Summary:
UCLA researchers from the Department of Mechanical and Aerospace Engineering have developed a novel circuit architecture that emulates neural synapses for concurrent parallel computing.
Background:
Almost all modern computer chips consist of computing and learning processes that are implemented sequentially. To improve computing power, new types of circuits modeled after human neural synapses, neuromorphic devices, have been developed. Neuromorphic platforms are particularly well-suited for machine learning applications, like training sophisticated neural networks for a wide variety of applications. However, the processing power of these devices are limited because the writing voltage signals for learning and reading voltage signals for signal processing cannot occur at the same time. Because of this, it is not possible to fully emulate a human synapse to enable parallel signal processing and learning with currently available hardware. There is a need to develop new circuit hardware and materials to build effective neuromorphic devices to drive machine learning applications.
Innovation:
Professor Yong Chen and colleagues in UCLA’s Department of Mechanical and Aerospace engineering have developed a synaptic resistor (synstor) that can perform parallel signal processing and learning for neuromorphic device development. The synstor is composed of three transistors and a memory capacitor, which can be easily fabricated in standard semiconductor foundries. These synstors are highly versatile and can be arranged in different geometries, depending on the application. They have shown that synstor circuits can concurrently perform signal processing and learning in real time to fly drones in changing environments faster than both human controllers and pre-trained neural networks. In addition, the synstor circuits are nine orders of magnitude more energy efficient than standard computing architectures. Techniques for manufacturing the technology in semiconductor chip foundries has been developed. In summary, this new circuit platform has significant potential to accelerate adoption of cutting-edge machine learning software with high energy efficiency.
Potential Applications:
• Artificial intelligence software
• Autonomous vehicles
• Robotics
• Computer vision/surveillance
• Environmental monitoring
• Energy management
Advantages:
• Compatible with present semiconductor manufacturing
• High energy efficiency
• Parallel signal processing and learning
• Specifically suited for machine learning software
Development-To-Date:
Researchers have designed, built, and tested chips composed of synstors to control drones, demonstrating improved computational power and energy efficiency.
Related Papers:
• Gao, D., Shenoy, R., Yi, S., Lee, J., Xu, M., Rong, Z., Deo, A., Nathan, D., Zheng, J., Williams, R. S., and Chen, Y. Synaptic Resistor Circuits Based on Al Oxide and Ti Silicide for Concurrent Learning and Signal Processing in Artificial Intelligence Systems. Advanced Materials, 2023, 35, 221048.
Reference:
UCLA Case No. 2023-092
Lead Inventor:
Professor Yong Chen, UCLA Departments of Mechanical and Aerospace Engineering and Materials Science and Engineering