"Moore's law is alive and well because it gets redefined every two years."
CHeers
Phil Hobbs
--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC
Optics, Electro-optics, Photonics, Analog Electronics
160 North State Road #203
Briarcliff Manor NY 10510
hobbs at electrooptical dot net
http://electrooptical.net
Well it turned into a sigma curve as it must. But it hit the buffers at ab solutely astronomical performance rates and numbers of active elements. It happened so fast that people don't really understand what they have in thei r possession. In terms of human level AI current CPUs are more than fast en ough, the main restraint is memory. The more the better. What is a neural net except an analog hash table? Then it is just a question of how you con nect these hash tables together and what learning mechanism you apply. That is all doable with current hardware. It is just taking humans a bit of tim e to catch up with the extraordinary hardware that has been created. Google DeepMind are showing what is possible. Even they though are not using the best types of neural net. Anyway memory density is still growing exponentially, with the possibility of nearly infinite memory in 3D cubes containing nanoparticles. Well nearl y infinite memory would allow construction of something well beyond human c apacities. I would say that within 5 years you are going to see some pretty weird thin gs happening. At a minimum you are going to see a surge in industrial autom ation. Particularly the cost of automation is going to fall dramatically. That's quite positive, if you want to do something or make something then t he burden of finding money, week in week out, for wages is gone.
Gordon Moore said something doubles every 18 months. Not two years. That two year rumor is just proof of the slowing down of progress. It was not a Law, it was a trend that is over, now.
From wikipedia, "The period is often quoted as 18 months because of Intel executive David House, who predicted that chip performance would double every 18 months (being a combination of the effect of more transistors and the transistors being faster).[17]"
"Moore's Law to roll on for another decade". Retrieved 2011-11-27. "Moore also affirmed he never said transistor count would double every
18 months, as is commonly said. Initially, he said transistors on a chip would double every year. He then recalibrated it to every two years in
1975. David House, an Intel executive at the time, noted that the changes would cause computer performance to double every 18 months."
Initially Moore *observed* a doubling every year in the number of transistors on a chip and projected this would continue for a decade. A decade later he changed the projection to a doubling every two years. This has held pretty well for decades now.
It's good for a while longer, maybe a bit slower progression, but the cost of design and masks for 5 nm multi-billion-transistor chips is going to mean that only mega-volume-sales chips will be at the leading edge. Maybe Gordon didn't think about the ultimate limit being cost and complexity, rather than resolution.
--
John Larkin Highland Technology, Inc
lunatic fringe electronics
Back in the palmy days of Mead-Conway VLSI scaling, transistors got faster as they got smaller. Thresholds went down with gate oxide thickness, so you could reduce VDD and save power. It was very nice--for a new device node, we just needed the litho folks to get off their duffs and improve the resolution.
'Taint like that any more--not since 65 nm or so. As transistors shrink, they now get slower. And leakier. And the threshold voltages get more and more poorly controlled due to dopant atom statistics. And we've about run out of entries in the periodic table to try fixing those problems. (Hafnium oxide gate dielectric, for instance.)
So you need to throw lots more transistors at the same functions to keep them working right, and turn off most of the chip most of the time to prevent it from turning to lava. (That's a bit of hyperbole, but less and less so.)
And chip area is limited to about 2 cm square by yield and thermomechanical stress issues. It's been stuck there for decades.
That's why Moore's law is over. You can concentrate on counting transistors, but if most of them are there to compensate for the badness of the rest, it's sort of academic.
The CPU speed of my desktop computers peaked in about 2006, at 3.5 GHz. My present Supermicro boxes run at 2.3 GHz. When I left IBM in 2009, their Deep Blue supercomputer ran at (iirc) 700 MHz for power reasons.
Cheers
Phil Hobbs
--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC
Optics, Electro-optics, Photonics, Analog Electronics
160 North State Road #203
Briarcliff Manor NY 10510
hobbs at electrooptical dot net
http://electrooptical.net
I get all kinds of inquiries from managers wringing their hands... "I spent all this money at BigNameExpensiveConMen, Inc, for analog functions on 65nm and they don't work for shit." >:-} ...Jim Thompson
--
| James E.Thompson | mens |
| Analog Innovations | et |
| Analog/Mixed-Signal ASIC's and Discrete Systems | manus |
| San Tan Valley, AZ 85142 Skype: Contacts Only | |
| Voice:(480)460-2350 Fax: Available upon request | Brass Rat |
| E-mail Icon at http://www.analog-innovations.com | 1962 |
The touchstone of liberalism is intolerance
ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here.
All logos and trade names are the property of their respective owners.