.png){kind=icon}
.png){kind=icon}
Return
Towards dynamic computing: Multi-dynamic neuronal-synaptic transistors for visual intelligence
.png){kind=icon}
PRE
AI
DOI：10.1016/j.nanoen.2025.111656.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
• Broad application spectrum. The MDNST ingeniously integrates four dynamics of biological visual systems, enabling dynamic neuromorphic computing and unlocking infinite possibilities for achieving machine vision comparable to biological vision. • Simplified architecture. The MDNST utilizes a unified transistor structure to transform the traditional heterogeneous system into a highly homogeneous and compact one, thereby improving the manufacturing efficiency of visual hardware systems. • Outstanding performance: The MDNST outperforms traditional computing both in the IoU and the accuracy. In a semantic segmentation task with multiple categories, the mean IoU increased from 65.0% under bright visual environments and 67.8% under dim visual environments to 74.3%, and the global accuracy rose from 90.5% and 91.6% to 93.5%, respectively.