GlobalFoundries has announced that its Fab 1 in Dresden, Germany has shipped a quarter million wafers based on 32nm high-K metal gate (HKMG) technology over the last year.
The chipmaker pointed out that it had exceeded its customer’s product shipment requirements and the 32nm yield challenges are well behind it.
GloFo wheeled in new AMD chief exec Rory Read to comment. He said that in one quarter, it was able to double yields on 32nm. It means that the company could finish 2011 having exceeded its 32nm product shipment requirements.
“Based on this successful ramp of 32nm HKMG, we are committed to moving ahead on 28nm with GlobalFoundries,” Read said.
On a unit basis, cumulative 32nm shipments for the first five quarters of wafer production are more than double that achieved during the same period of the 45nm technology ramp, he added.
This showed that the 32nm ramp has significantly outpaced the 45nm ramp, despite the integration of a number of new and complex elements in both design and process technologies.
Now that its 32nm HKMG supply is unconstrained, AMD can now increase output of its latest chips, including new Opteron processors for servers, FX CPUs for desktops, A-series “Llano” and “Trinity” APUs. Of course it is still miffed at the GloFo delays and is rumoured to be looking elsewhere for new suppliers.
According to anonymous murmurs in the chip industry, the situation for GlobalFoundries continues to look grim with rumours of discontent among disparate staff, with poor leadership affecting chip production.
GloFo has been having big trouble at 32nm and 28nm production with both using problematic gate first processing. This is being compounded by a split in production between two processes, an old AMD one and an IBM one, which has left a can of worms open on the table.
Trying to debug two processes is causing serious headaches, and a lack of cooperation between Dresden and the US bases is throwing the company even deeper into the mire.
It looks like it could be a few years before AMD’s production arm spinoff is truly counted among the major players in the industry.
A sore lack of cohesion is being blamed on a lack of quality leadership which could get the firm powering forward.
With some quality chips at its disposal this is likely to affect AMD too. Bulldozer is a good chip with plenty of raw performance, while Fusion should be dominating at low power applications, but production problems have ruined competitiveness.
Rumours persist that AMD will totally walk from GlobalFoundries.
Infineon has managed to put power semiconductors on a 300 millimeter wafer at the Villach, Austria site.
The German company had been trying to put out a pilot line for the 300mm wafers since October 2010, and it took a team of 50 engineers and physicists to make it happen. Infineon claims the achievement helps put it ahead of the pack for power semiconductors in energy efficiency, which is the talk of the town.
Infineon says Dresden will be the production site for its Power 300, and in the years up to 2014 it plans to invest roughly €250 million, while creating about 250 jobs.
In a statement, research director Dr Reinhard Ploss said that the development “marks a quantum leap in production technology”.
Infineon’s 300mm wafer chips, it says, run as smoothly as its power semiconductors on the 200mm wafers.
It claims the proof is in successful application tests running Metal Oxide Semiconductor Field-Effect Transistors, or MOSFETs, for high voltage applications.
Foundry contender GlobalFoundries said that it is now ready to install 300 millimetre wafer equipment in its Dresden and New York cleanrooms.
The company – spun off from AMD and financed by big dollars from Abu Dhabi – said that Fab 1 in Dresden will now be able to pump out 80,000 wafers a month when it’s at its peak with 100,000 square feet of clean room, space.
Fab 8, the factory it is building in New York State, has already admin staff working there, and GloFo started building another admin building next week.
The company claims that Fab 8 will be the most technologically advanced wafer fab in the world when it’s finished, with total clean room space of 300,000 square feet and the ability to churn out 60,000 wafers a month when it’s completed in early 2013.
It will be able to create chips with line widths of 28 nanometres and below, GlobalFoundries said.
A senior vice president of Global Foundries said that expanding its existing and new fabs using 12-inch technology will allow it to compete successfully against number one foundry TSMC.
