Rory Buszka
Partition Master
If you're a PC enthusiast and you haven't heard of AMD, Intel, NVIDIA, and ATI, you'd probably better just quit now. Before you hurt yourself. Trundle on down to the Fry's, or the Best Buy, and just buy a HP, Gateway, or Compaq that costs what you're willing to spend, and be happy with it.
However, far fewer enthusiasts may know about TSMC, the Taiwan Semiconductor Manufacturing Company. They're the largest independent chip foundry in the world, and they essentially work on contract, building chips for other companies that simply design them. In addition to manufacturing logic chips for VIA, Broadcom, and Marvell, they manufacture the full line of GPUs and motherboard core logic chipsets for both NVIDIA and ATI, as well all AMD chipsets produced since the acquisition of ATI by AMD. (As a side note, Intel's chipsets are manufactured on outmoded equipment formerly used to produce CPUs.)
Some time ago, AMD announced a gradual transition to 'fabless' manufacturing for many of their product lines as part of their 'asset light' strategy to get the company back on track, financially. On August 5, a news story appeared on our front page news section detailing a leak of details regarding AMD's upcoming Fusion CPUs. If you don't know what Fusion is, that's somewhat forgivable, since talk of Fusion development has more or less been towed-under by the tidal wave of press that Intel's similar Nehalem product has been receiving. In that news story on our front page, we learn that AMD plans to have the Fusion's dual-core Phenom-derivative CPUs manufactured on TSMC's brand-spanking-new 40nm process node, and the onboard GPU produced on a TSMC 32nm process node, leapfrogging Intel's 45nm Yorkfield and Wolfdale CPUs entirely.
In the months following the release of Intel's first 45-nanometer CPUs, the tech press buzzed with talk about how AMD would be able to make a comeback in the process war with Intel, after falling a generation behind. (AMD's current Phenom X3 (Toliman) and X4 (Agena) CPUs are produced on a 65nm process.) The first products to come out of AMD on a process smaller than their existing 65nm silicon-on-insulator (SOI) process were the Radeon HD3000-series GPUs. The The hot, power-hungry RV600 GPU was built on a 90nm process, but its mid-range and low-end cousins RV610 and RV630 were built on a 65nm process, which used far less power and produced less heat. However, with the HD3000 series came an across-the-board die shrink to 55nm, smaller than NVIDIA's process for G80-family GPUs and the G92 on which the 8800GT was based. This process shrink was made possible by the fact that TSMC, which had already made the Radeon GPUs for AMD, and for ATI before that, had developed the 55nm process and was able to manufacture it with acceptable yields.
Now, with Fusion, it's becoming clear that AMD plans to use TSMC's process know-how to achieve a similar die-shrink on their CPUs. In the ongoing war for smaller and smaller lithography processes (which decrease power consumption an heat output), this makes TSMC not unlike a war profiteer, providing the products used by both rival factions. The AMD and NVIDIA GPU war comes to mind, since both product lines are manufactured by TSMC. AMD took advantage of TSMC's 55nm process, while NVIDIA's GPUs continued to be made on a 65nm process, likely not more than several hundred feet away. In the news article we published, AMD also plans to use TSMC's 32nm process for the "Bulldozer" CPU, which suggests that TSMC plays a major role in AMD's 'fabless' future.
The major question in my mind is this: Can it be in any way thought of as wise for a major CPU manufacturer like AMD to base their future process-shrink successes (and overall competitive edge therein) on the capabilities of a company like TSMC, which provides chip-fab services for direct competitors? If a company like Intel (which presently owns a large number of fabrication facilities in various countries) were to see the benefits of going asset-light and contracting a chip foundry to produce their newest and most competitive products, would there be any other company than TSMC capable of handling that task?
Discuss.
However, far fewer enthusiasts may know about TSMC, the Taiwan Semiconductor Manufacturing Company. They're the largest independent chip foundry in the world, and they essentially work on contract, building chips for other companies that simply design them. In addition to manufacturing logic chips for VIA, Broadcom, and Marvell, they manufacture the full line of GPUs and motherboard core logic chipsets for both NVIDIA and ATI, as well all AMD chipsets produced since the acquisition of ATI by AMD. (As a side note, Intel's chipsets are manufactured on outmoded equipment formerly used to produce CPUs.)
Some time ago, AMD announced a gradual transition to 'fabless' manufacturing for many of their product lines as part of their 'asset light' strategy to get the company back on track, financially. On August 5, a news story appeared on our front page news section detailing a leak of details regarding AMD's upcoming Fusion CPUs. If you don't know what Fusion is, that's somewhat forgivable, since talk of Fusion development has more or less been towed-under by the tidal wave of press that Intel's similar Nehalem product has been receiving. In that news story on our front page, we learn that AMD plans to have the Fusion's dual-core Phenom-derivative CPUs manufactured on TSMC's brand-spanking-new 40nm process node, and the onboard GPU produced on a TSMC 32nm process node, leapfrogging Intel's 45nm Yorkfield and Wolfdale CPUs entirely.
In the months following the release of Intel's first 45-nanometer CPUs, the tech press buzzed with talk about how AMD would be able to make a comeback in the process war with Intel, after falling a generation behind. (AMD's current Phenom X3 (Toliman) and X4 (Agena) CPUs are produced on a 65nm process.) The first products to come out of AMD on a process smaller than their existing 65nm silicon-on-insulator (SOI) process were the Radeon HD3000-series GPUs. The The hot, power-hungry RV600 GPU was built on a 90nm process, but its mid-range and low-end cousins RV610 and RV630 were built on a 65nm process, which used far less power and produced less heat. However, with the HD3000 series came an across-the-board die shrink to 55nm, smaller than NVIDIA's process for G80-family GPUs and the G92 on which the 8800GT was based. This process shrink was made possible by the fact that TSMC, which had already made the Radeon GPUs for AMD, and for ATI before that, had developed the 55nm process and was able to manufacture it with acceptable yields.
Now, with Fusion, it's becoming clear that AMD plans to use TSMC's process know-how to achieve a similar die-shrink on their CPUs. In the ongoing war for smaller and smaller lithography processes (which decrease power consumption an heat output), this makes TSMC not unlike a war profiteer, providing the products used by both rival factions. The AMD and NVIDIA GPU war comes to mind, since both product lines are manufactured by TSMC. AMD took advantage of TSMC's 55nm process, while NVIDIA's GPUs continued to be made on a 65nm process, likely not more than several hundred feet away. In the news article we published, AMD also plans to use TSMC's 32nm process for the "Bulldozer" CPU, which suggests that TSMC plays a major role in AMD's 'fabless' future.
The major question in my mind is this: Can it be in any way thought of as wise for a major CPU manufacturer like AMD to base their future process-shrink successes (and overall competitive edge therein) on the capabilities of a company like TSMC, which provides chip-fab services for direct competitors? If a company like Intel (which presently owns a large number of fabrication facilities in various countries) were to see the benefits of going asset-light and contracting a chip foundry to produce their newest and most competitive products, would there be any other company than TSMC capable of handling that task?
Discuss.
