REGISTER | LOGIN
Breaking News
News & Analysis

Moore’s Law’s End Reboots Industry

CMOS scaling is ‘not dead, just resting’
6/26/2017 01:01 AM EDT
3 comments
NO RATINGS
2 saves
Page 1 / 3 Next >
More Related Links
View Comments: Newest First | Oldest First | Threaded View
Eudecio
User Rank
Rookie
Eletronics will go beyond moore's law.
Eudecio   6/27/2017 7:50:45 AM
NO RATINGS
That's what was said before modern physics.

realjjj
User Rank
CEO
...
realjjj   6/26/2017 7:34:16 PM
NO RATINGS
The main problem with such a panel is that they are focused on server and that's not a huge part of the market nor does it define it. I do like their attitude though.

Moore's Law is about cost per transistor and perf. Design costs are also a huge problem today. Cost per transistor is a manufacturing issue while design costs are an economic issue, ideally you want to re-architect both product and manufacturing to address those but don't ask me how LOL. Anyway, the main problem is perf scaling and how difficult that is depends on how the market defines performance, likely efficiency is what will matter most.

For memory, you got 4 types of scaling: horizontal, vertical, bits per cell and perf. Hard to claim that 3D NAND is great from that perspective as it is hitting all the limits it can hit. It can still scale but not very well or for too long.

Then you got the devices, where are things going? Server is fine with very high power and relatively average efficiency and it's a lot about moving data not processing it. The trend for the broader market is for devices to shrink, leading to severe power and thermal constrains. Glasses replace PCs , phones and TVs, robots start to take off, IoT is already here. You can't turn HoloLens into a real product if it uses a 5W SoC +DDR5, that's crazy. Ofc if these markets require extreme efficiency, server would follow. For server, folks also need to account for electricity costs as those are bound to decline and it does make a difference - it's another type of scaling, costs shift from OPEX to CAPEX, thermal gains in relevance vs power. When electricity costs approach zero, using idle capacity for cryptocurrency mining can be a real opportunity in reducing costs - can be done even today, creates economic value but pollutes so not doing it just yet would be preferable.

Moore's Law applies pressure from a single direction but nobody should ever do just that, it should be more like an implosion, applying pressure from every possible direction- when that's not enough anymore, change the game. Humans have problems with context too and the more context we ignore, the poorer our solutions are. Moore's Law today is like making the perfect chess move while the room is on fire.

fragro
User Rank
Author
So what is the problem after all?
fragro   6/26/2017 5:40:47 AM
NO RATINGS
It looks as if we are trying to find the answer to a question which has not been formulated precisely, yet.

Ok, the microprocessor kernel cannot be made smaller anymore, however, lets take a look at the current, growing set of demands, and then decide what we really would like to achieve.

Applications for computing:
  • Computing (yes, indeed:-)) of well defined mathematical structures
  • Database search and query
  • Forms based data capturing, maybe dynamically, multi user, globally on distributed data sets
  • Global search engines on unstructured data
  • Video editing
  • Speech recognition
  • Spectral graph analysis on complex graphs
  • Chaotic systems 
  • Neural networks for image recognition, etc.
  • Video Games
  • Artificial Intelligence based on logical deduction and induction
  • Augmented reality

And, after all, everything with the same little microprocessor based on a small set of serially interpreted instructions, going through the same little door?

And while the answer is already there, it looks as if Intel is the only company trying to provide a more or less strongly integrated tool set(?). 

 

 

 

 

 

Like Us on Facebook
EE Times on Twitter
EE Times Twitter Feed