, which is very risky. They have also publicly claimed on many occasions that their transistor density is much higher than that of their competitors at the same process node. Although the density is high, Intel has not solved the problems of heat generation and energy consumption. High density generates serious heat and has low high-frequency stability. Secondly, Intel did not choose to use EUV like TSMC, but continued to use ArF DUV, combined with double exposure and quadruple exposure to increase transistor density. However, if multiple exposures produce a slight offset, it will have disastrous consequences. For this reason, Intel also used SAQP technology, but an obvious disadvantage of DUV + SAQP is that the edges of the graphics are not as clear as EUV."

In addition, other experts mentioned the introduction of cobalt (Co) to replace some copper wires. The use of cobalt as a new material will bring more benefits, including improving electromigration performance by 1,000 times, and reducing interlayer through-hole resistance by half, which will inevitably greatly increase the durability of the chip. Applied Materials, which first studied this technology, once said that cobalt is the biggest change in the Internet in 15 years. However, it takes too much time to put this material into practical use, it is too expensive, and it is not necessary.

As an absolute leader in semiconductor technology, Intel has been the first to try almost all important innovations in the past few decades, such as HKMG materials, Gate-first or Gate-last, and the transition from planar transistors to three-dimensional FinFET processes. However, this time Intel's radical strategy of simultaneously adopting multiple technological innovations on 10nm finally hit a wall.

In fact, before 10nm, Intel had been stuck on the 14nm process for five years. Starting with the 180nm process in 1999, Intel had been updating its process every two years until 2014. In July 2015, Intel announced that the 10nm CPU originally planned to be launched in 2016 would be postponed to the second half of 2017.

At this point, Intel's Tick-Tock finally came to a halt, while TSMC reached the top relying on the still immature EUV.

Will Intel divest its foundry business?

The end of the Tick-Tock strategy has forced Intel to reflect on its stubbornness of "two-pronged approach" of chip design and production. Bob Swan tactfully and with difficulty revealed that he was considering using a third-party foundry. Then it was reported that Intel had reached an agreement with TSMC to start mass production of processors or graphics processors using TSMC's 7-nanometer optimized version of the 6-nanometer process next year, and booked 180,000 6-nanometer production capacity from TSMC next year. In addition, the industry also speculated whether Intel would divest its foundry business?

In an investor conference call at the end of last month, Bob Swan said that the 7nm process was delayed by 6 months than expected, mainly due to yield ramp-up, which was 12 months slower than planned. "Intel detected a defect model on the 7nm process, which caused the yield to drop. Intel said it has fundamentally solved the problem and believes that there are no fundamental obstacles (in terms of yield improvement). Despite this, Intel has prepared contingency plans to avoid other uncertainties that may occur."

This emergency plan is to be more pragmatic in process research and development, and consider using self-manufacturing, third-party foundry, or a mixed model of "manufacturing + outsourcing" to produce chips.

Intel emphasized that although the process technology is very important, it is only one of the six innovation pillars that drive Intel's product differentiation. Bob Swan described the outsourcing to TSMC as a "contingency plan." In Intel's view, in an ideal world, Intel's best choice is to have leading products based on its own process technology, which also benefits from the economic advantages brought by IDM-type companies. However, the company's focus is still "leading products", so the current focus is to ensure that leadership products updated every year are delivered to customers in a controlled manner.

Since Intel has repeatedly encountered delays since its 14nm products, how can it ensure that it continues to provide customers with leading products at the previous cadence of one generation per year? Bob Swan believes that the lesson Intel learned from the 10nm process is whether and when it should be manufactured internally or outsourced to ensure that there is still an option of combining internal manufacturing and outsourcing. "This is the lesson we learned from the 10nm process, which is to look at internal manufacturing and using third-party foundries more pragmatically. If the process does not progress as expected, we must ensure that we have contingency plans in advance to provide more options and flexibility." He explained, "The real milestone is that we can switch strategies and use third-party foundries as much as possible when needed, and will not delay product progress due to process complexity. This is a strength, not a weakness, which allows us to be more flexible and make the most informed decisions about effective ways to build products."

It is not difficult to find from Bob Swan's words that the annual important processor product updates are Intel's top priority. Now it is no longer obsessed with whether this advanced product is manufactured by itself or by others.

This is also true. As the process shrinks, fewer and fewer companies can afford or need the most advanced process nodes. Although performance and power consumption remain key elements of design, there are still multiple ways to achieve the goals. Industry insiders point out that more of the benefits gained from process shrinkage in the past will now be increasingly transferred to areas such as advanced packaging. In these areas, Intel's embedded multi-die interconnect bridge (EMIB) and FOVEROS 3D die stacking technology do have potential significant advantages.

The news indirectly confirmed this. The media reported that Intel reserved TSMC's 6nm production capacity and also pointed out that after the chip is manufactured, the packaging and testing still have to return to Intel's own packaging factory. Bob Swan said that Intel has reduced the impact of process delays on product progress through improvements in design methods including die disaggregation and advanced packaging.

However, some analysts still believe that Intel will sell or should sell its foundry business. Jefferies, an investment firm, said in a report that in this case, moving towards fabless will help Intel generate higher cash flow and capital returns. Susquehanna also suggested that Intel sell its own wafer fab to TSMC because it no longer has any chance of surpassing TSMC in the competition for high-end process technology.

Morgan Stanley pointed out that compared with AMD chips produced by TSMC's 7-nanometer process, Intel's 14-nanometer chip performance is still quite competitive in the market. And the company still occupies a leading position in the core PC and data center processor market. This shows that process advantages are not the only factor that dominates the market.