TSMC has been the industry leader for a long time, enabling them to bring out advanced geometries fairly quickly. However, they have grown so large that leading edge companies are reliant on TSMC's production prowess, and any delays on a new process could cost them millions in revenue.
This has been the case for ATI, the graphics chip division of Advanced Micro Devices. ATI first began manufacturing the RV740 GPU used in the Radeon HD 4770 at the end of 2008, using TSMC's 40nm process technology. However, they weren't able to get a high enough yield rate for volume production until April, leading to a delayed launch.
The problems have continued for the rest of the year, as ATI launched the first DirectX 11 graphics cards and transitioned an entire generation of graphics chips to the 40nm process. The Radeon HD 5800 series was launched in September and has been extremely successful, so much so that it became production constrained due to yield issues at TSMC. ATI was selling every 40nm Cypress chip it could produce, but couldn't meet demand until TSMC was able to increase the supply of 40nm wafers in November. This led to increased supply in December, just in time for ATI to meet holiday demand.
There are now six discrete desktop graphics cards using 40nm chips from ATI, with two more about to be released. ATI also recently announced DirectX 11 Mobility Radeon GPUs for notebooks at CES. The company has sold over 2 million DX11 GPUs, all built using the 40nm process. While this may sound like a lot, ATI sells around 35 million GPUs per year, most of them below the $100 mark.
DailyTech spoke with a TSMC spokesperson yesterday, who stated that TSMC's 40nm yields are now "approximately at the same level" as the more mature 65nm process. Semiconductors are made in lithography chambers, and the process can be comprised of several hundred steps. Usually a new manufacturing process is developed and refined in a test fab and then transferred to production lines in a process called Chamber Matching. This theoretically ensures standard conformity and higher yields. There were several problems with chamber matching on TSMC's 40nm lines, leading to yield problems despite using the same process and recipes.
This is good news for ATI, as they prepare to launch more sub-$100 graphics cards for the cost-conscious value market. Rival NVIDIA must also be celebrating, as they recently began volume production of their next-generation GF100 graphics chip using their new Fermi architecture. The GF100 is expected to have a die size larger than 500mm^2, which makes having a good yield rate even more critical.
Having resolved its production problems, TSMC is looking ahead to the future. The company is betting big on the next-generation 28nm process node, and is finishing shell construction of Phase 5 of its Fab 12 facility. Initial 28nm production will start in the third quarter, ramping up to volume production in the fourth quarter. 28nm devices are expected to cut power usage in half, increase performance by 50%, and cost less due to smaller die sizes.
The new section is adjacent to Phase 4, which began volume 40nm production in the third quarter of last year. TSMC has estimated that it will spend over $5 billion on building and equipping Phase 4 and Phase 5. The total cleanroom area for both phases is approximately 22,700 square meters, with 2 stories underground and 4 stories above ground.
TSMC began a large-scale recruitment campaign at the beginning of January, and expects to hire more than 3,000 staff, primarily semiconductor engineers and production staff.