X-bit labs

NVIDIA, Xilinx Disappointed with IBM Microelectronics’ Foundry.

Category: Video

by Anton Shilov

[ 11/07/2003 | 11:26 PM ]

NVIDIA and Xilinx are reportedly planning to increase orders to TSMC and UMC this quarter, as they are not satisfied with their work with IBM Microelectronics. There have been a lot of rumours concerning IBM’s problems with its fabrication technologies for 3 or 4 months, but this is the first time when we are hearing about some definite actions of IBM’s customers.

According to DigiTimes, IBM has been experiencing low yield at the 0.13 micron and 90nm nodes due to its problem on SiLK-based low-k dielectric – made by Dow Chemical – that has problems with thermal expansion mismatch with other materials in the transistor. IBM is not going to begin using other materials, such as vapor deposition based dielectrics, to address the issue for quite some time from now, hence, it becomes rather tricky for its partners to work with the East Fishkill, New York-based foundry.

The source notes TSMC’s and UMC’s ability to output 20% “more good die” on conventional 200mm wafers and also more mature 300mm wafers technology ready to go into mass production.

As a consequence of problems with IBM Microelectronics, some of its potentially large customers, such as Santa Clara, California-based NVIDIA Corporation and San Jose, California-based Xilinx, intend to increase their orders to the primary partners, such as TSMC and UMC and substantially decrease outsourcing to IBM. As a result, both companies do not plan to increase their orders to the US-based maker unless their Taiwanese partners are unable to satisfy their needs.

Xilinx currently orders practically all its 90nm and 0.13 micron FPGA programmable logic to UMC. NVIDIA also had to redesign its GeForce FX 5700 also known as NV36 – its first GPU to ramp up at IBM – for production at TSMC.