Is FPGA still the status quo for producing accelerated neural networks? This is for non-linear function approximation?
yes, neural network accelerator implementation on FPGA is possible.
Thanks Mr Raveendran for answering. Are the implementation techniques quite mature?
Ok. I suppose there are ethics with all this. Anyway assuming some positive uses Are possible, which I think there are, then could you give me some understanding of where the field is currently At (tutorials, textbooks, authors). Would be appreciated. Thanks
Hi.. this is a guideline
http://lib.tkk.fi/Diss/2005/isbn9512275279/article8.pdf
This author is one of the most popular in this area.
The True North chip has 1 million neurons, and 256 million synapses, and is the state of the art in silicon based NeuroMorphic chips. Not sure of Availability you have to order it from IBM and it is not yet commercially available.
A basic way is that you can educate neural network in computer with using object ortiented programmes (for example MATLAB). After this one, educated network (cell) values can be loded to MCU or MPU.
Thanks for your answers so far. I have found interest in using FPGA down the track. It seems like it is already a rapidly maturing field... I have a tentative plan for co-simulation with FPGA in the loop. I believe it is much easier to work with FPGA these days...
i'm just interested in feedforward multi-layer perceptron at this stage or recurrent neural networks. the exploration there is large enough and mature enough for what I need to test for the moment
fpga uses standard logic and tends to be expensive power-wise compared to neuromorphic chips. Your mileage may vary, depending on the size of your project.
were neuromorphic chips readily available then? Do you know any vendors or have access to data sheets?
I have seen that the European Commission is looking for partners for it's neuromorphic program as part of the human brain project. You have to apply as an institution, I believe. There are two companies that have made it through the 2nd round at DARPA, IBM is one of them.
- Badges
- Science topic