ZT: TSMC 40nm yield issues explained - 电工杂谈 - 电子工程师俱乐部

2dmin

28nm coming this year, 22nm in Q3 2012
By Paul Taylor
Fri Feb 26 2010, 06:14



<strong>TSMC HAS FINALLY</strong> come to terms with its 40nm ordeal. In a transcript published by EETimes, Shang-Yi Chiang, TSMC’s R&D boss, laid out six bullet points to explain the situation the company is coming from, what steps it’s taking to make amends and where it's going.
The 40nm yield issues were blamed on the company’s previous lack of experience with the new, smaller node. Without going into too many specifics, Chiang explained the delay in achieving a viable yield was due to it using 193nm shrink immersion on the wafers, which resulted in a high defect rate, and also its low-K process that would damage the dies when converted into a package. TSMC is looking at extreme ultraviolet and e-beam direct write as alternatives to shrink immersion depending, of course, on cost.
Having its clients breathing down its neck couldn’t have been easy. TSMC immediately started developing its 3rd generation manufacturing process on the node. Currently, output for 40nm wafers is 80,000 wafers per quarter, as only Fab 12 is manufacturing these. However, the company vowed it will be able to double that by year’s end, that is, 160,000 wafers per quarter, once Fab 14 is up and running.
He then moved on to explaining what the company plans to do next with its 28nm roadmap. It will start with a 28nm Low Power (28LP) silicon oxynitride, followed a quarter later by the 28nm high performance (28HP) high-K metal gate process. Despite it being the first high-K metal-gate process, Chiang believes that TSMC will be able to migrate the technology as it moves forward with 22nm and 20nm nodes. 28nm should be introduced by Q3 2010, we garnered.
Further along the roadmap comes the aforementioned 22nm node. Chiang expects to see a more mature second-generation high-K metal-gate process kick in. The first 22nm process will be a high-performance one (22HP), followed 2 quarters later by the low-power version (22LP). These should be in place by Q3 2012 and Q1 2013, respectively.
Considering TSMC’s major clients are the likes of AMD, Nvidia, Broadcom and Qualcomm. not to mention Intel, it seems that it's taking fairly important measures to reassure its clientele about its manufacturing. Of course we should expect a reaction from Global Foundries soon, as it are developing competing processes in their own house.
So, if you’re a GPU buff, late 2010 or very early 2011 would be a good estimate for a new generation of GPUs. That, or smartphones. We'll have to wait and see. >µ