A top engineer from Nvidia Corp. said that as chips become more complex in terms of transistors count and sometimes in terms of physical sizes, the industry will need to move to larger 450mm wafers in order to profitably build those devices. The engineer predicted that by 2020 chips will contain as many as a trillion transistors, or over 232 times more complex than the most complex graphics chip today.
"A diverse host of new technologies and methods will be needed to keep the industry on track to profitably deliver a trillion-transistor device by the end of the decade. [...] The industry needs to move to 450mm wafers to deal with the increasing number of masks and process steps required to make chips," said Sameer Halepete, vice president of VLSI engineering at Nvidia, reports EETimes web-site.
The larger 450mm wafers are supposed to decrease pricing of every single chip located on them, but fabs that can process such wafers will be much more expensive, which will require them to work at 100% utilization all the time to be economically feasible. A number of challenges exist on the route of 450mm semiconductor factories, but with more and more companies engaging into the work, they are projected to be generally solved by the middle of the decade.

With process technologies becoming more complex and harder to implement, it now takes many months to initiate volume production of new chips, whereas windows of opportunities for those chips remain the same. It is crucial to deliver a new graphics processor by Christmas season every year and once companies like Nvidia or AMD fail to do so, they essentially lose money. 450mm wafers allow to produce more chips at once, therefore, can either shorten cycles or leave them in current timeframes.
Another major problem in today's chip design is extreme complexity amid necessity to reduce power consumption while increasing performance. Back in the days, engineers optimized chip designs "by hands", but today it is virtually impossible and special computer programs are used to tune the designs. Those tools will need to get more sophisticated in the future to enable even more advanced chips. Nvidia is currently working with Mentor graphics on appropriate software.
"New process technologies are running out of steam in their ability to lower power because voltages are not decreasing significantly. Thus the next wave of improvements in energy efficiency will come from tools that can suggest optimizations in logic and circuit designs. It is hard for engineers to find more than half or two-thirds of these opportunities," explained Mr. Halepete.
Tags: Nvidia, Geforce, Quadro, Tesla, 450mm, Semiconductor
Comments currently:
5
Discussion started: 04/13/12 09:00:54 AM
Latest comment: 04/18/12 12:04:45 AM
Expand all threads |
Collapse all threads
[1-3]
1.
While 450mm wafers will show up in the not too distant future, it might be better for Nvidia to get their engineering sorted out so they don't need to scrap 80% of their chips due to engineering/design issues?
1
1
[
Posted by: beenthere

|
Date: 04/13/12 09:00:54 AM]
+ expand thread (2 answers)
- collapse thread
Agreed, plus almost total reliance on automated design software is slowly killing the advancement of gpus and especially cpus. They should do the hands on early on that should be scalable and efficient then use VLSI software from there.
0
0
[
Posted by: nforce4max

|
Date: 04/13/12 10:52:32 AM]
"they don't need to scrap 80% of their chips due to engineering/design issues?"
Source?
Design issues? I see a smaller chip, that runs cooler, consumes less power spanking HD7970 for less $ too. It's like a GTX670Ti beating AMD's flagship. I'd say Kepler architecture is superior for gaming than GCN is. 1536 shaders vs. 2048 SPs. AMD has its work cut out for them next round.
NV simply decided not to buy wafer capacity since they knew they weren't ready to launch their cards for a quarter. That's why we are seeing shortages since other customers of TSMC have purchased the available wafer capacity and now NV has to compete with them for wafer capacity.
The reason HD7970 isn't sold out is because no one wants it for $530-550.
Your comment that NV only has 20% yields (i.e., 80% of chips are scrapped) is unsubstantiated by any evidence. NV has demolished AMD in profits for the last 2-3 years, easily too. Do you think that would have happened if NV was yielding 20%?
The shift to 450mm wafers to reduce costs is beneficial for everyone in the industry, not just for NV. They are doing just fine selling 294mm^2 chips for $500 though.
It's funny how anything NV says to promote advancement/innovation in the industry is automatically painted in negative light. The transition to 450mm was expected regardless of NV's comments. It's an engineering problem, not just an NV issue. It's great to see reasoning for why 450mm wafers will be beneficial coming from one of the customers.
BTW, NV's "design issues" have just caused HD7970 to drop to $479 and HD7950 to $399. Not bad considering GTX680 isn't the flagship Kepler card.
http://www.hardwarecanuck...7970-price-drop-incoming/
I'd like NV to have more of these "design issues" you speak of so I can pick up an HD7970 for $349 once GK110 launches.
show the post
0
3
[
Posted by: BestJinjo

|
Date: 04/13/12 01:40:50 PM]
2.
0
1
[
Posted by: beenthere

|
Date: 04/13/12 02:27:53 PM]
3.
Financial statements can only be falsified for so long, I really wonder how NV have got away with the lies for so long
0
1
[
Posted by: alpha0ne

|
Date: 04/18/12 12:04:45 AM]
[1-3]