59 Simulating Hardware Neural Networks with Organic Memristors and Organic Field Effect Transistors
-
Published:2010
Download citation file:
Present day Artificial Neural Networks (ANNs) are emulated on serial computing machines, and as a result their ability to execute tasks requiring large computations in real time is severely limited. Here we propose a novel way of realizing a hardware neural network based on a minimal number of electrical elements per neuron. In its most basic form, our artificial neuron consists of a single transistor, one diode, two resistors, and, for a binary connection weight, one memristor per input. Via computer simulation, we show how this device architecture can exhibit simple neuronal behavior with an activation function that approximates a sigmoid function. We demonstrate that it can be trained to carry out a binary classification scheme using a single simulated neuron. We then apply it to a real-world problem and show that a network of eight simulated neurons can perform a pattern recognition task relevant to tethered locomotion. Our analysis is tailored toward the use of organic electronics as a means for hardware implementation.