Industry-Shaking Changes Ahead

I’ve been closely following the increasing power of system OEMs (such as Apple, Google, Facebook and Microsoft) over players in the semiconductor industry with an eye toward understanding how this will affect others in the semi value chain. However, I haven’t paid as much attention to the root causes of why these changes are occurring. McKinsey & Company has provided our industry food for thought regarding semiconductor industry drivers with this week’s McKinsey Insight article, “Moore’s Law: Repeal or Renewal?”

For context, I’ve documented some of my thoughts regarding how this shift in power will affect our industry in a somewhat controversial EE Times article (Wake Up, Semi Industry: System OEMs Might Not Need You). Further analysis is provided in a keynote speech I delivered at the Semico IMPACT Conference (Download the slides from the article, “Can Intel Catch Samsung? Can Anybody Catch Samsung?”)

System OEM power over semi makers is not a pendulum swing!
My main point has been that the trend of system houses and OEMs designing or specifying their own chips is more than a “pendulum swing” between semi industry verticalization and horizontalization. All of us in the semi value chain must adapt to a new reality where fabless semiconductor companies will have less influence in the design of some leading-edge products in certain industries.

The two tables below from the presentation slides show the current financial state of industry players, with system OEMs in green, semiconductor companies in gray, and IP companies in white.

The gist of my prior arguments is that huge amounts of cash on hand and high operating margins are enabling companies such as Apple, Facebook, eBay, and Google to create their own chip designs or specify features and IP implementations to semiconductor manufacturers.

McKinsey: Four kinds of Semiconductor Innovation
The McKinsey article has a different take on this issue by looking at whether Moore’s Law will survive, and if not, what will the semiconductor industry look like. First, the McKinsey authors highlighted four sources of innovation in the semiconductor industry, which I show in Exhibit 1 from the article, below. I have highlighted three areas where innovation is occurring today in the consumer electronics, automotive and mobility markets.

Of course, Moore’s Law has driven the explosive growth in systems-on-chip (SoCs) design starts. But we are also seeing new technologies, such as 2.5D silicon interposers and 3D-IC/through-silicon vias (TSVs) that are not dependent on Moore’s Law. Increased integration, whether on-die or in-package, will continue to be a major source of innovation and differentiation. The question the McKinsey team poses is whether semiconductor industry growth will continue to follow the trajectory of Moore’s Law, or adopt a flatter growth profile.

EUV and 450mm wafers: 2 critical technologies to continue Moore’s Law
The authors then identify two critical technologies that could enable Moore’s Law to continue as we have known it. The first is innovations in lithography, particularly extreme-ultraviolet (EUV) lithography. The second is the industry transition to 450mm wafers.

What makes the adoption of these technologies unique in the history of our industry is the huge cost of development and adoption. In past technology node transitions, semiconductor companies were able to quickly cover their costs of capital when developing new processes and purchasing new equipment. With EUV and 450mm, the capital costs are so huge that only a few companies will be able to participate. And the break-even points for paying off these investments will be very far into the future.

There are two potential benefits to adopting EUV and 450mm wafers: lower cost and better performance. With previous technology node transitions, Moore’s Law has always delivered both lower cost and better performance. However, it is not clear if either of these benefits will result from the future transitions, let alone both.

In other words, our industry could finally meet the point of diminishing returns.

As management consultants love to do, the McKinsey team created a two-by-two matrix to describe the possible outcomes if these benefits are or are not realized.

What is most interesting here is that “Leading-edge segments continue to consolidate and absorb capital expenditure” results from three of the four outcomes. In other words, the rich get richer. The only outcome that does not result in an oligopoly with two to four major players is if Moore’s Law ends.

Some OEMs will have more power in some markets over semis
As the larger consumers of semiconductors gain more power, they will have the option to integrate downstream to design their own chips. This is highlighted below in Exhibit 4 from the paper:

As I implied in the EE Times article and Semico presentation, the key word in the red box is “certain.” Only certain system OEMs in certain industries will take on the added expense to design their own silicon. If huge differentiation results in huge margins, the Googles of the world have the option to make their own chips. Of course, having this option allows them to drive down chip prices even if they choose to buy rather than make.

System house/OEM power is increased by slower semiconductor technology adoption
The “so what” of all this is that we can better understand the trends that are happening today, and perhaps predict a bit of the future by looking at the McKinsey analysis. Given a projection of whether Moore’s Law continues, one can explore the ramifications.

My personal, non-fact-based opinion is that scenario IV above, “Moore’s Law ends,” will not happen. However, it is possible that either performance increases or cost improvements could end with the transitions to EUV and 450mm (scenarios II or III).

What this would mean is we would have even more consolidation in our industry, not just with IDMs and foundries, but also with our major system house customers. These OEM customers would have even more power than they have today to compete directly with their semiconductor suppliers. This would result in a few leading-edge semiconductor vendors that can compete to meet the needs of leading-edge OEMs. The rest of the semi vendors would have to pick a niche where they can innovate and extract unique value, or exit the market.

Bottom line: Increased power by key OEMs over others in the semi value chain is not a cyclical pendulum swing between industry horizontalization and verticalization. The current trend is being driven by our slower rate of new technology adoption, and will only accelerate as it becomes harder and more expensive to introduce new semiconductor manufacturing technologies.