| Project/Area Number |
19K21083
|
|---|
| Project/Area Number (Other) |
18H05911 (2018)
|
|---|
| Research Category |
Grant-in-Aid for Research Activity Start-up
|
| Allocation Type | Multi-year Fund (2019)
Single-year Grants (2018) |
|---|
| Review Section |
0302:Electrical and electronic engineering and related fields
|
|---|
| Research Institution | National Institute of Advanced Industrial Science and Technology |
|---|
Principal Investigator |
Stoliar Pablo 国立研究開発法人産業技術総合研究所, エレクトロニクス・製造領域, 主任研究員 (40824545)
|
|---|
| Project Period (FY) |
2018-08-24 – 2021-03-31
|
| Project Status |
Completed (Fiscal Year 2020)
|
| Budget Amount *help |
¥2,990,000 (Direct Cost: ¥2,300,000、Indirect Cost: ¥690,000)
Fiscal Year 2019: ¥1,430,000 (Direct Cost: ¥1,100,000、Indirect Cost: ¥330,000)
Fiscal Year 2018: ¥1,560,000 (Direct Cost: ¥1,200,000、Indirect Cost: ¥360,000)
|
|---|
| Keywords | Artificial neuron / Neuromorphic system / Mott materials / Jeffress model / Neuromorphic systems / Emerging devices / Recurrent neural net. / Homeostasis mechanisms / Ferroelec. tunnel junct. |
|---|
| Outline of Research at the Start |
AI using neuromorphic circuits is expected to be advantageous in size, speed and energy. Still, human events are very slow-paced for standard electronics. That is not a problem for conventional computers, but neuromorphic systems work differently; they cannot go idle or do multitasking. The solutions using standard silicon technology are very bulky or power hungry. Here, neuromorphic building blocks, architectures and learning methods using resistive switching devices are developed. Overall, this project contributes to the general understanding of how to do AI using emerging devices.
|
| Outline of Final Research Achievements |
The results can be divided into 4 groups: (1) Study of the advantage of introducing innovative devices into neuromorphic systems. I demonstrated the use of ferroelectric devices to control the power dissipation during learning and to reduce the size of circuits operating at human timescales. I also demonstrated the use of Mott devices (a kind of semiconductors materials) to extend the operational range of artificial neurons. (2) Development of working neuromorphic systems based. In particular, I implemented several bioinspired functionalities, and a system that mimics the way humans and animals detects sound directionality. (3) I studied recurrent-neural neurons stability during learning. In particular, I studied the stability limits, connectivity requirements and learning time of a system to learn arbitrary sequences. (4) I develop and built instrumentation to support the project.
|
| Academic Significance and Societal Importance of the Research Achievements |
AI using neuromorphic circuits is expected to be advantageous in size, speed, and energy. Nevertheless neuromorphic systems operating at human timecales are bulky and ineficient. Overall, this project studies how emerging devices can be more efficient roadmap for AI-human interaction.
|
