• The Journal of Information Technology
• /
• v.5 no.2
• /
• pp.55-67
• /
• 2002

This paper presents an adaptive back propagation algorithm that update the learning parameter by the generated error, adaptively and varies the number of hidden layer node. This algorithm is expected to escaping from the local minimum and make the best environment for convergence to be change the number of hidden layer node. On the simulation tested this algorithm on two learning pattern. One was exclusive-OR learning and the other was $7{\times}5$ dot alphabetic font learning. In both examples, the probability of becoming trapped in local minimum was reduce. Furthermore, in alphabetic font learning, the neural network enhanced to learning efficient about 41.56%~58.28% for the conventional back propagation. and HNAD(Hidden Node Adding and Deleting) algorithm.

• Journal of the Institute of Electronics Engineers of Korea TE
• /
• v.39 no.2
• /
• pp.66-75
• /
• 2002

This paper presents an adaptive back propagation algorithm that its able to enhancement for the learning efficiency with updating the learning parameter and varies the number of hidden layer node by the generated error, adaptively. This algorithm is expected to escaping from the local minimum and make the best environment for the convergence of the back propagation neural network. On the simulation tested this algorithm on three learning pattern. One was exclusive-OR learning and the another was 3-parity problem and 7${\times}$5 dot alphabetic font learning. In result that the probability of becoming trapped in local minimum was reduce. Furthermore, the neural network enhanced to learning efficient about 17.6%~64.7% for the existed back propagation. 

• Proceedings of the Korean Vacuum Society Conference
• /
• 2015.08a
• /
• pp.211.2-211.2
• /
• 2015

In order to investigate the effect of front channel in DAL (dual active layer) TFT (thin film transistor), we successfully fabricated DAL TFT composed of ITZO and IGZO as active layer using the solution process. In this structure, ITZO and IGZO active layer were used as front and back channel, respectively. The front channel was changed from 0.05 to 0.2 M at fixed 0.3 M IGZO of back channel. When the mol concentration of front channel was increased, the threshold voltage (VTH) was increased from 2.0 to -11.9 V and off current also was increased from 10-12 to 10-11. This phenomenon is due to increasing the carrier concentration by increasing the volume of the front channel. The saturation mobility of DAL TFT with 0.05, 0.1, and 0.2 M ITZO were 0.45, 4.3, and $0.65cm2/V{\cdot}s$. Even though 0.2 M ITZO has higher carrier concentration than 0.05 and 0.1 M ITZO, the 0.1 M ITZO/0.3 M IGZO DAL TFT has the highest saturation mobility. This is due to channel defect such as pores and pin-holes. These defect sites were created during deposition process by solvent evaporation. Due to these defect sites, the 0.1 M ITZO/0.3 M IGZO DAL TFT shows the higher saturation mobility than that of DAL TFT with front channel of 0.2 M ITZO.

• Proceedings of the KIEE Conference
• /
• 2000.07c
• /
• pp.1824-1826
• /
• 2000

Up to date, the dual scanning method has been adopted to decrease address-ing period in AC PDP. In this case, addressing period can be reduced, but the driving circuit cost should be increased. In this study, to increase addressing speed we have studied the relationship between addressing speed and cell structure. That is to say, we varied the thickness of dielectric layer on the front glass, the thickness of white back and the height of barrier rib on the rear glass. So, we found that the addressing time was decreased 4% with decreasing 5um thickness of dielectric layer on the front glass and 2um thickness of white back on the rear glass. Also in case of decreasing the height of barrier rib, addressing time was decreased about 4% per 10um.

