NOVEMBER 2012: It is said that with the exception of Dresden (hub for companies like GlobalFoundries and now Infineon with its recent acquisition of Qimonda’s assets), Europe might find itself open to attack on 450mm fabs, given its predominantly 200mm/150mm manufacturing base.
Europe’s chip companies will be subjected to a constantly ongoing closure and restructuring effort, damaging the employees’ morale and affecting the costs and productivity. Finding and keeping good operations personnel will be very difficult indeed, given this business strategy is an operations career’s dead-end.
Future Horizons has long stated that the ‘fab-smart’ strategy remains the only true solution. This means continuing to build in-house state-of-the-art fabs while outsourcing a modest amount (5-15 per cent maximum) to foundries in order to smooth the supply and demand peaks and build external fab demand high enough to justify equipping the next module of in-house expansion.
In this way, any expansion in capacity enters production ‘fully loaded’ from the beginning whilst simultaneously improving response time to near-term demand fluctuation. The foundries do not like this option of course, but this is the only real competitive reality for integrated device manufacturers.
There is a parallel opportunity for an European-based 450mm wafer fab focused on more mature technology node processing.
In 450mm process, there have hardly been advances in Europe as far as silicon readiness (scratches eliminated and particle count dramatically reduced) is concerned, according to Malcolm Penn, chairman and CEO, Future Horizons. “They have not made anything beyond a few sample wafers yet.”
As for the particle, edge and thin-film measurements capabilities required for 450mm fabs, he adds that this is just a part of the whole transition process. All has to be ready in time for the first module.
Where will the fabs be?
So when are the first 450mm fabs likely, and where? Intel has announced that its first 450mm pilot plant will be at Hillsboro in Oregon in the new D1X fab. It will be used to debug the production process before moving it to other high-volume sites.
The global semiconductor industry is going to make a major transition from producing 300mm wafers to 450mm wafers. This move to 450mm wafers also means that companies can set up 450mm fabs (fabrication units) that will lead to reductions in die cost, and later, production cost.
A wafer is a thin slice of a semiconductor material, such as a silicon crystal, used in the fabrication of integrated circuits (ICs) and other micro devices. It is believed that 450mm wafers will cost less per unit output than 300mm wafers. It is also said that 450mm production will make it easier to stay in the game because it is cheaper. The 450mm fab and wafer could become a reality by 2018.
Currently, this fab is in a very early construction stage, planned for a 300mm line initially and a 450mm pilot line sometime later. The second plant will be in Arizona. The location of the third plant is yet to be decided. It could be in the US, Israel or Ireland.
Only the University of Albany’s College of Nanoscale Science and Engineering (CNSE), IMEC, Intel (Oregon), Samsung and TSMC are expected to be able to research and develop 450mm wafer processing during this decade.
Intel is the prime driver in the move to 450mm wafer processing. It has announced its intent to run in the first line on a proven process/IC design combination at the 15nm node before moving to 11nm once everything is proven capable of operating at this node properly.
It will be interesting to see how TSMC allocates wafers on its first 450mm fab and how much it charges for them. TSMC is perhaps the most complicated 450mm business case, given that its first 450mm product will have to be a logic IC. The key decision will therefore be whether the first 450mm process will be high-speed or low-power.
Samsung’s first 450mm fab is expected to be a memory fab. It’s not certain whether this would be for DRAM or Flash.
Industry observers believe that the first 450mm wafers will be produced at node n-1, but they will be flipped to node ‘n’ as soon as the 450mm platform bugs are ironed out with comparable or better wafer fab yields and performance than the established past-practice.
This might be sooner rather than later for Intel, given its high-performance fab business model, and later rather than sooner for TSMC, where stable yields and low cost are the key driving factors. A few fabless semiconductor firms, though, are expected to be prepared to share the potential die cost risk with TSMC in order to benefit from the early 450mm experience.
Samsung’s roll-out is a bit hard to predict, as it will be strongly influenced by whether the first parts are DRAM, NAND Flash or even SoC logic.
