Review Index:
Feedback

GLOBALFOUNDRIES: 2013 and Beyond

Author: Josh Walrath
Subject: Editorial
Manufacturer: GLOBALFOUNDRIES

Two Technologies to Watch

Two implementations really stand out on GF’s roadmap.  The 14XM FinFET node is a low power product aimed at mobile applications at 14 nm.  It will not be a good choice for large, power hungry products like GPUs, but it will be appropriate for low power SOCs.  GF is being very aggressive in getting to this node as soon as possible, which is supposed to debut in a 2014 timeframe.  With the massive growth of mobile computing, this could be a good money maker for GF and a good line for current SOC producers to utilize.  This is again not a performance oriented line, but it does pave the way for future products that will be.  The combination of low power consumption and 14 nm features should enable a new generation of mobile products which will better and have more features than what we see today.  I fully expect to see ARMv8 based processors utilize this node.

View Full Size

FD-SOI is looking to shake things up.

The second really interesting area is that of Fully Depleted SOI.  Currently GF is one of the top producers of Partially Depleted SOI, as all of AMD’s current APUs and CPUs utilize this wafer type at 32 nm.  FD-SOI takes things a step further in terms of materials, performance, and physical and electrical characteristics.  GF is the primary production partner of ST’s 28 nm FD-SOI process technology.  What we have seen so far with this 28 nm FD-SOI process is pretty impressive.  ST-Micro recently showed off a 3 GHz Cortex A9 based part that does not break the bank when it comes to TDPs.  This tablet based part is able to go to 3 GHz at standard voltages and go into a low speed mode at 1 GHz at very, very low power levels.

The first generation of FD-SOI products we see will be based on 28 nm with planar gates (as compared to FinFETS).  FD-SOI can utilize much the same equipment as bulk silicon, so ease of manufacturing offsets the increased cost of the wafers.  These products will be gate-first so as not to conflict with many current design methodologies and production flows.  Initial indications point to 28 nm FD-SOI being very competitive with Intel’s 22 nm Tri-Gate product.  Die sizes on Intel’s process will of course be smaller, but not as much as some would expect.  There are some features on Intel’s process that are upwards of 26 nm.  Where FD-SOI really shines is not only in power consumption, but also in transistor switching performance.  Intel’s 22 nm Tri-Gate is very power efficient at moderate speeds, but once products exceed a certain speed then power consumption and heat production dramatically increases.  FD-SOI exhibits a more linear curve when increasing clockspeed vs. power/voltage.

Where things start to get very interesting is at the 22/20 nm node.  FD-SOI is able to achieve competitive electrical characteristics with Intel’s 22 nm process while continuing to utilize planar gate designs.  It does not require the 3D FinFET structures to achieve the power saving features that we have seen with Intel’s process.  This is still using the gate-first methodology.  Reports have shown that a gate-last approach will deliver even better performance and characteristics.  For this particular generation of process nodes, it certainly appears as though gate-last 20 nm FD-SOI will be one of the best general and high performance process nodes available to 3rd parties.

View Full Size

Fab 8 with all of its machinery installed and ready to go.

It does not look like GF will go full bore into FD-SOI, and they will offer a 20 nm bulk/HKMG/planar product.  GF will not go FinFET until 14 nm, but they feel their 22/20 nm bulk, planar process will be the right combination of performance and cost.  IBM and Samsung are also pursuing this particular technology in their own Fabs.  This looks in fact to be the primary process at this node, and only client demand will shift production from bulk silicon to FD-SOI.  We must also consider that it does take time to develop a FD-SOI process even though it re-uses IP from bulk processes.  28 nm FD-SOI will not be ready for risk production until 4Q 2013 and volume production in 1H 2014.  While FD-SOI certainly appears to be an outstanding technology, we will not see consumer level products based on these wafers for some time to come.

For now GF is churning out a lot of 32/28 nm products from AMD and Rockchip.  Other partners have announced their intention of utilizing GF’s 28 nm bulk process; these include Qualcomm, STMicro, Adapteva, and RAMBUS.  This is certainly a few more major clients than what GF has had in the past.  In addition to these 28 nm partners, they have a whole range of clients using the older 45 nm process and above for a variety of chips that include high voltage ICs, automotive IC products, and mixed signal products (prior Chartered partners).

Lastly GF has shown the ability to create TSVs on their 20 nm bulk process.  TSVs are trans silicon vias, which can connect dies stacked vertically.  This technology is still in its infancy and has not been applied to any shipping product that I know of.  The technical challenges are great, but the ability to stack chips in such a way should improve performance dramatically in certain situations.

