TSMC在北岛上让ATI浪费半年？

CoreSolo

http://www.semiaccurate.com/forums/showthread.php?t=1934

What can I say about the execution of TSMC lately besides the phrase "in total mess and chaos"? The problem and issues with their 40G (for making desktop GPUs for ATI and NVIDIA), the sudden cancellation of 32 nm bulk silicon process and the move to gate-last techniques for 28 nm bulk silicon process. Many people are saying there's something going on inside TSMC, well I'll let you be the judge on this.

And the intelligence from China just came in this very moment, here goes: "TSMC's cancellation of 32 nm bulk silicon process caused ATI to waste half a year of time, [the foundation chip* of] Northern Islands may tape-out on 40 nm instead".

So what actually happened? The 32 nm version of that foundation chip was taped-out already and was scheduled to be released in Q3 2010 if all goes well. But then TSMC cancelled 32 nm gate-first bulk silicon process in short order and worked on 28 nm gate-last bulk silicon process instead, and that effectively made AMD's efforts over the past half a year in vain.

And that's not about it, it's probably some more bad news for the red team: TSMC's 28 nm will not come this year (official schedule: risk production in Q4 2010, and ramp up in Q1/Q2 2011), and that means AMD can't wait for TSMC's 28 nm to get ready in Q4 2010 for taping-out the foundation chip of the Northern Islands family. If AMD did wait for TSMC's 28 nm for re-tapeout, then at least it will be a Q2/Q3 2011 release instead of the initial aim of Q4 2010 introduction, early 2011 with expected delays.

So, what're the options AMD currently have if they wanted to get the chip out by that timeframe? The only option: 40G from TSMC.

Why not GlobalFoundries? Well, 32SHP SOI is not a good option for a performance chip (it will be good for the integrated graphics part in the APU), and 28 nm bulk silicon will not be considered for the foundation chip, albeit Northern Islands family is dual sourced. And since the tape-out will be made on 40 nm instead of the smaller 32/28 nm process, that would probably mean some effects to the hard specifications, such as "shader core" count, TMU, ROPs and things like that.

Also, currently all GPUs are made on gate-first process technologies, moving to gate-last would mean some more homework will have to be done, and that takes time, not only AMD but also NVIDIA, Xillinx and Altera, notice all of them are "transistor count and density comes first" kind of customers (AMD adds one more: "transistor budget/die area comes first"), while GlobalFoundries did say about designs fabricated using gate-last techniques may result in 10% bigger die compared to the same design using gate-first techniques. That would result in serious considerations for the red team to continue on using TSMC for future GPUs.

So, another "OUCH award" to TSMC for the poor execution and wasting everybody's time. And let's see if the dual source decision will give ATI a clearer picture about the two pure-play foundry partners' capabilities and decide on what to do next with their GPU fabrication outsourcing. Probably AMD will consider UMC also?

* the foundation chip here means the performance GPU of the Northern Islands family.

什么？

AMD设计32nm的北岛，结果TSMC取消的32nm，而28nm又得等明年，AMD只能重拾40nm

PS：TSMC这不是逼ATI用GF么

mindfury

老外论坛上一个普通YY帖子罢了。。

zsy19860820

台积电已经被所有的fans骂了好几遍了

红发IXFXI

tsmc的工艺是#@%#￥%#￥%#￥%