The first surprise of the opening keynote for Semicon West was on the slides that were cycling on the screen as the room filled up. Somehow our book Fabless had managed to be in the rotation.
The opening keynote was by Mark Adams, the President of Micron. He was talking about upcoming big changes in the semiconductor environment, although given his background there was an especial interest in memory. Memory has been a weird market. From 2002-2012 nobody made any money. It was basically a commodity DRAM market and the companies in it, of which there were over 40 at the beginning of the era, overbuilt capacity leading to a price war. At the current node and with the worldwide downturn, the number of companies was down to 4 (Toshiba, SKHynix, Micron and Samsung) and they didn’t overbuild capacity. Indeed, a large part of the growth of the semiconductor industry in the last year or two has been simply the firming of prices in the DRAM market.
He sees several big challenges going forward, some of which are unique to memory but mostly not:
- the end markets are getting more diverse. The main market a decade ago was PC, that has already been eclipsed by mobile, and whatever the internet of things turns out to be it will for sure be consist of hundreds of products
- increased complexity of system level solutions, particularly 3D and novel packaging
- decisions about foundry vs captive. Well, for everyone except the memory manufacturers and Intel (and, I suppose, Samsung) that decision has already been made
- the fundamental semiconductor technology is at a crossroads, running into some fundamental limits requiring extensive innovation. In particular, for Micron, the upcoming end of planar in DRAM.
- changing capital allocation models making capital harder to deploy and raising the importance of just how it is deployed (not to mention the cost of fabs means more capital is needed to build one)
- economics don’t scale like they used to, a process node shrink does not generate as much return as before and so shrinking is less important than other innovation
- finally, the customer landscape is evolving, moving (for memory) from just PC and networking towards embedded, server, mobile, client storage and more
A new more cooperative partnership model is required since no one company can do it on its own and since even memory, never mind SoC, is no longer a commodity.
It is obviously more exaggerated for memory, but all semiconductor business face the fragmentation of their market. They will need to partner with their customers closely to ensure they build the right products, but also they will inevitably be building a broader portfolio of products with relatively lower volumes for each one.
And developing the basic semiconductor process will also require partnering to do the development, to develop the equipment needed to manufacture, and to develop the new materials that will be required.
Success will depend on 5 factors: