Design Con 2015
Breaking News
News & Analysis

Memristor emulates neural learning

4/15/2010 03:00 PM EDT
4 comments
NO RATINGS
More Related Links
View Comments: Newest First | Oldest First | Threaded View
SOY
User Rank
Rookie
re: Memristor emulates neural learning
SOY   4/17/2010 9:52:37 AM
NO RATINGS
Johnson-san, Hi, thank you for your valuable articles. I think this is one of reconfigurable computing approaches (or simply FPGA based computings, and or Adaptive Computing from DARPA). And curbon nano-wire based PLD (nanoPLD) researched by Prof. Andre DeHon and the memoristor based computing researched by hp have common scense or face to common direction. Difference between them is probably how many states held by memory element (cross-point, nano-wire may be only one bit, but memoristor may hold multi-bit). Xbar network is the key that behaves both of the memory and interconnect. I request you to report not only the structure (or architecture) but also application and HOW TO PROGRAM at initial state or HOW TO CONFIGURE at initial state. Currently FPGA support JTAG based programming, but it is no scense. And I think this is solution for application-specific computing that means application is statically mapped to the device and does not change to other application (does not make context-switch). Of course, forcing the switching is possible, but it breaks data flow on device, this means a cost involving extra time and space if switching is taken.

R_Colin_Johnson
User Rank
Blogger
re: Memristor emulates neural learning
R_Colin_Johnson   4/16/2010 3:37:24 PM
NO RATINGS
Memristors are just a part of the solution. IBM, Hewlett Packard, HRL and their university partners are all looking at these other issues too. Look for a flurry of results over the next 6 to 18 months.

SOY
User Rank
Rookie
re: Memristor emulates neural learning
SOY   4/16/2010 11:53:31 AM
NO RATINGS
>Memristors do not provide a solution for this problem. As mentioned in the article, crossbar behaves as interconnection network, and its crosspoint is the synapse-like element.

Mapou
User Rank
Rookie
re: Memristor emulates neural learning
Mapou   4/16/2010 4:02:20 AM
NO RATINGS
Interesting. However, even though simulating a synapse or a neuron is nice but what is even more important is the ability to physically link those neurons to hundreds or thousands of other neurons in the network. Oftentimes, the neurons can be very distant. Memristors do not provide a solution for this problem. * Also, synaptic learning in the sensory cortex is not based on the concurrency of the spikes but on afferent signals arriving about 10 milliseconds before the post-synaptic action potential. See the work of Henry Markram for more on this.

Flash Poll
Like Us on Facebook

Datasheets.com Parts Search

185 million searchable parts
(please enter a part number or hit search to begin)
EE Times on Twitter
EE Times Twitter Feed
Top Comments of the Week