Can Intel overcome TSMC to be the world’s strongest lithography machine?
In October 2020, Lee Jae-yong – “The Crown Prince of Samsung” – had a particular business trip to the Netherlands. He decided to do so for only one reason – meeting the ASML’s Board of Management to speed up the transaction of 9 extreme ultraviolet lithography machines (EUV) to Samsung. Without those vital systems, TSMC can further surpass the Korean company in semiconductor technology. However, about three months later, Lee Jae-yong was accused and sentenced for bribery.
Questions about ASML have started to rise. “What kind of machine do they produce that various tech companies willingly pay a huge amount of money on?” The answer is simple but will shock many people: ASML holds one of the most crucial factors in manufacturing semiconductor chipsets – photolithography machines. The device can ‘engrave’ on silicon thin films using light, creating the nanoscopic features into microchips. Developing such technology is costly, explaining why these photolithography machines can cost 25% of total expenses in a chipset production line. With the price of $150 million each and many logistics required for transportation, not all companies can afford this high-end technique. Only three leading-edge chipmakers hold most of the amount of it – Samsung, Intel, and the world’s largest foundry – TSMC. The Taiwan-based company leads the semiconductor race with the state-of-the-art EUV machines bought from ASML. That explains how important ASML is in the world’s high-end chipsets production chain. This Dutch-based company’s position in semiconductor manufacturing is irreplaceable. Even the representative of this company once stated: “If we cannot distribute these photolithography machines, Moore’s Law will suffer from stagnation.”
Despite such bold moves from Samsung, the most active in developing the business with ASML is Intel – the counterweight of TSMC. In this year’s January, Intel has surpassed TSMC, acquiring the latest and first High-NA photolithography machine from ASML with an aperture number of 0.55. Since the new CEO of Intel, Patrick Paul Gelsinger, took office, he has announced that this company will invest more than $100 billion in building foundries in the US and EU. Also, Intel has successfully closed the deal of buying the Israeli company Tower Semiconductor. Over the past few years, Intel has tried its best to take back its place as the number one company in the semiconductor manufacturing business. Still, the alliance between TSMC and AMD is not that easy to be eliminated.
The fall of the king
Can Intel fulfil their mission of defeating TSMC and revive its golden era in the past? We must first understand how Intel fell from the top and lost its position to answer this question.
In 2014, Intel was the name that dominated semiconductor manufacturing with its latest 14nm manufacturing process. With the great profit from the microchip foundry industry, Intel had poured in a massive investment in the foundry business, leading the world in the density of transistors in various microchips. But the first place of Intel was soon lost to the hands of TSMC and AMD’s alliance. The root of all Intel’s downfall lies in the IDM model that they had steadfastly pursued for many years.
Integrated Device Manufacturer – as Intel defines its business – covers the three essential triads of semiconductor manufacture: design, manufacture, and sell its integrated circuit products. This model benefits from having a strong production capacity and the comprehensive implementation of the company’s strategies. However, it has a critical drawback: It demands a high investment, and the production would be longer than other competitors.
From Intel’s perspective, the design department is responsible for designing and planning the process update, and the technology and manufacturing (TMG) department is responsible for implementing the process. Until the 14nm process, the tick-tock rhythm proved to be still stable and tight. But after that, TMG started focusing on improving transistor density rather than manufacturing processes. That is why the appearance of “strange phenomena” such as 14nm +, 14nm ++, 14nm +++, while mass production of the 10nm process is far away.
Meanwhile, TSMC is pioneering the foundry and “all-in” model of the semiconductor manufacturing process, overtaking Intel.
The resurrection of AMD and TSMC
Intel’s miss also saved the life of an old rival of the company, AMD. Before 2016, AMD was almost entirely defeated by Intel. According to data from Mercury Research, Intel accounted for 92.4% of the PC CPU market share and 99.2% of the server CPU market share in 2015. But at the decisive moment, TSMC’s process increased. Becoming more advanced, AMD decided to abandon the IDM model, sell its foundry, then invest in TSMC’s 7nm process.
Finally, with TSMC’s advanced process and performance-to-market ratio, AMD has recovered with a share of around 40% of the CPU market by 2021.
Faced with the backward situation, Intel also cannot sit idly by. In 2020, then-Intel CEO Bob Swan reorganised the technology division to decentralise and pushed for the abolition of the IDM model. The company also wanted to transfer the semiconductor manufacturing process to the TSMC foundry, but this plan was scuttled before it was implemented.
A year later, the new CEO of Intel, Patrick Paul Gelsinger, took office. He decided to expand production capacity and licence the x86 architecture to win back customers, aiming to take down TSMC.
The foundation for Intel’s counterwork plan
In theory, as a long-standing giant in the semiconductor industry for more than 50 years, Intel has the advantage of accumulating technology and resources.
