Janus Henderson Investors: The end of Moore’s Law is creating winners and losers in the semiconductor space

Key takeaways

• Semiconductor companies have been scrambling to sustain Moore’s Law, the foundation of technology innovation, resulting in a shift in the structure of the tech sector.
• While US-based Intel, to which Moore’s Law traces its origins, maintained momentum, it has suffered from the decision to not supply chips for the iPhone, increased competition and not following the outsourcing trend.
• The many complexities faced by the semiconductor industry, including geopolitical tensions, have resulted in a shifting tech landscape, presenting investment risks and opportunities.

According to Moore’s Law, the density of transistors on an integrated circuit can roughly double every two years. This relatively simple observation in 1965 by Gordon Moore, one of the founders of Portland, Oregon-based Intel, has been the foundation for all of the innovation in technology we have seen in half a century since then. The ability to increase performance while cutting the cost, leading to better, faster and cheaper products has been the key enabler for the development of technology and technology companies’ share gains. Not long after Moore’s Law was first defined, man landed on the moon for the first time and the entire computing power of NASA at the time has been available in a single iPhone while the latest NVIDIA Ampere (micro)chip used in artificial intelligence training crams 54 billion transistors onto a chip less than three square centimetres in size.

The demise of Moore’s Law has spurred innovation in technology

In a dynamic sector such as technology where there are few constants, the fact that Moore’s Law has endured is truly remarkable. However, the challenges of sustaining Moore’s Law have long been debated; Jensen Huang, the CEO of semiconductor giant NVIDIA, is only the latest to proclaim its demise and it is one of the reasons he is pursuing the acquisition of rival ARM Holdings. In most cases to date, the challenges have been met with innovative solutions. When ASML was struggling to meet the lithography resolution requirements required to print these extraordinarily fine circuits as they came up against the physical limits of light itself, it led to the novel solution of using the higher refractive index of water to reduce the blurriness, creating immersion lithography. 15 years ago, Dennard scaling ̶ the ability to reduce power consumption in-line with Moore’s Law – started to break down.

An Intel processor used to be one big core with the frequency, or the processing speed, rising exponentially every year. But since the clock rate of an Intel Pentium 4 in 2002 reached 3Ghz (or 3 billion cycles per second), an extraordinary gain compared to the first commercial PC in 1974 that used an Intel 8080 at 2MHz (2 million cycles per second), gains have been much more gradual. Today, Intel processors only just reach 5GHz. The issue has been that at such nanometre geometries the transistors are so close to each other that there is electrical leakage resulting in major heat (thermal) issues capping the performance. An associated issue is ‘dark silicon’ which is the portion of transistors that cannot be powered given these thermal constraints. The solution has been to build chips with lots of smaller cores rather than one big core and that is the multi-core architecture used today by all Intel, Advanced Micro Devices (AMD), NVIDIA or Apple processors. The recently announced NVIDIA/ARM deal marries multi-core leaders in graphics processing units (GPUs) and lower power, mobile central processing units (CPUs).

At Intel, we have a problem

Despite the challenges above, Moore’s Law kept going notably under its historical steward, Intel, which led the industry in terms of the smallest transistors and the most powerful computing chips. Part of that unique capability was in contrast to US and Japanese peers, which increasingly outsourced semiconductor manufacturing to foundries. For Intel, the commitment to producing as well as designing their own processor chips remained sacrosanct. However, problems began to emerge in 2013/14 at the 14 nanometre manufacturing node, where once again Intel was first to market but suffered some delays. With limited competition, it was not such an issue in the PC and server markets that Intel supplied. However, their reported decision to turn down making chips for the iPhone back in 2005 laid the foundations for future trouble. Intel were not alone in significantly underestimating the success of the iPhone. Fast forward to today and Taiwan Semiconductor Manufacturing Co. (TSMC), the world’s largest foundry, makes over 200 million iPhone processors for Apple each year at the very limits of Moore’s Law. Apple’s innovation cadence, the scale of the smartphone market and the substantial profits TSMC has reaped meant that when manufacturing issues really hit Intel at the 10 nanometre node (resulting in a four-year delay), Intel’s competitive challenges were much more serious. AMD, Intel’s main competitor, followed the trend of outsourcing production first by spinning off its manufacturing arm into GlobalFoundries, and then after several more years of disappointing execution, made a momentous decision to shift production to TSMC even at the cost of a significant exit payment to GlobalFoundries.

As Moore’s Law hit the buffers at Intel in Oregon, over in Hsinchu Science Park in Taiwan, TSMC continued to release a new iPhone chip each year like clockwork, and now enables AMD to produce chips that match the performance of Intel for the first time. With Intel recently announcing delays to it 7 nanometre process, investors are now contemplating the demise of Moore’s Law at Intel, a shift to an outsourced foundry model, the likelihood of significant market share gains for AMD for years to come, as well as NVIDIA’s aspirations to take share from Intel in the datacentre following the proposed ARM acquisition.

Key questions for the semiconductor industry

The current situation has the ability to reshape not just the semiconductor but the technology sector in its entirety, especially when considering the escalating US-China tensions. Has Moore’s Law finally run out of steam? Or is this just an Intel problem? Will TSMC be the future steward of Moore’s Law? Or will there be a new ‘Morris Law’ named after Morris Chang, the founder and long-time CEO/chairman of TSMC?