Its new fab in New York State, Fab 7 in Singapore and Fab 1 in Dresden will all ramp up processing of wafers, said Mojy Chian, from GloFo. Singapore will pump out 50,000 wafers and a month and Dresden 60,000 wafers a month. He said: “We’ll announce further capacity expansion in the next few weeks. The numbers compared to TSMC will be very comparable.”
Chian said that while it had no immediate plans to join the 450 millimetre wafer party, he believed that creating extra capacity on 12-inch wafers would hold it in good stead for some time.
Global Foundries, spun off from AMD, borrowed its foundry model when it merged with Singapore firm Chartered, he said. Because it is a global company, he said, Abu Dhabi could well be a candidate for a future fab. Sources told TechEye that AMD will lessen its share in Global Foundries as the year wears on.
Chian described two chip designs from customers that he said illustrated the kind of collaboration that made GloFo different from other foundries. Those are the AMD Fusion, already in production, which uses 32 nanometre SOI process technology, high k metal gate, four CPU cores and 11 copper metal layers. AMD codenames that the Llano – it’s already in production said Chian.
The other chip he described was the ARM Cortex-A9. “The level of collaboration between a [chip] designer and a foundry is unprecedented,” he said.
He claimed that close collaboration between designer and foundry was a sine qua non. R&D for 22-12 nanometre chips costs around $1,300 million and a fab between $4.5 billion to $6.5 billion. The 22/20/17 nanometre fab in New York State will cost around $5 billion, he said.
Plastic Logic said that it’s likely a colour version of the QUE ProReader will be ready for action towards the end of next year.
And it also confirmed that the black and white version, which it showed off at this year’s CES, will be available to buy in June – the 3G version will cost around the $700 mark.
Plastic Logic – a spin off from the Cavendish Labs in Cambridge, has a flexible display which consumes very little power. 1,000 content providers have already signed up to collaborate on the ProReader including the Wall Street Journal and several scientific journals.
The device, showed off by the company here at Dresden, weighs less than 500 grams, is shatterproof, thin, uses touch navigation and a wi-fi connection.
The company said that earlier delays to the black and white version were because it wanted to ensure process stability and had to change a few of the process steps.
The problems with a colour version are that Plastic Logic wants to keep the flexibility of the product and is already cooperating with other unnamed companies to make colour filters on plastic foil and merge flexible colour filters with its own flexible display.
The company said that it had a number of other applications in mind for its technology, including advertising, and that as it ramped volume products could become significantly cheaper because the transistor used in the process is relatively simple and avoided the complexities of other designs.
The chief technology officer of TSMC, Jack Sun, outlined the plans the foundry has for moving to 450 millimetre wafers. There appear to be some problems.
One of the problems is the sheer cost required to create a 15 nanometre process on a 450 millimetre and it’s not pretty. The estimate is that such a fab will cost maybe as much as $20 billion and it appears the only companies to have agreed on such a plan are Intel, Samsung and TSMC.
But if they’ve agreed on the plan, they don’t seem to have agreed on any kind of date. Sun was asked to put a specific date on such a move, and he wouldn’t reveal that to the audience here in Dresden.
He said: “It’s an industry infrastructure issue. We all have to contribute to this effort.”
No-one else seems to have jumped on the 450 nanometre troika and it’s clear that the costs involved for other semiconductor players are immense. A delegate at IEF 2010 asked him about this. Sun said: “There will be financial incentives in the future and people will find ways to invest.”
Of course, it isn’t just the effort to move to the larger wafer that’s the problem. The technology of producing 15 nanometre dies faces many challenges.
A source at another pure play foundry told TechEye that his company believed that there was still a lot of life left in the 300 millimetre wafer size. Plus it’s expensive.
TSMC’s chairman Morris Chang expects things to get bleak for foundries and chip makers this year. According to China Economic News, Chang sees business going downhill in the final two quarters of 2010 and predicts one will have to drink tee, sack workers and wait until demand picks up again next year.