TSMC has announced that its first 450mm pilot line was planned for the Fab 12 Phase 6, starting with 20nm technology. The timing of this pilot line is said to be 2013-14, with 450mm production line planned for 2015-16. If met, this timescale would be one to two years more aggressive than Intel’s. It is known that TSMC’s 20nm process is well advanced with ARM building a Cortex A-15 chip to prove both the TSMC process and ARM’s cell library at this node.
Samsung, on the other hand, has made no public announcements on either the locations or timescales for its 450mm plans. Given the cost-effectiveness of its 300mm lines, it may wait for the others to make the initial 450mm moves first.
Role of G450C
According to Gartner, now that the Global 450 Consortium (G450C) has been established as a proving ground for 450mm equipment, development is moving forward rapidly.
Moving to 450mm process is an economic decision. For 450mm process to succeed, G450C must show that it can produce semiconductor dies less expensively than the 300mm process currently in production.
While the G450C pilot line will be ready to accept equipment in 2013, it is highly unlikely that 450mm will enter production before 2018.
The money needed to R&D the 450mm production lines will place a significant strain on semiconductor equipment manufacturers, as the spending needed on 450mm alone is greater than the current R&D spending by the equipment manufacturers.
Gartner has made the following recommendations for 450mm fabs:
1. Semiconductor equipment makers must be engaged in 450mm development, either by participating in the G450C or Interuniversity Microelectronics Centre (IMEC), or by working closely with one of the G450C semiconductor manufacturers.
2. Semiconductor manufacturers will need to communicate openly with the equipment makers to ensure that the equipment is developed correctly and that the equipment makers are enabled to do what they do best.
3. The G450C will need to engage the lithography makers to ensure that lithography will be available when the semiconductor manufacturers are ready to move to production.
Reducing manufacturing costs
In 2007, Gartner had estimated the R&D costs of equipment for the 450mm transition to be at least $16 billion. More recent equipment industry studies suggest that the cost will be even higher, between $25 billion and $40 billion.
For the 450mm programme to be successful without financially crippling the equipment industry, the semiconductor manufacturers will need to assist to see adoption of 450 mm by 2018.
IMEC and International Sematech Manufacturing Initiative (ISMI) have well-established programmes focused on the challenges posed by manufacturing with 450mm wafers. One of the biggest hurdles in the transition to 450mm process is lithography. Another major barrier facing 450mm manufacturing is whether both 193 immersion (193i) and EUV lithography tools will be able to match their 300mm counterparts in terms of wafer throughput. A bigger issue is the source power needed to scale for 450mm manufacturing. The primary motivation behind the move to 450mm wafers is to reduce manufacturing costs.
When is the transition likely?
The transition to 450mm manufacturing will happen by 2020. However, the expected manufacturing cost savings will be only 10 per cent, extending the return on investments in R&D by ten years or more.
Semiconductor companies driving the move to 450mm—Intel, Samsung, TSMC, GlobalFoundries and IBM—are spearheading the G450C for 450mm initial development at Albany Nanotech.
In addition, the IMEC research consortium in Leuven, Belgium, has set up a lab for the development of 450mm equipment and processes. Equipment manufacturers seeking alternatives to the G450C—perhaps looking to have multiple 450mm projects—will be able to work with IMEC to scale their 450mm equipment.
Since the G450C has space for about 50 tools at its pilot line, not all of the equipment makers will be able to show off their wares at Albany. The IMEC facility provides them another alternative for 450mm development.
Semiconductor manufacturers will most likely have 450mm development labs working in parallel with the G450C and IMEC to help accelerate the move to 450mm wafers.
Best case roll-out
The 450mm technology roadmap will be driven by the current 300mm advanced technologies roadmap, despite the fact that its early production life will start one or two generations in arrears.
Based on this model, the best-case roll-out timing is early test tools in 2012, demo tools in 2012-13 and first-generation production tools available in 2015. This would place the 450mm transition node at 11nm, with initial production at 16nm. The technology wall of death for the remaining 300mm-based producers would start to bite home in the 2020 time-frame.
The author is an executive editor at EFY