Research

- - -

Computers have led to an information revolution and artificial intelligent systems that simulate the learning functions of the human brain. The world’s fastest supercomputer, Summit, may have a computing capacity comparable to that of the human brain. However, Summit consumes the equivalent power of 7000 homes (~15 MW), and the brain only consumes a power of a light bulb (~20 W). Computers execute algorithms on physically separated logic and memory units in digital serial mode, which fundamentally restrains computers from handling “big data” efficiently in complex dynamic environments, and limits the developments of emerging intelligent systems such as self-piloted unmanned aerial vehicles (UAVs). By contrast, the brain simultaneously processes and learns from “big data” via trillions of synapses and neurons in analog parallel mode, and facilitates parallel processing and real-time learning with an energy efficiency more than five orders of magnitude superior to that of the supercomputer.

In this project, we plan to (A) perform research on devices including synaptic resistors (synstors), memory resistors (memristors), and neuristors to emulate the analog short- and long-term memory, convolutional signal processing, and correlative learning functions of synapses, and the nonlinear dynamic functions of neurons. (B) We will develop a synstor and neuristor integrated circuit (SNIC) that operates in analog parallel mode, facilitates processing and real-time learning, is more than six orders of magnitude more efficient than that of the supercomputer, and consumes a power of ~1 mW. (C) We will integrate multiple SNICs with distributed networks of sensors and actuators to demonstrate multifunctional intelligent systems with structural health-monitoring (SHM), automatic navigation, and real-time learning in self-piloted aerial vehicles. (D) We will also establish theoretical models for transforming intelligent functions of the brain to SNICs and intelligent systems.

The multifunctional intelligent system based on SNICs will lead to platforms for machine learning from “big data” with speed, power efficiency, and memory capacity significantly superior to digital computers, and the embedded SNICs permit the establishment of intelligent behaviors in mobile systems in complex and dynamically changing environments. This project could have potential DoD impacts to (1) transform off-line machine learning in supercomputers to real-time learning in embeddable SNIC-based systems; (2) establish intelligent behaviors in SNIC-based sensor systems such as state perceptions, self-awareness, and self-learning in SHM and intelligence, surveillance, and reconnaissance (ISR) systems; (3) facilitate self-driving, autonomous control, and learning functions in unmanned systems such as self-piloted UAV; (4) improve agility, survivability, reliability, responsiveness, and reduce energy consumption, failure, risks in military systems. The project will also train students with cutting-edge technologies to prepare future careers in DoD projects.

Prof. Yong Chen of UCLA will be the PI of the project, and Profs. Lei He of UCLA, Profs. Daniel Inman and Jun Zhang of University of Michigan, Ann Arbor, Prof. Fu-Kuo Chang of Stanford University, Prof. Stanley Williams from Texas A&M University, Prof. Andy Sarles of University of Tennessee, and Prof. Jianhua (Joshua) Yang of University of Massachusetts will be co-PIs of the project. The program will leverage collaborations among researchers from broad disciplines including neuroscience, bioengineering, applied mathematics, computer science, materials science, electrical, mechanical, and aerospace engineering. The program will be guided and advised by AFOSR projector managers and also an advisory board, which consists of members from DoD laboratories including AFRL and Sandia National Laboratories (SNL), and from the aerospace industry including Boeing, GE, United Technologies Aerospace Systems, and Raytheon.