FinFET Ramp: Changing Market Dynamics?

Intel, TSMC and Samsung are involved in an interesting strategy game, with no obvious winner because there are too many variables.

popularity

Rolling out a new semiconductor technology always has its share of challenges, but it seems like the 14nm finFET process node is starting off with more than its share of delays and speculation.

This week Intel revealed some of the details for its new microarchitecture, Broadwell, and their first product, the Intel Core M processor, to be manufactured using their second-generation finFET, 14nm technology.

Plagued by rumors of yield issues and a slow computing market in 2013, Intel delayed the release of their newest 14nm product line by almost one year.
Back on July 16, during the TSMC Q2 2014 financial conference call, the company reported that their 16nm finFET process would not begin ramping until 2H 2015. That is a delay of approximately six months from the original Q4 2014 ramp target.

There have been some technology issues, but Semico believes these delays are also market-driven. In the past, new products could be released for the early adopters willing to pay a premium just to get in on the higher performance. Huge volumes were not required from day one. Today, the ramp to high volume occurs much faster and the products require high efficiency and low cost to support huge consumer driven mobile markets.

Does this mean a shift in the market dynamics? Are the challenges of these complex technologies having a negative effect on the industry and its players? Because of their delayed introduction, TSMC doesn’t expect to be the number one foundry provider of 16nm/14nm FinFET volume in 2015. Intel’s computing sales took a dive in 2013. Could a 14nm processor have helped their sales last year?

Semico believes Intel’s timing may work in their favor. Computing sales have begun to show signs of recovery. The 2-in-1 form factor is gaining acceptance. The rollout of 14nm with a solid Core M processor for tablets could be Intel’s pathway into the mobile market.

TSMC invested in 20nm for volume production in 2014 and will have 300K 20nm capacity by the end of this year. TSMC experienced rough introductions at both 40nm and 28nm. This strategy gives TSMC more time to fine-tune its first-generation finFET process. A successful large volume ramp would be a feather in TSMC’s cap. As the company executives pointed out, they may not be the No. 1 provider of 16nm/14nm FinFET in 2015, but they expect to retake their No. 1 position in 2016.

But there is another player in this equation, too. Samsung is still anticipated to rollout their 14nm product on schedule by the end of this year. The company demonstrated a 14nm device in June at DAC 2014. Samsung is also in a unique position. Even if they do run into yield issues or other technical difficulties on their first 14nm product, Samsung has more options on how to deal with the product rollout because they control their own end-market product strategy. If they can roll out a 14nm, low-cost application processor in high volume and be the first to bring the next ‘must-have’ device to market, Samsung will have a lot to boast about. But the more likely scenario with a first-time product as complex as a 14nm finFET device would be that there’s at least one glitch—or more likely a few. Volumes could be constrained.

Would they be willing to implement a strategy of limited release for the early adopters who are willing to pay a higher price? Are consumers willing to stand in line for a high-performance Samsung Galaxy product?

These market strategies and moves are more entertaining than a good game of chess. For more information on the finFET players, the foundry market implications, manufacturing capacities and wafer demand forecast, contact Rick Vogelei at [email protected] for details on Semico’s new finFET report.



  • Byungchun Yang

    Joanne,

    I am sorry but I disagree with your comment on the reason of Intel’s delayed 14nm node Broadwell release. I believe that there is always a market even if a chip costs $5000, for example, if the 14nm chip shows any better performance such as a faster speed (e.g., 3.5GHz) or less leakage for a longer laptop use, or both. Please don’t get it wrong. There are always enough rich people who would pay higher money for a better product, and even $5K/chip is not a burden to them at all.

    The real reason is thus that the 14nm chip is not giving any benefit to the customers; compared to previous node, speed is slower for a higher via resistance, yield is lower although more chips can be obtained for a smaller size theoretically, and leakage is higher due to more closely populated lines and other features. I guess BEOL portion of their chips got damaged more since smaller Cu vias and Cu lines mean nothing but a disaster! Look patent US8,207.060 for solution, if you want. The current BKM damascene process has lived up to its life time, and new approach is needed although some more investment is needed.