Breaking News
News & Analysis

IBM Puts Brain On-a-Chip

8/7/2014 06:55 PM EDT
19 comments
NO RATINGS
3 saves
< Previous Page 2 / 2
More Related Links
View Comments: Newest First | Oldest First | Threaded View
Page 1 / 2   >   >>
R_Colin_Johnson
User Rank
Author
Videos on IBM-Brain-on-a-chip in Actiion
R_Colin_Johnson   8/11/2014 3:55:52 PM
NO RATINGS
Here's the links to a couple of videos on how its IBM's brain-chip use spikes to communicate plus how they can work together to perform pattern recognition problems (with "Background" by IBM).
 
How Data Flows:  http://youtu.be/5izS3lAZHmI
Background: IBM's brain-inspired architecture consists of a network of neurosynaptic cores. Cores are distributed and operate in parallel. Cores operate—without a clock—in an event-driven fashion. Cores integrate memory, computation, and communication. Individual cores can fail and yet, like the brain, the architecture can still function. Cores on the same chip communicate with one another via an on-chip event-driven network. Chips communicate via an inter-chip interface leading to seamless scalability like the cortex, enabling creation of scalable neuromorphic systems. 
 
Public Safety Demo:  http://youtu.be/pYJgqd3g6Ks
Background: A video camera on Hoover Tower at Stanford University is looking down at the plaza, below. A simulated network of IBM TrueNorth chips takes in the video data and locates interesting objects. Objects might look interesting to the system because they are moving or have a different color or texture than the background. The system then further processes those portions of the interesting video to determine what the objects are. It is trained in several specific categories, such as buses, cars, people, and cyclists. In a monitoring application, the camera would only need to communicate when it found an interesting object, rather than continually streaming video to a central location. 


AZskibum
User Rank
Author
Almost asynchronous design
AZskibum   8/11/2014 9:53:21 AM
NO RATINGS
One of the more fascinating things about this chip is that it is almost a purely asynchronous design. Notice that the only clock used is 1 kHz to "discretize neuron dynamics."

prabhakar_deosthali
User Rank
Author
Re: Brilliant, but...
prabhakar_deosthali   8/11/2014 3:21:15 AM
NO RATINGS
I am also of the opinion that suc h a complex chip will require much more advanced programming language and development environment for the application engineers to be able to develop working products .

I would not be surprised that in a current scenario , a supercomputer will be erquired to program such a complex chip!

mithrandir
User Rank
Author
Brilliant, but...
mithrandir   8/11/2014 12:16:59 AM
NO RATINGS
I really love the concept. But what I'm really looking forward to is seeing what sort of real life applications this chip can do. The programming language and environment is also something which can be pretty complex for more the more 'hands-on' engineer.

I dabbled in Neural networks a bit before and experimented by replacing traditional control loops with the neural ones but the results are pretty underwhelming. Sure the neural networks were better but not by much and the sheer amount of data sets I needed plus the effort to train it offline wasn't all that appealing to me. 

Perhaps the developing environment can take a bit of sting off the process. Anyone else use neural networks for more complex stuff?

 

R_Colin_Johnson
User Rank
Author
Re: The competition
R_Colin_Johnson   8/8/2014 5:52:58 PM
NO RATINGS
LeCun makes some good points but he also says "would be capable" which indicates to me that his design is not a reality yet.

lonecowboy
User Rank
Author
Brain On-a-Chip!
lonecowboy   8/8/2014 2:07:56 PM
NO RATINGS
More information here regarding the people involved with this project :-> http://vlsi.cornell.edu/~ni49/

My former Conexant colleague Professor Alyosha Molnar is one of them!

alex_m1
User Rank
Author
The competition
alex_m1   8/8/2014 1:56:42 PM
NO RATINGS
Yann lecun, another AI expert , is working on another chip, claim his chip could do much better , 3000 gsops/watt.

https://www.facebook.com/yann.lecun/posts/10152184295832143

Intellijoule
User Rank
Author
IBM Politicans
Intellijoule   8/8/2014 12:35:30 PM
NO RATINGS
The comment about chip versus the politicians is not correct!  You must be talking about USA politicans who have no brains at all. Vote for Netanyahu or Hero Putin if you want brains and action!  IBM and Apple will change the way computing occurs at the personal level. Great going IBM!! 

Intellijoule
User Rank
Author
IBM Brain chip
Intellijoule   8/8/2014 12:19:39 PM
NO RATINGS
IBM, Politicians,

Intellijoule
User Rank
Author
IBM Brain chip
Intellijoule   8/8/2014 12:19:24 PM
NO RATINGS
IBM, Politicians,

Page 1 / 2   >   >>
Most Recent Comments
michigan0
 
SteveHarris0
 
realjjj
 
SteveHarris0
 
SteveHarris0
 
VicVat
 
Les_Slater
 
SSDWEM
 
witeken
Most Recent Messages
9/25/2016
4:48:30 PM
michigan0 Sang Kim First, 28nm bulk is in volume manufacturing for several years by the major semiconductor companies but not 28nm FDSOI today yet. Why not? Simply because unlike 28nm bulk the LDD(Lightly Doped Drain) to minimize hot carrier generation can't be implemented in 28nm FDSOI. Furthermore, hot carrier reliability becomes worse with scaling, That is the major reason why 28nm FDSOI is not manufacturable today and will not be. Second, how can you suppress the leakage currents from such ultra short 7nm due to the short channel effects? How thin SOI thickness is required to prevent punch-through of un-dopped 7nm FDSOI? Possibly less than 4nm. Depositing such an ultra thin film less then 4nm filum uniformly and reliably over 12" wafers at the manufacturing line is extremely difficult or not even manufacturable. If not manufacturable, the 7nm FDSOI debate is over!Third, what happens when hot carriers are generated near the drain at normal operation of 7nm FDSOI? Electrons go to the positively biased drain with no harm but where the holes to go? The holes can't go to the substrate because of the thin BOX layer. Some holes may become trapped at the BOX layer causing Vt shift. However, the vast majority of holes drift through the the un-dopped SOI channel toward the N+Source,...

Datasheets.com Parts Search

185 million searchable parts
(please enter a part number or hit search to begin)
Like Us on Facebook
EE Times on Twitter
EE Times Twitter Feed