Taiwan Semiconductor Finally Goes FinFET, But It's Not All Roses
At IEDM 2013, a premier conference at which the latest semiconductor technologies are often revealed, Taiwan Semiconductor finally detailed its next-generation 16-nanometer FinFET semiconductor manufacturing technology. This is a great step for the company as well as its numerous fabless customers, including Qualcomm, Broadcom, and NVIDIA.
While it would seem as though TSMC and its customers would be well-positioned to take on products based on Intel's 14-nanometer process, it's important to understand that not all 14-nanometer and 16-nanometer processes are created equally.
TSMC's 16-nanometer is really 20-nanometer with FinFETs
When it comes to semiconductor manufacturing, there are two major "parts" of the manufacturing process: the "front end of line, or FEOL, and the "back end of line," or BEOL. The FEOL refers to the actual devices, like transistors, that are built onto the semiconductor. The BEOL refers to the metal interconnect layers which basically serve to "wire up" all of these individual components. While the FEOL is incredibly important for the transistor-level performance of the device, the size of the metal layers are a critical component of density.
TSMC shrunk its BEOL from the 28 nanometer node to 64nm (metal layers are much larger than the transistors) at the 20 nanometer node (to get roughly twice the density), the company isn't shrinking its BEOL at the 16 nanometer node. This implies that there will be negligible density improvements in going from the 20-nanometer node to the 16-nanometer node. With that in mind, the move from traditional planar transistors to a new structure known as FinFETs buys TSMC significant performance and power improvements.
What kinds of performance improvements will TSMC customers get?
According to TSMC's presentation at IEDM, chip designers will see about a 35% improvement in performance at the same power consumption, or up to a 55% reduction in power consumption at the same performance level. Keep in mind that this refers to transistor-levelimprovements, so they may not translate directly to chip-level improvements -- a chip is made up of billions of transistors.
At any rate, while this seems impressive, this does suggest that as far as performance goes, TSMC's 16-nanometer node will have roughly the same performance as Intel's 22-nanometer FinFET node. Want proof? When Intel announced its 32-nanometer node, and when TSMC announced its 28-nanometer node some years ago, Intel's transistors had a distinct performance advantage over TSMC's, as measured by NMOS and PMOS drive currents at fixed leakage current.
When Intel moved to FinFETs at the 22-nanometer node, it cited 37% speed gains at the same power, or 50% power reduction at the same speed. That's just about what TSMC is reporting that it's seeing here.
As far as density goes, TSMC will be meaningfully ahead of Intel's 22-nanometer process, but according to materials presented at Intel's investor meeting, it will be behind Intel's 14-nanometer process by about 35%. Further, as Intel claims up to 40% higher performance at the same power, or 67% less power at the same performance as its 22-nanometer node, it should have a meaningful transistor-level performance advantage over TSMC's process.
What does this mean?
This means that as far as transistor-level performance goes, TSMC will have a tough time competing with Intel. Can TSMC's customers still out-execute Intel on the design side? It's certainly possible if Intel still hasn't quite gotten its designs sorted out. But given that Intel's designs have been rapidly improving, it's going to be pretty tough for the fabless players to compete if Intel can deliver a knockout set of high-end mobile products.
With that in mind, Intel's share gains -- if it delivers as promised -- won't happen overnight. TSMC serves a rather broad swath of clients in a wide range of end markets. Further, with the rumors that Apple has chosen TSMC for its 20-nanometer parts, any share that TSMC's other mobile chip clients lose will probably be more than offset by some very attractive business from Apple.
To further drive this point home, most of the end markets that require TSMC's leading-edge silicon are growing, so even if some of TSMC's clients lose share to Intel in tablet/smartphone chips over the next few years, it is still poised to see nice unit growth. Also, Intel's highest-volume tablet/phone products for late 2014/most of 2015 will actually be built at TSMC.
Foolish bottom line
TSMC's announcement that its 16-nanometer process has gone into "risk (limited quantity) production" is encouraging, although there is still a long road to "high-volume production" -- TSMC claims it'll be in late 2014/early 2015. The industry dynamics should be quite exciting to watch -- and invest in -- over the next several years.
More compelling ideas from the Motley Fool
As every savvy investor knows, Warren Buffett didn't make billions by betting on half-baked stocks. He isolated his best few ideas, bet big, and rode them to riches, hardly ever selling. You deserve the same. That's why our CEO, legendary investor Tom Gardner, has permitted us to reveal The Motley Fool's 3 Stocks to Own Forever. These picks are free today! Just click here now to uncover the three companies we love.
The article Taiwan Semiconductor Finally Goes FinFET, But It's Not All Roses originally appeared on Fool.com.
Ashraf Eassa owns shares of Intel. The Motley Fool recommends Apple and Intel. The Motley Fool owns shares of Apple and Intel. Try any of our Foolish newsletter services free for 30 days. We Fools may not all hold the same opinions, but we all believe that considering a diverse range of insights makes us better investors. The Motley Fool has a disclosure policy.
Copyright © 1995 - 2013 The Motley Fool, LLC. All rights reserved. The Motley Fool has a disclosure policy.