Advanced Micro Devices said last week that it would ship its first products made using 20nm process technology at TSMC in 2015. The company specifically decided not adopt the 20nm fabrication process among the first in order to ensure high yields and thus low production costs. The chip designer intends to use 20nm fabrication process to make different kinds of products.
“20nm is an important node for us. We will be shipping products in 20nm next year and as we move forward […],” said Lisa Su, senior vice president and chief operating officer of AMD. “If you look at our business, it is quite a bit more balanced between the semi-custom, embedded, […] professional graphics […] as well as the more traditional sort of client and graphics pieces of our business. [20nm] technology plays in all of those businesses.”
This is the first time when AMD confirmed that it will use TSMC’s 20nm fabrication process to make graphics processing units, semi-custom Fusion system-on-chips, solutions for embedded applications as well as code-named SkyBridge accelerated processing units (which will feature either ARM Cortex-A57 or AMD Puma+ cores).
It should be noted that Taiwan Semiconductor Manufacturing Co.’s 20nm manufacturing technology is a universal fabrication process designed mostly with various system-on-chips in mind, which means that the technology is not tailored particularly for accelerated processing units or graphics processing units.
Today TSMC offers several versions of its 28nm production 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, but which will be used for graphics processors as well.
TSMC’s 20nm process technology should help AMD to boost functionality and performance of its products and cut down their power consumption.
According to market rumours, the yield of chips made using TSMC’s 20nm process technology is pretty low, which is a problem for fabless designers of semiconductors as nowadays TSMC charges them for the whole wafer, not individual chips (i.e., they pay for both functional and faulty chips).
Tags: AMD, Radeon, Fusion, GCN, 20nm, TSMC, APU, GPU, Semiconductor, Puma+, Skybridge
Comments currently: 7
Discussion started: 07/25/14 12:24:03 PM
Latest comment: 08/23/14 02:00:17 AM
Expand all threads
| Collapse all threads
You guys do realize this is just the nice way of saying "no 20nm product for you in 2014", right ? TSMC has been bragging about the volume of 20nm production [of vaporware], there has been speculation regarding NVidia's (in)ability to release 20nm Maxwell in 2014, now this is a confirmation from AMD that they won't be releasing 20nm product in 2014 either.
Bottom line being: TSMC is full of hot air, no real 20nm products are going to hit the market in 2014. No wonder Qualcomm is working out an arrangement with Samsung (their 14nm looks promising).
07/25/14 12:24:03 PM]
- collapse thread
"TSMC has been bragging about the volume of 20nm production"
Actually it has been the opposite, they have been cautious and admitted the low run quantities being produced will not be enough to satisfy the markets.
By holding back and stockpiling quality and quantity they at least wont flood the market with an ill prepared chip in the rush to be the first.
07/27/14 12:42:43 AM]
Face it they all... TSCM, Nvidia, and AMD have had to tread lightly on this subject. They all see it as a PR disaster to utter anything truthful or that paint TSCM in a bad light. But right you are this is saying 20nm is not ready for primetime, and likely won't be till sometime like end of Q2 for any true volume/market release.
I'm still crossing my fingers that "Tonga" comes out of GloFo, as that would be the first true revelation that TSCM GPU monopoly could be waning and there could be some true divergence in the landscape.
07/29/14 01:17:16 PM]
Sounds like great news in 2015.
It means that AMD is narrowing the nm gap,
from 10nm to 6nm difference with Intel.
08/08/14 08:22:24 PM]
AMD has been here before many times. They have learned a few things about moving to smaller nodes vs the demand for their products.
I believe it will be at least mid-year before AMD actually begins shipping a 20nm product. I'll bet they do start hyping up up early next year though to try to amp up sales when the chips are available.
I also believe that it will be well into 2016 before a majority of AMD's new products are 20nm.
The Old Fart
08/17/14 04:20:56 AM]
As has been documented already in the drop from 32nm to 28nm to 22nm the primary performance gain is lower power consumption, not increased clockspeed. Thus there is no panic to move to 20nm especially in a down PC market.
08/21/14 01:07:17 PM]
- collapse thread
It simply means that the power budget will need to be spent in transistors rather than clockspeed, which is fine given the later is (generally) less power-efficient even on mature HP processes.
In the past, this transition would of primarily hurt AMD, as they usually went for small, efficient designs that clocked higher (and burned power at those high clocks) to compete. Now, with designs like Hawaii (and even later bins of Tahiti/Pitcairn) AMD seems to have embraced more units at a lower clock. Of course on the flipside, you have nvidia, whom have had a role reversal in the exact opposite way. If they were to stick to a a similar design mentality as something like GK104 (which is highly dependant on clockspeed, GM204 may be as well) they would have a lot of issues.
20nm will simply require more logic (proportionally) per market segment to compete than before. Because of this, it makes sense to want more mature yields. You can bin a chip to leakage factors (a primary reason it made sense to jump earlier before) but when you're not only talking more transistors, but more transistors proportionally on a process with less wiggle-room for clocks, you want as many good chips as possible.
That all said, it's worth noting while she may indeed mean 20nm, '16nm' (which will primarily increase clockspeed instead of density and be more suitable for GPUs) could also be classified as '20nm'...16nm of course is (mostly) simply 20nm with finfets.
08/23/14 02:00:17 AM]
Add your Comment
Enter your username and e-mail address. Password will be sent to you.