Comment Re: Probably not. (Score 1) 114
You simply make it stop being a secret just before the exam starts.
You simply make it stop being a secret just before the exam starts.
Not 14 mm. Their claims of 28 nm are doubtful. They buy litho tools from outside and they don't have permission to buy 28 nm grade tools.
It has dual FP units. So what. Blue Gene also looks good FP wise and the design is mostly crap.
Intel uses 14 nm. SMIC is full of shit. They mostly steal their process from TSMC and TSMC has been doing poorly of late.
It is not that easy. Even if China threw its entire financial and manned weight at the problem it would take them 10-15 years to catch up and that's assuming they did not slip up along the way. There's just too many things they don't know how to do. Those "Europeans" are the Dutch which are a member of NATO. They won't export advanced litho tools to China either. Nor will the Japanese.
What the Chinese will get is like the 65 nm tools they have. Four processes behind i.e. 6 year old tools.
1.1 GHz. Two integer and two floating point pipelines. 7-stage, 40-bit physical address. 8 KB L1 cache, 96 KB L2 cache. These are worse specs than those of a 32-bit AMD Athlon processor from 1999.
But hey it has 16 cores. They are manufacturing it at 65 nm after all. Intel uses 14 nm right now:
http://en.wikipedia.org/wiki/6...
That's four manufacturing process generations behind what Intel uses. It you assume Moore's Law happens every 18 months this would be like using a six year processor even if they had bleeding edge processor design. Looking at the specs of their processor shows their design skills actually are even worse than their manufacturing skills. Like I said AMD had a better core design than that in 1999. That's 16 years ago.
Unfortunately it is quite common to see people underestimate how chip manufacturing works. Last time Apple said they would just switch semiconductor manufacturers, even despite having their own chip design team, it took them 2-3 years. Just to port the design to a different factory. Let alone design and manufacture the chip manufacturing tools, R&D the semiconductor process, the chip design tools, the chip design, ad nauseum.
It is non-trivial to reverse engineer advanced lithography equipment. This equipment is sold in small numbers and is highly controlled. Without the tools they cannot manufacture a damned thing.
Well they are. The design is made in the USA or Israel and the manufacturing is done in Oregon, Arizona, New Mexico, Israel, etc. The list of sites is here:
http://en.wikipedia.org/wiki/L...
Costa Rica is where Intel does wire bonding and puts the metal head spreader on the chip. Wire bonding is labor intensive so cheap labor is important to make it cost effective. It is not where they manufacture the chip. Malaysia is where AMD does wire bonding and attaches the metal heat spreader.
Intel and AMD usually say 'diffused in XXX' which is the hard and export restricted bit, and 'assembled in YYY' which is hand labor intensive non-export restricted bit.
All they need is to declare it a secret.
The number of clients who buy Xeons and Xeon Phis is quite limited. It's less hard to track it down than you think.
Intel does not manufacture CPUs in East Asia. The only 'Asian' manufacturing site is in Israel.
Apple assembles things in China. The actual chip manufacturing is done in Taiwan and the USA.
They don't have the machine tools, they don't have the knowledge to do advanced chip design, they can't do it. Period.
Sure they might do it with a monster project in like a decade, but by then the industry will have moved beyond the current point.
They can but it will be a lot more expensive and you can bet getting the tens or hundreds of thousands of chips they need for a supercomputer WILL be noticeable.
"Been through Hell? Whaddya bring back for me?" -- A. Brilliant