• Journal of the Korean Institute of Telematics and Electronics C
• /
• v.35C no.3
• /
• pp.79-89
• /
• 1998

The Error Back-Propagation(EBP) algorithm is widely applied to train a multi-layer perceptron, which is a neural network model frequently used to solve complex problems such as pattern recognition, adaptive control, and global optimization. However, the EBP is basically a gradient descent method, which may get stuck in a local minimum, leading to failure in finding the globally optimal solution. Moreover, a multi-layer perceptron suffers from locking a systematic determination of the network structure appropriate for a given problem. It is usually the case to determine the number of hidden nodes by trial and error. In this paper, we propose a new algorithm to efficiently train a multi-layer perceptron. OUr algorithm uses stochastic perturbation in the weight space to effectively escape from local minima in multi-layer perceptron learning. Stochastic perturbation probabilistically re-initializes weights associated with hidden nodes to escape a local minimum if the probabilistically re-initializes weights associated with hidden nodes to escape a local minimum if the EGP learning gets stuck to it. Addition of new hidden nodes also can be viewed asa special case of stochastic perturbation. Using stochastic perturbation we can solve the local minima problem and the network structure design in a unified way. The results of our experiments with several benchmark test problems including theparity problem, the two-spirals problem, andthe credit-screening data show that our algorithm is very efficient.

• Proceedings of the Korean Nuclear Society Conference
• /
• 2005.10a
• /
• pp.832-833
• /
• 2005

• Proceedings of the Korean Institute of Electrical and Electronic Material Engineers Conference
• /
• 2010.06a
• /
• pp.258-258
• /
• 2010

In this work, back bias effects in the program of the silicon-oxide-nitride-oxide-silicon (SONOS) cell using two-step pulse sequence, are investigated. Two-step pulse sequence is composed of the forward biases for collecting the electrons at the substrate terminal and back bias for injecting the hot electrons into the nitride layer. With an aid of the back bias for electron injection, we obtain a program time as short as 600 ns and an ultra low-voltage operation with a substrate voltage of -3 V.

• Proceedings of the Korean Vacuum Society Conference
• /
• 2011.02a
• /
• pp.394-394
• /
• 2011

최근 태양전지의 후면에서 통상적으로 사용되는 Al을 이용한 후면의 BSF형성과 그에 관한 연구보다 계면의 recombination을 줄이기 위하여 passivation 특성이 좋은 층을 후면에 형성하고 국부적으로 BSF를 형성하는 back contact을 형성하여 특성을 향상시키는 연구가 많이 이루어지고 있다. 본 연구는 이러한 local back contact을 boron-BSF를 이용하여 형성하고 passivation layer는 oxide를 이용한 구조를 SILVACO 2-dimension simulation을 이용하여 그 특성을 분석하였다. Boron-local back contact 구조에서 boron-BSF의 doping concentration, depth, lateral width, boron-BSF spacing 가변을 통해 태양전지의 특성변화에 대해서 spectrum response를 통한 QE 분석 및 I-V를 분석하여 최적화하였다.

• Proceedings of the Korean Society of Precision Engineering Conference
• /
• 1995.10a
• /
• pp.415-418
• /
• 1995

This paper describes a new technique for thickness measurement of a very thin layer less than one-quarter of the wavelength of ultrasonic wave using ultrasonic pulse-echo method. The technique determines the thickness of a thin layer in a layered medium form the amplitudes of the total reflected waves from the back side layer of interst. Thickness of a very thin layer few inch deep inside the media can be measured without using a very high frequency ultrasonic transducer over 100MHz which must be used in the conventional techniques for the precision measurement of a thin layer. The method also requires no inversion process to extract the thickness from the waveform of the reflected waves, so that it makes possible on-line measurement of the thickness of the layer.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.7 no.5
• /
• pp.1044-1051
• /
• 2003

This paper surveys the EBP(Error Back Propagation) learning, the Cross Entropy function and the LBL(Layer By Layer) learning, which are used for learning the MLP(Multi Layer Perceptrons). We compare the merits and demerits of each learning method in the handwritten digit recognition. Although the speed of EBP learning is slower than other learning methods in the initial learning process, its generalization capability is better. Also, the speed of Cross Entropy function that makes up for the weak points of EBP learning is faster than that of EBP learning. But its generalization capability is worse because the error signal of the output layer trains the target vector linearly. The speed of LBL learning is the fastest speed among the other learning methods in the initial learning process. However, it can't train for more after a certain time, it has the lowest generalization capability. Therefore, this paper proposes the standard of selecting the learning method when we apply the MLP.