Sang-Gyun Gi

Sang-Gyun Gi

Auto-generated profile, based on publicly available information

Sang-Gyun Gi

School of Electrical Engineering and Computer Science, Gwangju Institute of Science and Technology, Gwangju, South Korea

Research Focus & Metrics

Academic Career Length

9Years (2018-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
15 May 2019 in IEEE Transactions on Electron Devices
Verified
Journal Article
Closed access
IYInjune Yeo
Injune Yeo
MCMyonglae Chu
Myonglae Chu
SGSang-Gyun Gi
Sang-Gyun Gi
HHHyunsang Hwang
Hyunsang Hwang
Byung-Geun LeeByung-Geun Lee
Byung-Geun Lee
In this study, a circuit technique and training algorithm that minimizes the effect of stuck-at-faults (SAFs) within a memristor crossbar array of neural networks (NNs) are presented. To improve the network performance in the presence of SAFs, a conventional transimpedance amplifier, which is used for summing the currents that flow through the memristors, is modified to ensure that the amplifier output is within the appropriate operating range. Further improvement in the network performance is achieved by using the proposed training algorithm, which utilizes the locations and values of faulty memristors for network training. A feedforward NN employing 32 x 32 memristor crossbar arrays is implemented to verify the performance improvement in the NNs using the proposed circuit technique and training algorithm.

27 October 2020 in IEEE Transactions on Industrial Electronics
Verified
Journal Article
Closed access
IYInjune Yeo
Injune Yeo
SGSang-Gyun Gi
Sang-Gyun Gi
GWGunuk Wang
Gunuk Wang
Byung-Geun LeeByung-Geun Lee
Byung-Geun Lee
This article presents a hardware- and power-efficient RRAM-based neural network capable of online learning. The network is modularized in consideration of scalability and consists of 11 modules. Each module comprises two 25 × 25 RRAM crossbar arrays, four analog multiplexers, and one neuron chip. A stochastic neuron that performs both quantization and activation of the column current flowing through the RRAM array is used to perform online learning without using the high-resolution analog-to-digital converters and digital-to-analog converters. The impact of the initialization of the RRAM device on the network performance is analyzed and a fast yet effective method of initialization is also presented. The network consists of three fully connected layers with 100 input and 10 output neurons. The two hidden layers are made up of 50 and 25 neurons, respectively, and a total of 13 × 103 synapses are implemented by combining 21 RRAM arrays. The neuron chip, which includes 32 neurons, is fabricated through a 0.18-μm standard CMOS process. The network achieves a classification accuracy of 93.4% for the reduced Modified National Institute of Standards and Technology database (MNIST) dataset and the energy efficiency of 38.1 teraoperations per second per Watt (TOPS/W), the highest efficiency among recently reported state-of-the-art RRAM-based neural networks.

17 July 2020 in IEEE Transactions on Nanotechnology
Verified
Journal Article
Closed access
WCWooseok Choi
Wooseok Choi
SGSang-Gyun Gi
Sang-Gyun Gi
DLDonguk Lee
Donguk Lee
SLSeokjae Lim
Seokjae Lim
CLChuljun Lee
Chuljun Lee
HHHyunsang Hwang
Hyunsang Hwang
Hardware neural networks (HNNs) which use synapse device (SD) arrays show promise as an approach to energy efficient parallel computation of massive vector-matrix multiplication. To maximize the inference accuracy of pattern recognition HNNs, we propose a highly reliable 2-terminal SD adopting WOx films for resistor materials. We utilize the intrinsic properties of the $WO_x$ film, which exhibits substantial changes in conductivity from $10^{-8}$ to $10^4 \Omega^{-1} \mathrm{cm}^{-1}$ across multiple phases by varying the oxygen vacancy concentration. By optimizing the oxygen stoichiometry and electrode materials, we can form nanoscale $WO_x$ -based SDs with excellent conductance uniformity and I-V linearity. To evaluate the performance of array-based HNN, MATLAB and SPICE simulations are conducted considering parasitic resistance effects in arrays. Our results show a significant improvement in inference accuracy using $WO_x$ -based SD arrays even in advanced node scaling. Through experimental hardware implementation, 16 x 16 pixel images are correctly classified and we show the potential of $WO_x$ -based SD for future large-scale HNN applications.

20 July 2022 in IEEE Transactions on Industrial Electronics
Verified
Journal Article
Closed access
SGSang-Gyun Gi
Sang-Gyun Gi
HLHyunkeun Lee
Hyunkeun Lee
JJJingon Jang
Jingon Jang
Byung-Geun LeeByung-Geun Lee
Byung-Geun Lee
This article presents a resistive random access memory (ReRAM)-based convolutional neural network (CNN) accelerator with a new analog layer normalization (ALN) technique. The proposed ALN can be used to effectively reduce the effect of the conductance variation in ReRAM devices by normalizing the outputs of the vector-matrix multiplication (VMM) in the charge domain. The ALN achieves high energy and hardware efficiencies because it directly processes the normalization of the VMM outputs without storing their values in memory and is merged into the neuron circuit of the accelerator. To verify the effect of the ALN through experiments, a VMM accelerator that consists of two 25 × 25 sized ReRAM arrays and peripheral circuits with ALN is used for a convolution layer with digital signal processing in a field programmable gate array. The MNIST dataset is used to train and inference a CNN employing two VMM accelerators that work as convolution layers in a pipelined manner. Despite the conductance variation of the ReRAM devices, the ALN successfully stabilizes the output distribution of the convolution layer, which improves the classification accuracy of the network. A final classification accuracy for the MNIST and Fashion-MNIST datasets of 96.2% and 83.1% is achieved, respectively, with an energy efficiency of 9.94 tera-operations per second per Watt.

1 August 2018 in IEEE Transactions on Electron Devices
Verified
Journal Article
Closed access
SGSang-Gyun Gi
Sang-Gyun Gi
IYInjune Yeo
Injune Yeo
MCMyonglae Chu
Myonglae Chu
KMKibong Moon
Kibong Moon
HHHyunsang Hwang
Hyunsang Hwang
Byung-Geun LeeByung-Geun Lee
Byung-Geun Lee
This paper presents a new method for modeling the nonideal conductance response (CR) of synaptic devices. Unlike previous studies, which utilize physical device properties for modeling, this paper only uses the measured CR data. This allows the proposed modeling method to be easily applied to various types of synaptic devices without considering the unique physical properties of each device. An efficient piecewise linear approximation method which offers a tradeoff between computational complexity and simulation accuracy of neural networks is also presented to generate a linear device model out of nonlinear CR data. In addition, model parameters, which reflect the nonideal characteristics of the CR such as abrupt and asymmetric conductance changes, conductance variation, and limited conductance dynamic range, are introduced to evaluate the network performances in the presence of the nonidealities. By adjusting the model parameters, the desired CR satisfying the network performance requirements can be derived for device development. A three-layer neural network employing the device model has been designed and trained for the MNIST data set in order to demonstrate the application of the model to system-level simulations and verify the effectiveness of the modeling method.

30 November 2025 in Journal of Sensor Science and Technology
Verified
Journal Article
Open access
SGSang-Gyun Gi
Sang-Gyun Gi
SPSehwan Park
Sehwan Park
JLJunseop Lee
Junseop Lee
SLSeungjoo Lee
Seungjoo Lee
JCJihyuk Cho
Jihyuk Cho