Apoptotic self-organized electronic device using thin-film transistors for artificial neural networks with unsupervised learning functions

Abstract

Artificial neural networks are promising systems for information processing with many advantages, such as self-teaching and parallel distributed computing. However, conventional networks consist of extremely intricate circuits to guarantee accurate behaviors of the neurons and synapses. We demonstrate an apoptotic self-organized electronic device using thin-film transistors for artificial neural networks with unsupervised learning functions. First, we formed a "neuron" from only eight transistors and reduced a "synapse" to only one transistor by employing the characteristic degradations of the synapse transistors to adjust the synaptic connection strength. Second, we classified the synapses into two types, "concordant" and "discordant" synapses, and composed a local interconnective network optimized for integrated electronic circuits. Finally, we confirmed that the device is feasible and can learn multiple logical operations, including AND, OR, and XOR.

(Invited) Neuromorphic Application of Oxide Semiconductors

“…In any case, this characteristic is available to modified Hebbian rule, a learning rule we proposed for our network, which is explained in detail elsewhere. 6…”

“…Therefore, we are investigating "brain-type integrated system", namely, neural network built only by hardware, which can be compact, low power, robust, and integrated on everything in future. [5][6][7][8][9][10][11] In order to realize such system, simplification of the processing elements, such as neuron elements and synapse elements, three-dimensional structure, and low cost fabrication are required.…”