REGISTER | LOGIN
Breaking News
Blog

10nm Chips Promise Lower Costs

Handel Jones
6/15/2015 07:00 AM EDT

 38 comments   post a comment
NO RATINGS
2 saves
View Comments: Newest First | Oldest First | Threaded View
Page 1 / 4   >   >>
HJ88
User Rank
Author
Re: Is Moore's Law back?
HJ88   7/16/2015 5:43:30 PM
NO RATINGS
Dorian,

Thanks for your inputs.

We have wafer cost increase of 17.7% for 10nm over 16FF+.

Available gates per wafer increase is 39.3%.

We expect wafer cost for 10nm to be about $5.2K.

Key issue with 28nm is that depreciation declines by 50%+ in 2016, and cost per 100M gates will be around $0.90. 28nm technology has long lifetime.

Hope the explanation helps.

HJ88
User Rank
Author
Cost per Gate at 10nm
HJ88   7/14/2015 1:47:50 PM
NO RATINGS
If FD SOI is migrated to 10nm, cost per gate will be 25% to 35% lower than FinFETs with comparable power consumption and performance.

Supply chain is being built for FD SOI.

Dorian_
User Rank
Rookie
Re: Is Moore's Law back?
Dorian_   7/3/2015 4:38:28 AM
NO RATINGS
Handel, thank you very much for your inputs and this detailed analysis.

After reading your article I tried to understand how you get this output. At 10nm, assuming 21,930 gates/mm2, a parametric yield impact just below 50% (lets say 49%), gates per wafer would be around 679,000M. Then, $1.31 per 100M gates would means that 10nm shows a very strong increase in wafer cost compared to 16/14nm: wafer cost would be up by 120%. Your figures about previous nodes show that the most notable wafer cost increase was at 20nm with a wafer cost of $2982 (vs. $2126 at 28nm), i.e. a 40% increase.

I understand capital expenditure is increasing with 10nm, but does it really explains such an increase of wafer cost for 10nm?

Looking forward to hearing your feedback. Thanks again.

IJD
User Rank
Author
Re: Is Moore's Law back?
IJD   6/29/2015 10:14:36 AM
NO RATINGS
The physical design costs are higher because the tools are much more complex (to deal with double patterning) and the runtimes are much longer (many more complex rules and checks). More advanced tools would hopefully reduce this, but funnily enough the CAD vendors want a lot more money for them -- partly because they sell fewer of them, because there are a lot fewer designs than in 28nm -- so tool costs now exceed engineer costs per seat.

Software costs rise if system complexity rises, for example doing a multi-standard AP for mobile. For the same system complexity the software costs are the same, but 16nm has much higher physical design costs (and 10nm even more so). This is the category that the vast majority of cost-sensitive designs fall into, total device revenue is just too low to amortize the huge NRE costs.

Put more simply, most people in the chip design business aren't going to get their hands on 16nm (10nm even more so) for a long time -- if ever -- because it just doesn't make economic sense for them.

alex_m1
User Rank
Author
Re: Is Moore's Law back?
alex_m1   6/29/2015 8:30:49 AM
NO RATINGS
@IJD : thanks.

From what i'm reading ,around 70% of the NRE is software, and almost everythingis design costs and not physical artifacts like masks. So isn't it just a problem that  might be solved with better design tools(optimistically), and not a permanent barrier ?

IJD
User Rank
Author
Re: Is Moore's Law back?
IJD   6/29/2015 6:34:26 AM
NO RATINGS
Alex_m1, a rough rule-of-thumb is that a 1W power reduction saves about $1/year in electricity costs. Whether this justifies going to 14nm depends on the power saving, and whether this is actually seen by whoever has to pay for the chip. Most consumers don't think this way and wouldn't pay extra, many professional applications like data centres do and would if it made sense.

The fact remains that due to much higher NRE cost, to get the same ROI you need to sell several times as many 14nm chips (or get several times the revenue, to be more accurate -- but we assume cost for the same function will be similar) compared to 28nm. It's the high cost of entry to 14nm needing higher revenue that's the big problem for many applications, not the cost per chip.

HJ88
User Rank
Author
Re: Is Moore's Law back?
HJ88   6/24/2015 11:22:52 AM
NO RATINGS
Key issue for 28nm is that depreciation is nearing the end of cycle, and wafer costs will be $1,800 to $1,900 in 2016 for 10 layer metal.

The result is wafer selling prices with 50% gross profit margin can be $3,600 to $3,800.

A wide range of specialty wafer processes are being developed for 28nm, and it is significant that Xilinx, having obtained revenues of $1B from 28nm, is projecting design wins for many applications.

It is this type of competitive environment that 20nm and 16/14nm has to compete against.

10nm will be the high revenue technology node successor to 28nm.

However, it is likely that there will be special versions of 10nm, which could be called 7nm.

alex_m1
User Rank
Author
Re: Is Moore's Law back?
alex_m1   6/19/2015 6:18:14 PM
NO RATINGS
One interesting question is the TCO - for the same chip, if we calculate cost + energy cost , by how much do we see reduction in TCO for 14nm over 28nm ?


Because if this cost reduction is big , and 28nm costs the same as 14nm , i could see 14nm dominating in 14nm chips for businesses and probably consumers.

sranje
User Rank
Author
Re: Is Moore's Law back?
sranje   6/19/2015 4:53:39 PM
NO RATINGS
Dear IJD - thank you very much for your clarification. At 14nm a mobile processor functions are well defined, including for Intel integrated (BB+AP) mobile processor.

For the LTE BB certainly, silicon-integrated digital portion of Wi-Fi combo in BB also.

For processor we know that it will be IA x86-based rather than using ARM cores -- one of remaining questions will be - how many cores (4 or 8 - not likely there will be 10 of them as in the latest from Mediatek)

Hence if we assume that the complexity is well known and understood -- my understanding from the entire discussion is that because Intel has much greater interconnect density -- Intel's new mobile processor will have dramatically different cost than from foundries - correct?   Perhaps - but that assertion may or may not be correct - we will see.

Once again, many, many thanks

IJD
User Rank
Author
Re: Is Moore's Law back?
IJD   6/19/2015 11:02:14 AM
NO RATINGS
The issue is not whether 14nm can meet the market price requirements, it depends what is put into the chip. It the complexity is unchanged 14nm will probably be a little more expensive than 28nm but if it's half the power this is well worth it. If the system designer instead throws double the complexity at it, the  cost will be double and the power will be the same as the less complex 28nm chip -- or half the power of a same complexity one.

What's clear is that going to 14nm is a no-brainer if you need the power saving, it's much less clear whether it will be cheaper for the same function -- especially in devices like modems where the I/O and analogue area (which doesn't shrink as fast as digital) is significant.

Page 1 / 4   >   >>
Like Us on Facebook
EE Times on Twitter
EE Times Twitter Feed