Changing Economics In Chip Manufacturing

Swapping architectures will do more than just improve performance. It will alter the power structure in the semiconductor industry.


The foundry and equipment businesses are poised for significant changes that could affect the balance of power far beyond just the semiconductor manufacturing sector.

It’s no secret that the number of companies developing new chips at 7nm is shrinking. There will be even fewer at 5nm. The business case for moving forward is that density must provide a competitive edge. But that density improvement will require an investment of hundreds of millions of dollars per design, which means that end markets either need to be completely price insensitive or volumes will have to be in the hundreds of millions of units.

Market opportunities can change, of course. But as of now, the only markets that appear to warrant those kind of investments are mobile handsets, FPGAs and server processors. The fundamental problem here is that it takes a global ecosystem to develop each process. IP vendors already are balking at the cost of developing multiple iterations of IP, beginning at version 0.01, for each of the foundries playing in this space. And while some manufacturing equipment can be stretched out several nodes, equipment makers are looking at fewer customers at these nodes for equipment that is likewise increasingly expensive to develop.

This is the same scenario that pushed 450mm wafers to the back burner, where even the pilot light seems to have gone out. R&D investments may provide better returns in such areas as advanced packaging, new materials, photonics, and manufacturing equipment for neuromorphic and quantum architectures.

This is not necessarily bad news for the equipment and foundry sectors, although the picture isn’t quite as rosy for the handful of companies that are banking on more feature shrinks. In fact, all of this churn gives the manufacturing sector far more clout in dealing with big systems vendors. There are enough business prospects at 28nm and above in new and existing markets, and enough possibilities for improving performance and power with new architectures, to give foundries and equipment makers some breathing room for the first time in many years.

Rather than just vying for a piece of the flattening PC and mobility markets, new opportunities are cropping up around the globe in markets that traditionally have been insignificant consumers of semiconductors—automotive, augmented/virtual reality, drones and aerospace, industrial equipment, and medical devices. And while the individual orders may be smaller, the total available market will be significantly larger.

In these kinds of semi-customized engagements, foundries have quite a bit more leeway on pricing. They are not betting the bank on a $1 billion process and struggling just to keep up with what has turned into an arms race with uncertain payback. At the same time, equipment makers can sell lots more modernized, flexible equipment that can facilitate a shift from one manufacturing job to the next.

All of this casts a shadow—at least for now—on the big systems vendors, which have been taking a lopsided share of the profits. They will be competing in a totally different supply-and-demand scenario, where playing off one foundry or OSAT against the next doesn’t necessarily work anymore. The economics of the semiconductor industry are shifting as the world becomes increasingly connected, and the biggest deals may account for less than a lot of smaller ones.

  • realjjj

    The process might matter less where you don’t need computing power. IoT is the only major area where cost might favor an older process.
    In glasses you need very high perf and insane efficiency. In robots (and that includes cars and drones) you need very high perf and better efficiency than in server or PC. Glasses would also harm TVs, tablets, PCs and there would be a shift to higher perf and much higher efficiency.

    What is needed is a breakthrough on the cost side, scaling on the horizontal being less relevant. A true monolithic 3D that would enable costs to scale much better than Moore. The focus on high perf and horizontal scaling, starting with the choice of the FET , is all wrong. Those times are gone. The focus should be on costs, accelerators (this includes neuromophic) and parallelism.
    The growth is there ,if the industry grabs that growth in 10 years or 30 years, depends on the roadmap. It’s not about servers, not about CPU , not about the horizontal anymore.All that is too little and not fit for the purpose anymore. Even in server accelerators are starting to have a major impact but area is too important to allow for much higher efficiency.