 

Wrapping it Up

This seems to be an overly positive article about GLOBALFOUNDRIES, but the truth of the matter is that GF has not yet executed on their sub 28 nm plans.  They have only relatively recently been able to ramp production of bulk 28 nm HKMG products for partners, and they are still ramping production at Fab 8.  They are still running full bore to keep production of AMD’s upcoming Steamroller based APU in house, as compared to losing that production to TSMC.

View Full Size

The three current 28 nm products from GLOBALFOUNDRIES.

FD-SOI is a bit of a risk, but as more and more research comes out it seems to be the wafer material of the future.  Even Intel is considering FD-SOI for their sub-14 nm process nodes, and some believe that contracts will be signed very soon to enable production at Intel for that technology.  Still, selling potential clients on such a technology is a hurdle.

GLOBALFOUNDRIES will have to show some progress very soon, otherwise ATIC is not going to be pleased.  We have seen new officers appointed from ATIC to help run the company, and eventually ATIC will want GF to perform up to expectations and be a major player in the pay for play foundry market.  The first several years were pretty rough, but it seems as though the ship has been righted and we are headed to some interesting and compelling solutions in foundry technology.

April 17, 2013 | 09:21 AM - Posted by Anonymous (not verified)

Don't all the big CPU fabs use photolithography equipment from ASML and others, including Intel, and why is GLOBALFOUNDRIES having issues! All of the big CPU foundrys
have access to FINFET, and Companys like Intel, samsung, AMD, and Nvidia are all using ASML's and others suppliers equipment in their chip fabs!

April 17, 2013 | 12:53 PM - Posted by Josh Walrath

Lithography is only one part of the whole process.  Yes, it is very expensive, and only a couple of companies do the optics for it.  Foundries buy the equipment to do this, but all other aspects of chipmaking take a lot of time and effort (not to mention money) to get running correctly.  First we have the mix of materials that can do compressive and expansive strain, then we have other materials that need to be deposited.  Then we move onto the metal layer deposition, etching, washing, and repeating for as many layers as is required by these complex designs.  It is amazingly complex.  Getting an individual process to work, with the materials and machines available to them, is a very long process.

April 17, 2013 | 01:48 PM - Posted by Anonymous (not verified)

Yes but most of the equipment for chip fabs around the world, comes from a relatively few specialist companys that are the suppliers for Intel. Gloablfoundries, Samsung, etc!
ASML, being the largest of these specialist campanys! Is GlobalFoundries AMD's only chip foundry source, and if it is GlobalFoundries needs to get these new process lines going as quickly as possable, or AMD may need to get a second source supplier! Eventually the process node shrinks are going to hit the physical process node barrier for circut size! The costs of going below 5nm may be byond a single company's ability. Now that leaves Intel's, AMD's, and Nivida's processes being at the same level in not to many years from now! If AMD can hold on, even if it has only a few x86 products currently, it can begin to catch up to Intel in the process node via Gloablfoundries, beacuse I do not see Intel surpassing AMD or Nvidia in GPU design, by the time the rest of the industry catchs up to Intel's process node lead!

January 25, 2014 | 05:31 AM - Posted by Anonymous (not verified)

You need more than litho tools to make chips. You also need the materials science know-how and production recipes to build the transistors, etc, and there are many different approaches to overcoming the problems that arise as you scale down.

Using the same tools as someone else doesn't magically confer their knowledge and skill set to you. Give a man on the street the same paint brush Van Gogh used and it won't make him an expert painter.

April 17, 2013 | 09:25 AM - Posted by Anonymous (not verified)

Correction AMD(now using GLOBALFOUNDRIES) as AMD sold their chip fabs!

April 17, 2013 | 11:42 AM - Posted by Anonymous (not verified)

KO-KO-KO KUDAH-TAH-TAH

April 17, 2013 | 01:11 PM - Posted by qubiq (not verified)

Plans look great, let's just wait for execution...

May 5, 2013 | 07:15 PM - Posted by Anonymous (not verified)

Fab 8, 450mm tools going to be installed in phase 2 or the "expansion" of the existing fab. Thank you Albany Nanocollege!

Post new comment

The content of this field is kept private and will not be shown publicly.
  • Lines and paragraphs break automatically.
  • Allowed HTML tags: <a> <em> <strong> <cite> <code> <ul> <ol> <li> <dl> <dt> <dd> <blockquote><p><br>
  • Web page addresses and e-mail addresses turn into links automatically.

More information about formatting options

By submitting this form, you accept the Mollom privacy policy.