Furthermore, he believes the global semiconductor industry will grow four to five percent from 2011 to 2014, in contrast to an 18 percent growth rate this year. He was upbeat about his own line of business, stating foundries will grow 29 percent in 2010. He also added that only two to three semiconductor companies would be able to afford moving to the 15-nm process when the tough gets going. Chang communicated his prophesies at the Global Semiconductor Alliance Memory Conference, which is currently underway in Taiwan.
Market researcher Gartner meanwhile says demand for 300mm wafers is going to rise more than 30% this year due to shelves needing to be filled again. Apparently 3 million 300mm wafers will be used in the last quarter of 2010. Gartner however says fabs and chipmakers shouldn’t sink all their money into making more wafers, as capacity is growing fast than demand.
GlobalFoundries, which spun itself off from AMD the struggling, is going to invest up to $2.7 billion this year to boost 300mm wafer capacity. When all is said and done, Fab 1 in Dresden will be churning out 60.000 wafers per month, Fab 7 in Singapore 50.000 and the new fab currently being built in New York will produce 40.000 a month. That would put GloFo in second place by wafer capacity, directly behind TSMC which is aiming to hit the 200.000 wafers/month benchmark.
GlobalFoundries said it will invest up to $2.7 billion during 2010 to ramp up capacity at its facilites worldwide. The former better half of AMD declined to give a breakdown of capital expenditure per individual fab, however the company did say the major part will be invested in Fab 1. Most of the money will be used to for “the ramp of 32/28nm HKMG technology”.
Fab 1 is situated in Dresden, chip capital of the federal state known in Germany as “Silicon Saxony”, and is planned to achieve an output level of 60.000 wafers per month when fully built out.
Fab 7, situated in Singapore, will have 50.000 square feet more cleanroom space and churn out approximately 50.000 wafers a month, whilst focusing on mainstream technologies ranging from 130nm to 40nm. Fab 8, currently being built in New York’s Saratoga County, will produce more than 40.000 wafers per month when things are all nicely running.
Rival TSMC is currently building Phase 4 and Phase 5, which are two new buildings located at the foundry’s Fab 12. TSMC wants to bring its capacity to over 200.000 wafers per month by the end of this year, whereas smaller competitor UMC aspires to make 100.000 per month.
GlobalFoundries and ARM are using Mobile World Congress (MWC) in Barcelona as a platform to announce new System-on-Chip (SoC) platform technology to power next gen wireless products boasting 40 percent increases in computing performance and 30 per cent improvements in power efficiency.
Word on the Catalonian street is that the new chip manufacturing platform will also sport 100 per cent increase in standby battery life – which translates to about 10 to 15 hours for the average smartphone.
The British Fish n’ Chippy will be working with Glofo on two variants of the new platform including 28nm super low power (SLP) for mobile and consumer applications and 28nm high performance (HP) for applications that need “maximum performance.”
The platform will of course be based on ARM’s very own Cortex- A9 processor, along with a healthy dose of the firm’s optimised physical IP, combined with Glofo’s 28nm Gate-First High-K Metal Gate (HKMG) process.
ARM is banking on the fact that the partnership should help it push even further into the embedded space and secure a plethora of design wins with manufacturers of embedded devices like smartphones, smartbooks and tablets.
To add to its value proposition, ARM reckons the partnership addresses “increasing design and manufacturing complexities while reducing time to volume production at mature yields.”
Glofo says it expects to start production on the platform in the second half of 2010 at its Fab 1 in Dresden, Germany.
The fab Fab also expects to come up with “significant performance gains” over the previous generation 40/45nm technologies, meaning higher performance within the same thermal envelope. Practically, this means better application performance and increased multi-tasking capabilities on mobile devices.
It should also mean longer talk time, more multimedia playback time and even some half decent gaming and graphics before your battery dies.
ARM isn’t putting all its eggs in the Glofo basket, however. The firm has also established what it deems “strategic relationships” with other members of the IBM Joint Development Alliance to push ahead with the development of processors and physical IP tuned to the HKMG process.
And to prove it isn’t bluffing, ARM will be showing off its first 28nm wafer with HKMG technology at Mobile World Congress this week.
Stay tuned for pictures.