An Artificial Neural Network at Device Level Using Simplified Architecture and Thin-Film Transistors

Kasakawa, Tomohiro; Tabata, Hitoshi; Onodera, Ryo; Kojima, Hirotsugu; Kimura, Mutsumi; Hara, Hiroyuki; Inoue, Satoshi

doi:10.1109/ted.2010.2056991

Cited by 29 publications

(13 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This learning method is different from Hebb's learning rule [42], which raises the connection strength of synapse when the neighboring neurons are both in firing states. In the literature [43], this learning method is called the modified Hebb's learning rule, where it was confirmed that AND, OR, and XOR circuits are remembered based on this learning rule using poly-Si TFTs. We use the oxide semiconductor a-IGZO as synapses, which is a device with a characteristic that the conductance deteriorates when current flows [38].…”

Section: Oxide Semiconductor Synapsementioning

confidence: 99%

“…Therefore, researches to realize neuromorphic computing built on actual hardware are actively being carried out [15,16]. It is possible to achieve high performance, large scale integration, low power consumption, etc., using the neuromorphic computing because the biological brain can achieve them [17][18][19]. In such a system, it is necessary to simplify the configuration of the neuron and the synapse and make it inexpensive.…”

Section: Non-neumann-type Computingmentioning

confidence: 99%

“…In Hebbian learning rules, the synapse connection strength increases when the states of both neurons connected to the synapse are the firing states and maintains or decreases otherwise. On the other hand, in the modified Hebbian learning rule [43], the synapse connecting strength is expressed by the relationship between those of the concordant synapse and discordant synapse. The synaptic connection strength changes depending on the state of adjacent neurons, which is similar to Hebbian learning rules.…”

Section: Learning Principlementioning

confidence: 99%

See 2 more Smart Citations

Influence of characteristic variation of oxide semiconductor and comparison of the activation function in neuromorphic hardware

Ikeda

Yamane

Takishita

et al. 2020

NOLTA

Self Cite

View full text Add to dashboard Cite

As the amount of data that people handle increases, the conventional Neumann-type computer architecture is reaching its limits. Therefore, research on hardware implementation of machine learning systems is being actively conducted. In this paper, we have implemented and evaluated neuromorphic hardware that realizes human brain neurons and synapses using oxide semiconductor of amorphous In-Ga-Zn-O (a-IGZO) and a cellular neural network. It was confirmed how variations of initial resistance and deterioration rate of the oxide semiconductor affect operation accuracy of the neuromorphic hardware. Furthermore, we clarified that an activation function suitable for the hardware implementation is a ReLU function.

show abstract

Section: Oxide Semiconductor Synapsementioning

confidence: 99%

Section: Non-neumann-type Computingmentioning

confidence: 99%

Section: Learning Principlementioning

confidence: 99%

See 1 more Smart Citation

Influence of characteristic variation of oxide semiconductor and comparison of the activation function in neuromorphic hardware

Ikeda

Yamane

Takishita

et al. 2020

NOLTA

Self Cite

View full text Add to dashboard Cite

show abstract

“…Consequently, an astronomical large number of transistors may be integrated, however, yield rates will be low, which are similar to human brains. Therefore, we are proposing artificial neural networks using TFTs [4].…”

Section: Introductionmentioning

confidence: 99%

33.2: Neuron MOS Devices using Thin‐Film Transistors

Kimura

Shimada

Matsuda

2015

Symp Digest of Tech Papers

Self Cite

View full text Add to dashboard Cite

show abstract

“…Here, we demonstrate an artificial neural network using poly-Si TFTs [3]. It may be possible to integrate a largescale artificial neural network comparable to the human brain.…”

Section: Introductionmentioning

confidence: 99%

Artificial neural network using poly-Si TFTs - verification of multiple overwriting -

Fujita

Miyatani

Kasakawa

et al. 2011

2011 International Meeting for Future of Electron Devices

Self Cite

View full text Add to dashboard Cite

Artificial neural networks are promising systems for information processing that have many advantages. Here, we demonstrate an artificial neural network using poly-Si TFTs. It may be possible to integrate a large-scale artificial neural network comparable to the human brain. Particularly in this presentation, we verify that the artificial neural network can be re-organized by multiple overwriting.

show abstract

An Artificial Neural Network at Device Level Using Simplified Architecture and Thin-Film Transistors

Cited by 29 publications

References 15 publications

Influence of characteristic variation of oxide semiconductor and comparison of the activation function in neuromorphic hardware

Influence of characteristic variation of oxide semiconductor and comparison of the activation function in neuromorphic hardware

33.2: Neuron MOS Devices using Thin‐Film Transistors

Artificial neural network using poly-Si TFTs - verification of multiple overwriting -

Contact Info

Product

Resources

About