There is a lot of evidence that, at the very least, Moore’s Law as we know it will never be the same again. Manufacturing a chip with billions of transistors at single-digit nanometres is incredibly challenging. There is also a lot of propaganda among the protagonists with TSMC currently producing at 5 nanometres while Intel is at the more cumbersome 10 nanometres. Despite this, most independent observers think TSMC is roughly on par with Intel today. Also, TSMC’s main customer at 5 nanometre is Apple, and the size of the processor used in an iPhone is very different to those used in servers by a factor of five to ten times. Yielding a larger chip is exponentially harder, which is why the number of leading-edge customers at TSMC has dwindled to pretty much only Apple, Qualcomm and Huawei/HiSilicon. AMD and NVIDIA use TSMC’s 7 nanometre process, while AMD goes even further to overcome the challenges by chopping up the chip itself into ‘chiplets’ and then reconnecting them via advanced packaging. As we reach the limits of both Moore’s Law and the capabilities of lithography, Intel has also announced it is going down this route from 7 nanometre. These smaller ‘chiplets’ or ‘tiles’ as Intel refers to them, are relatively easier to manufacture, albeit at the expense of higher cost and performance compromise.

At Intel, we have a plan

Intel publicly remains committed to manufacturing chips internally; that is engrained in the company’s DNA. While Intel outsources many chips already, core processors are made internally. Investors have questioned this commitment given the ongoing delays and recent management commentary that they have a Plan B for 7 nanometre if further delays occur, namely outsourcing to TSMC. There are a few factors to consider here. Intel wanted to avoid repeating the 10 nanometre debacle where customers that rely on Intel for their product roadmaps grew increasingly exasperated as their plans were severely disrupted for years. Having a firm back-up plan is intended to reassure customers. Intel has also said that this shift to tiles enables them to be more flexible in outsourcing as it affords them the option to outsource one tile, rather than the whole chip, if needed. There are also geopolitical considerations. The current US administration as well as potential future occupants of the White House is committed to strategically supporting the US semiconductor industry, notably the ability to manufacture leading-edge chips. Being reliant on a foundry on an island off the coast of China likely keeps many US CEOs as well as the Pentagon awake at night. Currently, the US leading-edge semiconductor manufacturer is Intel, with Samsung having a smaller fabrication plant (fab) in Austin, Texas, and TSMC planning an even smaller fab in Phoenix, Texas, in 2024. The idea that Intel would outsource completely to TSMC with the associated national security issues and thousands of high-end semi engineering job losses involved seems unlikely. While not impossible, especially if Moore’s Law continues to be challenged at Intel, Plan A for the company appears to be getting back on track internally with Moore’s Law.

Facing both challenges and opportunities

For TSMC, a significant opportunity continues to present itself in high performance computing, which is needed as the phenomenal growth the company has seen in smartphones over the past decade subsides, albeit with a near-term 5G boost. For the next couple of years TSMC appears to have a strong claim to be the steward of Moore’s Law. Hopes for full scale Intel outsourcing are likely overly optimistic. Intel’s challenges create opportunities for customers like AMD or NVIDIA/ARM to gain market share, to the likely benefit of TSMC, while there could be incremental opportunities at Intel, some likely temporary, but others more permanent.

TSMC also needs to continue to walk the tight rope of escalating US/China tensions. The US government’s recent new restrictions specifically targeted to sever the TSMC and Huawei/HiSilicon relationship* were the latest reminder of how impactful geopolitics can be. TSMC has already built a fab in China and has now committed to building one in the US, while its competitor in China, Semiconductor Manufacturing International Corporation (SMIC), has recently announced substantial new spending to build more capacity to serve domestic customers. It does however run the risk of being added to the US entity list, which requires a license for any product where 25% or more of the intellectual property is of US origin. TSMC will be navigating these opportunities and risks for years to come.

Summary

These complexities are a reminder that in the dynamic world of technology, where stock indices reward past rather than future success, active management of technology stocks remains important. It was not long ago that Intel’s market capitalisation was higher than TSMC’s, but now TSMC is close to double that of Intel’s. At the same time, NVIDIA recently surpassed Intel, and AMD is half-way there.

The shifting landscape of the semiconductor industry seems ripe with opportunities for investors, providing these are viewed through the right lens.

*In May 2020 the US government announced rules that targeted Huawei and its chip subsidiary HiSilicon, requiring a licence for any shipments from manufacturers that use US technology or equipment.

Glossary

Clock rate is an indicator of a processor’s speed. It refers to the frequency at which the clock generator of a processor can generate pulses which are used to synchronise the operations of a processor’s components. In general, a higher clock rate means a faster central processing unit (CPU).

Immersion lithography is a photolithography resolution enhancement technique for manufacturing integrated circuits (ICs) that replaces the usual air gap between the final lens and the wafer surface with a liquid medium that has a refractive index greater than one. The resolution is increased by a factor equal to the refractive index of the liquid.

Dennard scaling postulates that as transistors get smaller their power density stays constant, so that the power use stays in proportion with area. This allowed CPU manufacturers to raise clock frequencies from one generation to the next without significantly increasing overall circuit power consumption.

A GPU or Graphics Processing Unit performs complex mathematical and geometric calculations that are necessary for graphics rendering.

A CPU or Central Processing Unit is the primary component of a computer that processes instructions. It runs the operating system and applications, constantly receiving input from the user or active software programmes.