Will 10nm Be The Last Big Node?

Foundries have some very tough choices ahead, and not enough information to make them.

popularity

There is a great deal of attention being paid to established nodes these days and everything up to and including 10nm. What comes after that remains a mystery.

Intel and a handful of others will keep pushing to the next nodes, of course. Still, where the commercial foundries—including Intel—place their next big bets is a matter of ongoing debate. There is no doubt that 7nm and 5nm will be possible. Yet as markets for semiconductors become much more rationalized on the basis of power, performance and cost, the number of chipmakers that will be able to justify the cost of moving forward every couple of years has been shrinking. That trend will continue, but with unpredictable results

In the past, this became the basis for node skipping, where companies would develop chips at one node and skip the next node. For example, they could jump from 65nm to 28nm, leaving out 40/45nm. Node skipping is much harder these days, because multi-patterning and finFETs require new design techniques, new tools, and a whole bunch of new manufacturing rules.

But that’s only a piece of the puzzle. The entire semiconductor market has become much more complex over the past several process nodes. As the mobile market has consolidated, the number of companies vying for a socket at advanced nodes has shrunk. This is hardly bad news for the chipmakers, of course, in large part because of the uptick in demand for complex semiconductors at established nodes. This shift is being driven by the automotive, medical, industrial and consumer sectors, with the Internet of Things as a common communications fabric running between them. All of those markets utilize established process nodes, and there is a big effort underway to tweak those process technologies to reduce current leakage.

At the same time, though, the number of players developing applications processors for the mobile space has consolidated greatly, and two of them—Samsung and Intel—have their own fabs/foundries. That leaves ARM-based chips for Android devices—of which Samsung is the largest—and Apple, in the extreme high-volume mobile applications processor market.

The other major drivers for advanced nodes are server and networking processors, along with FPGAs, which require ever-greater density. But the speed at which they move forward, or whether they opt for other architectures such as 2.5D and 3D, or different substrates and processes, raises some new questions about where foundries and equipment makers will focus their attention and investments. Rather than process technology, it may be materials and architectures that provide the most bang for the buck, and that could have some massive ramifications for the entire supply chain.

All of this culminates after 10nm, depending upon lithography, materials, markets, the cost of equipping fabs, expected utilization, and even such developments as 450mm wafers. But the more questions that surface, and the more variables, the bigger the gamble for everyone—to the tune of billions of dollars for some companies, tens of billions of dollars for others.