Long-term trends in the technology segment are often the result of structural change. One such change in the semiconductor space was the shift to outsourcing specialized functions, including technology development in the form of intellectual property, chip design, fabrication, and packaging and test.
That change, along with the move to more specialized workloads and the need for original equipment manufacturers and service providers to differentiate, has sparked another trend, toward customized processors and System on Chip solutions.
The fragmentation of semiconductor design and manufacturing began several decades ago with outsourcing of packaging and final test because it was viewed as a low-level function that provided little differentiation or value.
As of late, however, using multiple dies in 2D and 3D structures has made packaging a valuable function and a new pillar of innovation in the semiconductor industry. (The other three are lithography, transistor design and material technology.)
Over the past two decades, most semiconductor companies moved to outsourcing the front-end die manufacturing, referred to as “fabrication,” to companies like GlobalFoundries, Samsung, TSMC, UMC and SMIC.
Outsourcing the fabrication allowed companies to share the high-cost and capital costs of fabs and new process development, both of which have increased exponentially with each new generation. As a result, most semiconductor companies are now referred to as “fabless semiconductor manufacturers.”
Many companies have outsourced semiconductor design to specialty design houses or have begun licensing IP from other companies. This allows companies to share the expense of the design experts and have access to designers with expertise in specific areas. It has resulted in a huge growth in IP licensing.
Some companies, like Arm, originally adopted this model out of necessity, but others have added IP licensing to their business model. Companies ranging from the large integrated design manufacturers like Samsung to SoC vendors like Qualcomm, to electronic design and automation tools vendors like Cadence, now license various processing cores, on-chip interconnects, and other forms of semiconductor IP. As a result, it has become easier than ever to design a custom chip or to outsource its design.
Custom Chip Development
In addition to the relative ease of chip design, the need for more specialized chips with higher performance and efficiency has been pushing many companies to develop their own custom chips. The most prominent segment is cloud services. Large cloud service providers like Amazon, Google and Microsoft long have optimized their software, systems and even entire data centers.
The last area for those companies to optimize is at the chip level. By developing chips that are designed for specific functions like data mining, Web services, or artificial intelligence, companies can reduce their operating expenses while improving efficiency through lower power consumption, higher data throughput, and denser system and data center configurations.
All this has a significant impact on the return on investment for capital expenditures. However, the cloud is not the only area where we see custom silicon.
We now see consumer electronics companies also following this trend. The three largest smartphone vendors — Samsung, Huawei and Apple — all design their own chips, and some of the smaller vendors have begun considering it as well.
Even in specialized segments like military and aerospace, companies have been moving toward this model because of the limited options they have from off-the-shelf products. The push for autonomous vehicles also has been driving automotive equipment companies and OEMs, like Tesla, to develop specialized chips for the demanding AI functions that will be required.
In the early days of the electronics industry, it was not uncommon for companies to develop their own chips because the companies were vertically integrated, like the old Henry Ford manufacturing model. Today, however, companies are using custom chip to optimize and differentiate.
While there will still be a market for off-the-shelf semiconductors, the highest growth is in the area of customized semiconductors that are being designed either in-house or with design partners. Note that while it is still expensive to begin manufacturing a new chip — often tens of millions or more — the benefit outweighs the investment for a growing number of companies.
The New Semiconductor Startup Mantra: ‘Buy Me’
Another structural change in the semiconductor industry has been consolidation, especially over the past five years. It began with the rush to acquire technology and solutions for more advanced SoCs for the emerging Internet of Things era, but it has continued with the excitement around autonomous vehicles and now AI.
This trend is likely to continue in other areas, such as radio frequency, with the growth of 5G and other wireless interconnects. This structural change, however, also has driven a change in the business model of startups. Fewer semiconductor startups aim to build the company into the next Arm, Intel or Qualcomm. Instead, they are just interested in getting far enough along to sell the technology, company, or both.
A major trend in semiconductor consolidation began in 2014 with the IoT hype and NXP’s acquisition of Freescale to form a formidable force in the semiconductor industry, especially in embedded solutions ranging from consumer products to industrial applications.
A second wave hit with the growing interest in autonomous vehicles. Suddenly, everyone from Intel to Samsung had automotive groups focused on autonomous driving.
A third wave was starting to get under way with AI, and likely RF, around the rollout of 5G, but recent geopolitics stalled any major mergers or acquisitions. The two most notable ones would have been Qualcomm’s acquisition of NXP, which was blocked by the Chinese government, and Broadcom’s acquisition of Qualcomm, which was blocked by the U.S. government.
However, this consolidation trend also resulted in a mindset change in most semiconductor startups. It has never been uncommon for semiconductor startups to be acquired — in fact, there are many serial entrepreneurs who have started semiconductor startups. Unlike before, though, very few appear interested in growing these startups into the next leading semiconductor company.
This is most evident in the plethora of semiconductor startups focused on artificial intelligence. With a huge investment in AI and machine learning, there are hundreds of entities developing new AI cores and chips, most commonly referred to as “neural processing units,” including many new semiconductor startups.
Shortsighted Business Plans
The business plans of most semiconductor startups today involve developing a chip or technology and then selling the company or its intellectual property to an established IP, semiconductor company, or OEM. In fact, many of the startups are banking on being purchased by the companies that provide some of their funding, which doesn’t always happen.
I find this trend disturbing. I find it hard to believe that a major cloud service company or OEM would be willing to risk its future on a startup that has no plan to function as an ongoing entity. As a result, I believe that most of the semiconductor startups in the market today will fail because of the growing competition from in-house resources and larger established semiconductor companies.
Startups always have provided valuable innovation, and they have led to change in the semiconductor landscape coinciding with major inflection points in the market. However, the lack of vision of many of these startups to become the next innovator or leading semiconductor company is disturbing and disappointing.
I will agree that it is a challenge to build a new semiconductor company, but with the market inflection points being driven by new technologies like AI, 5G and cloud computing, there is an opportunity today for new semiconductor companies to emerge and be the next generation of industry leaders.
It doesn’t stop there, with technologies like quantum and neuromorphic computing in the future. When the pre-Socractic Greek philosopher Heraclitus said that “change is the only constant,” he was referring to life, but it also applies to technology. The industry needs the innovative energy that startups provide. Hopefully, we will soon see more startups willing not only to invest in new technology, but also to build a long-term business.