In this paper, we propose a learning rule based on a back-propagation (BP) algorithm that can be applied to a hardware-based deep neural network (HW-DNN) using electronic devices that exhibit discrete and limited conductance characteristics. This adaptive learning rule, which enables forward, backward propagation, as well as weight updates in hardware, is helpful during the implementation of power-efficient and high-speed deep neural networks. In simulations using a three-layer perceptron network, we evaluate the learning performance according to various conductance responses of electronic synapse devices and weight-updating methods. It is shown that the learning accuracy is comparable to that obtained when using a software-based BP algorithm when the electronic synapse device has a linear conductance response with a high dynamic range. Furthermore, the proposed unidirectional weight-updating method is suitable for electronic synapse devices which have nonlinear and finite conductance responses. Because this weight-updating method can compensate the demerit of asymmetric weight updates, we can obtain better accuracy compared to other methods. This adaptive learning rule, which can be applied to full hardware implementation, can also compensate the degradation of learning accuracy due to the probable device-to-device variation in an actual electronic synapse device.
An inference system using gated Schottky diode (GSD) is proposed for highly reliable hardware-based neural networks (HNNs). We explain the characteristics of the GSD and present circuits that take into account the characteristics of the device. The reverse current of the GSD, which is the synaptic current, is saturated with respect to input voltage, which results in immunity of input and output noise and overcoming the IR drop problem in metal wire. In order to take advantages of this saturated I-V characteristics, pulse-width modulation (PWM) of input data instead of amplitude modulation is proposed. In addition, by applying identical pulses to the bottom gate, the synaptic current of the GSD increases linearly, which makes it easy to transfer the calculated weights to the conductance of GSDs. By considering these characteristics, electronic circuits for PWM, current sum, and activation function are designed. Through SPICE simulation, we evaluate the inference accuracy of a 2-layer neural network. The classification accuracy rate of 100 images of MNIST test sets is 94%, and it is comparable to the reference accuracy obtained with software.INDEX TERMS Neuromorphic, synaptic device, gated Schottky diodes, hardware-based neural networks, inference system.
The operation principle and near-linear potentiation mechanism of reconfigurable gated Schottky diodes (GSDs) are analyzed using calibrated device simulation. The reconfigurable GSD has two bottom gates and SiO 2 /Si 3 N 4 /SiO 2 gate insulator stack. According to the polarity of the bottom gate bias, electrons, or holes are induced in the poly-Si active layer and the type of Schottky diodes is reconfigured. In the same manner, the reverse-biased current of GSD is modulated by applying bottom gate bias or storing charge in the Si 3 N 4 charge storage layer. The reverse-biased current of GSD is exponentially proportional to the charge stored in the Si 3 N 4 layer. By representing the amount of stored charge as a logarithmic relation to the number of potentiation pulses, the number of potentiation pulses, and the current of GSD has a power relation. It has been demonstrated that the GSD current exhibits near-linear potentiation characteristics when the exponent of the power relation is close to 1.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.