During Computex Taipei 2012, we got an opportunity to talk with Jason Gorss and Subramani Kengeri, Head of the Advanced Technology Architecture at GlobalFoundries to discuss the latest trends from the foundry.

First and foremost, GlobalFoundries expectation of the foundry business exceeding the growth pace of the overall semiconductor industry is coming to life – while the semiconductor industry is growing at a 5% annual rate, foundry business is growing by 10%, i.e. twice as fast. The semiconductor industry is expected to grow from $304 billion in 2011 to $384 billion in 2016. At the same time, leading edge foundry (65nm and below) are expected to double, from $15 to $34 billion. In order to capture as much pieces of the silicon pie, GlobalFoundries is pushing forward with the 20nm process, scheduled to arrive next year. 

How did GlobalFoundries make 20nm a cost-effective technology platform?
The major change for the 20nm process is adoption of Gate Last, which was the key differentiator between Common Platform and Intel, TSMC and other players. Now you can be a TSMC customer and switch to IBM/GlobalFoundries/Samsung without too much engineering drama.

The manufacturing advantage is going back and forth between GlobalFoundries, TSMC and Intel. Intel lead the market with 45nm, then TSMC launched 40nm. Intel came back with 32nm, then TSMC and GlobalFoundries both started shipping 28nm semiconductors. In April, Intel launched the 22nm process (after several months of delays, untypical for the 800 pound manufacturing gorilla), and that lead will be continued onto 2013, when 20nm process from GlobalFoundries will take the lead, before all three foundry groups meet at 14nm.

With the 20nm node, GlobalFoundries worked hard with its partners inside the Common Platform Alliance (IBM and Samsung) to make the node as SoC-friendly as possible, yet keeping the same performance characteristics that enabled very high clocking of the 32nm and 28nm parts.

First transistors at 20nm show interesting comparison between GlobalFoundries and its direct competitors
First tests on a test ARM Cortex-A9 core at 20nm show an interesting comparison between GlobalFoundries and its direct competitors

The 20nm Low-Power-Mobile node directly competes for mobile SoC business from companies such as Qualcomm, Texas Instruments or even NVIDIA Corporation. The interesting player in this game will be Qualcomm, which is scheduled to ship its 28nm chips from GlobalFoundries during the third quarter. Even though neither GlobalFoundries or Qualcomm want or are allowed to talk to us in regards to their mutual cooperation, our well-connected sources are telling us that one of key customers, Taiwanese HTC was both ashamed and angry for having to ship its flagship phone, the One X (which we reviewed here) with an old 40nm Snapdragon S3 processor.

Such problems should be the matter of the past at 20nm. GlobalFoundries claimed significant lead in the key Performance-Power-Cost metric, as well as marginally lower battery life based on transistor design alone.

Shrinking the die from 90nm to 20nm opens up new possibilities in terms of transistor density
Shrinking the die from 90nm to 20nm opens up new possibilities in terms of transistor density

20nm also represents an interesting paramount in the silicon industry. Just like Intel’s 22nm, the 20nm process is being geared as the process that will revolutionize the mobile industry. While we won’t see 20nm designs until very late in 2013 (unless somebody pulls a surprise), leading mobile processor vendors will be able to develop a billion transistor design in a smartphone i.e. tablet friendly package.

Crossing the Lab to Fab Chasm: What after 20 and 14nm nodes?
Crossing the Lab to Fab Chasm: What after 20 and 14nm nodes?

After the 20nm reaches its peak in terms of Gate Last HKMG, and first FinFET (i.e. 3D transistors) roll out the door, we can expect 14nm FinFET process coming from all the major industry players. This will also be the first time when we are going to see collision of worlds: Common Platform, TSMC, SMIC and Intel – all on 14nm, all with FinFET transistors. Intel will still probably use bulk silicon, with Common Platform i.e. IBM/GlobalFoundries/Samsung utilizing fully-depleted Silicon-on-Insulator.

SOITEC, leading provider of SOI wafers made significant progress and from what effectively was a dead technology for the 28nm and 20nm node – brought a lot of development forward all the way to 28nm. While GlobalFoundries said they will not offer standard SOI on the 28nm and 20nm, Fully Depleted SOI is making an entrance in a pretty big way, two years ahead of the schedule.

All in all, the 20nm roadmap shown to us by GlobalFoundries holds ground for now. Time will tell if the market is going to dive in for all the products that 20nm process will make possible.

Watch this space.