3 ms·
Analog computing makes a lot of sense in the context of genetic algorithms and "deep learning" and I wouldn't be surprised if there's already some ASICs under d
by janekm 9y ago
Analog computing makes a lot of sense in the context of genetic algorithms and "deep learning" and I wouldn't be surprised if there's already some ASICs under design using those principles.
One big challenge is that the design kits from the foundries aren't likely to include all the analog computer cells that would be needed (but perhaps for example a current mirror into a MIM capacitor could make for an integrator?).
- shmageggy 9y agoGenetic algos are inherently discrete, so I can't see analog computing being all that useful for them.
- orbifold 9y agoYou can look into neuromorphic computing, there are several projects that use digital-analog hybrid systems to implement neural networks.
- read_only 9y agoAt the ISCA conference last week, Yoshua Bengio, of DNN fame, talked about their early efforts in analog computing. Here are some related slides of his: http://www.iro.umontreal.ca/~bengioy/talks/Brains+Bits-NIPS2016Workshop.pptx.pdf http://www.iro.umontreal.ca/~bengioy/talks/Brains+Bits-NIPS2...