A neuro-chip with temporal learning: test results for signal/shape generation

We briefly describe a recently designed neural network with temporal learning. The developed learning rule extends the temporal learning to realize a forward instantaneous update scheme, suitable for complete analog hardware implementation. The single chip configuration is a two-dimensional scalable array with learning constructed via simple scalable CMOS building blocks. We report on extensive test results of learning prototype periodic temporal signals which may be displayed as a closed curve or a figure-8 shape in the state space. Several experimental tests demonstrating the instantaneous learning by the chip of these shapes are reported.