2005 Fiscal Year Final Research Report Summary
STUDY ON SELF ORGANIZATION NEWRAL NETWORK USING MICRO MAGNETIC SENSOR ARRAYS FOR VARIABLE-SYNAPSE COUPLING DEVICE
| Project/Area Number |
15560294
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
Electron device/Electronic equipment
|
|---|
| Research Institution | Nagoya University |
|---|
Principal Investigator |
UCHIYAMA Tsuyoshi Nagoya University, Faculty of Engineering, Associate Professor, 工学研究科, 助教授 (00203555)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
SUZUKI Tatsuya Nagoya University, Faculty of Engineering, Associate Professor, 工学研究科, 助教授 (50235967)
|
|---|
| Project Period (FY) |
2003 – 2005
|
| Keywords | Variable-Synapse Coupling / Neural Network / Micro Magnetic Sensor / MI element / Multiple and Summation Operation / Delta-Sigma Operation / Profanity of firing of neural / High Order Brain function |
|---|
| Research Abstract |
The key functional devices which plays important role in the neural network, were variable- synapse couplings. However, it would be lead to both much increase of capacity of memory and complexity of the system, when we use a conventional digital system. In this study, to avoid this problem and to create a new type chip neural network device, we propose a new quick response variable-synapse coupling device based on micro-magnetic sensor arrays.
The variable-synapse coupling device is extension device of a coil-pickup MI sensor device. The device operation for the synapse coupling was investigated. We obtained such result that operation speed of neural network would be much increase, when we use conventional C-MOS digital circuit in combination with the proposed synapse couplings.
We also newly propose magnetic feedback neural network circuit. It was shown that the pulse neural network based on the magnetic feedback circuit works as a delta-sigma modulation function. It is concluded that the system is useful for conversion of analogue signal to digital signal in addition to high speed neural network operation.
In summary, the proposed neural network is adapted for high precious signal processing which use not only probability of firing of neuron but also time correlation between firing. In another word, the high order brain function may be achieved, when we use pulse function neural network based on micro magnetic sensor arrays as variable-synapse couplings.
|
