Many criticize the world's largest contract maker of chips of inability to supply enough chips made using leading edge 28nm process with high-k metal gate (HKMG) technology. In a bid to avoid a similar situation with 20nm fabrication process, Taiwan Semiconductor Manufacturing Co. plans to cut some corners and options.
The difference between chips for different applications nowadays is so significant that in order to build them contract makers of semiconductors develop special versions of each process technology in a bid to meet particular requirements. Nonetheless, TSMC intends to offer only one flavour of its 20nm fabrication process, but to consider possibility to introduce "half-node" 16nm or 18nm technologies.
Nowadays Taiwan Semiconductor Manufacturing Company offers four versions of 28nm process technology: 28LP (poly/SiON) for low-power cost-efficient chips, 28HPL (HKMG) for low-power applications, 28HP (HKMG) for high-performance chip designs and 28HPM (HKMG) that combines elements of high-performance and low-power process technologies and is mostly designed for microprocessors for tablets, superphones and notebooks. With 20nm, TSMC will not give any choice and there will be only one option, the company said at its annual technology symposium.
Shang-yi Chiang, executive vice president and co-chief operating officer at TSMC, said the firm might offer an 18nm or 16nm process node after 20nm if lithography technology is not available to make 14nm devices cost effectively, reports EETimes web-site. There is a huge difference between 20nm and 14nm fabrication processes across the industry since at 14nm a number of innovations, such as different - FinFET 3D - transistor structure, come into play. In case it is impossible to make 14nm chips in a cost efficient manner initially, chip designers may rely on half-node technologies like 16nm or 18nm.
Another major concern for TSMC is that extreme ultraviolet (EUV) lithography that is needed to produce chips at 14nm process will not be available by 2015 because it consumes too much power and no appropriate power sources are available at this time for it. Manufacturers of lithography, such as ASML Holding NV, have managed to tweak 193nm immersion lithography tools for 14nm fabrication technology, but that might require triple-patterning on some layers and double patterning on many layers in order to achieve adequate image fidelity, which would make it too costly for volume production.
Tags: TSMC, 20nm, 14nm, 16nm, 18nm, 28nm, Semiconductor, EUV
Comments currently: 3
Discussion started: 04/19/12 12:36:36 AM
Latest comment: 03/09/16 03:55:57 PM
Expand all threads
| Collapse all threads
This actually isn't such a terrible idea at all.
Figure the standard operating voltages for HP and HPL, 28nm or otherwise. One is tuned for low-leakage aimed slightly above 1v (Krait is somewhere around 1.05v iirc). The other is aimed at high-performance/leakage that can increase performance up to around 1.3v (going by 7000 series radeon series) depending on the chip.
Disregarding the fact the average between those is exactly 1.175v, think about where most graphics chips have been aimed for both 40nm and 28nm. Nvidia put OVP at 1.175v for both processes, while AMD stock-volted 6870/6970 and 7970 all at 1.175 volts. While certainly not the best baseline perf/w, which is typically 1.05v ala many AMD xx50 series parts, nvidia stock voltage for pretty much all chips, and of course all mobile chips using the LP processes, 1.175 has been proven to be the point where the extra clockspeed afforded by a higher voltage breaks the leakage/power curve and returns diminish; it could be argued all-around this is the level where the amount of transistors in a design for almost any market should be aimed.
When you look at this on a per-market level, it cuts both ways.
Obviously we should very small chips for mobile companies; it should allow a pretty straight-forward migration from 28n because much of the tdp improvement given by process node will be used for the higher-voltage/clockseed.
On the flip-side, this should allow graphics to consume less power/volt/mm, while still hitting their goals for the process if similar to 28/40nm. It may suck for overclockers, but bodes well for greater IPC/larger chips (which is something we may just need to supply adequate bandwidth). Considering the small (less than ~200mm, 256-bit) gpu is dying a slow death to CPU IGPs and the company that would use this tech for their processors aims them in the greater than 200mm range, it correlates pretty well.
04/19/12 12:36:36 AM]
- collapse thread
Good points and by far a better read than the news article itself!
I was first thinking it might be a bad idea TSMC shooting itself in the foot, then I realised it already is stretching too much and that's where majority of their problems regarding 28nm nodes come from. On the other hand (pardon the pun) it might not be the best idea to cut off patient's limbs just to keep it alive. Probably premature, but I'd wager these plans change substantially before soon. TSMC has reached a certain fame of great appetites and I highly doubt they'll settle for a single dish dinner. It might drive their customers, picky as they are, to order out (Intel, GF) even more often than some now speculate they might (e.g. Qualcomm). It better be a 450mm dish then TSMC!
04/19/12 10:14:44 AM]
Add your Comment
Enter your username and e-mail address. Password will be sent to you.