In terms of process technology, most people give the general impression that the smaller the number of nanometers in the process, the more advanced the process. But, the core metric to really measure chip technology that you should look at is “transistor density” – only the higher the density, the higher-end the new product.
Therefore, from the outside, Intel seems to be at a disadvantage when it is constantly crushing itself on the 14nm process and delaying mass production of the 10nm process products until 2019, while TSMC has already stepped up, developing into the 7nm process. But if you compare the transistor densities of the two, they tell a completely different story.
The transistor density of the 10nm Intel chip is 100.8 MTr/mm2 (each square millimetre has 100.8 million transistors), and the transistor density on the 7nm TSMC chip is 96.49 MTr/mm2. Thus, Intel’s 10nm process technology level is comparable to TSMC’s 7nm.
In other words, what Intel’s TMG division did then wasn’t entirely in vain. Intel’s technical level is still at the top of the industry and can completely match TSMC.
In terms of resources, the open licensing of x86 architecture to the outside is also considered a drastic move. According to Bob Brennan, vice president of engineering for Intel’s newly formed Foundry Services (IFS) division, this is “the first time in Intel’s history that both the hard and soft cores of the x86 architecture will be licensed. for customers who want to develop chips”.
Simply put, x86 is the mainstream chip architecture standard, holding an absolute position in the PC market today, equivalent to the place of Microsoft Windows in computers. Intel is in the leading role in all CPU markets with this architecture. Only Intel, AMD and VIA Technologies (Taiwan) have the right to use the x86 architecture.
Intel has decided to sell “entrance tickets” to everyone with access to x86. Of course, that will come with the condition that customers want to be licensed. After designing the chip, they need to find a foundry in Intel’s wafer division. And this will undoubtedly have an impact on orders using TSMC’s ARM architecture.
Compared to the internal factors, Intel also has another advantage for this counterattack: it is an American company. This will give Intel dual perks: core customers and critical devices.
However, despite all of these advantages, Intel’s rise to the semiconductor industry’s top is still unlikely. Because it still has potential unresolved concerns.
Obstacles to Intel’s Plan
When Intel’s new CEO Patrick Gelsinger took office, he gave various reform signals. But this CEO also did what is considered to be ignoring the past mistakes, which is to bring back the IDM model that his predecessor wanted to abandon.
While this IDM2.0 plan will outsource some advanced manufacturing capacity from third-party foundries, Intel will enhance its technology and R&D roadmap, developing its own products. This also means that Intel will adhere to the IDM model from design, manufacturing, packaging and testing.
In this way, Intel will face two potential worries. On the one hand, it was the inherent slowness of the IDM process in the past. On the other hand, this model also makes Intel “enemies everywhere” globally.
Regarding the first, IDM means that when a processor upgrade is needed, the circuit diagram, the wafer foundry, and the factory responsible for packaging and testing must be upgraded simultaneously. And there may still be problems of poor cooperation between departments in the past.
You must know that the more advanced the semiconductor process, the more difficult it is to implement and the more expensive the core device will be. In addition, it will undoubtedly be much higher than a TSMC that only specialises in production in terms of cost. It can be said that following the IDM model has determined that these huge costs need to be borne entirely by Intel.
In contrast, its rival TSMC simply focused on the foundry and didn’t need to repeat the design, packaging, and testing process. As a result, it will have a considerable capital advantage and can run faster on its way to more advanced manufacturing processes. In 2021, TSMC’s capital expenditure was around $30 billion, while Intel, which has higher revenue, was only $17.9 billion.
In short, TSMC is vital because it has formed an “active” cycle for approaching advanced semiconductor process technology. It can negotiate products and then invest all of its profits in more advanced process technology, thereby maintaining its process leadership again.
And Intel, which is steadfast in the IDM model, will find it challenging to beat TSMC because of its disadvantage in capital and technology.
The answer for the labour problem
Last but not least, Intel will have to pass through the labour shortage challenge. The entire semiconductor manufacturing industry has a unique feature: its heavy reliance on technical talent. In essence, the technological process strength of a foundry is accumulated by the engineers’ know-how.
And what is the current status of talent in the semiconductor manufacturing industry in the US, the headquarters of Intel? The answer is: Not optimistic.
During the 1990s, American semiconductor companies entered the process of a vertical division of labour, especially after TSMC created the division of labour model in wafer manufacturing. It caused many semiconductor giants such as Texas Instruments and AMD to withdraw from the wafer manufacturing process, and the focus of the semiconductor manufacturing industry began to shift. Asia has two representatives, Korea and Taiwan.
The result of outsourcing foundries is the dismissal of talented and professional engineers.
Obviously, in the context of other professions in the US, such as real estate, trading and stock speculation, finance raising, a lack of vision and interest in the chip industry is inevitable. Intel’s future is not bright whenever “American elites are still on Wall Street”.