Memristor Research Makes Artificial Brain a Closer Reality

The Memristor or RRAM (resistive random-access memory), is a device with a function similar to a synapse in the human brain; it is also known as an electronic synapse. The smart chip has real-time learning abilities and can handle tasks that machines were previously unable to perform.

“If the chip developed by memristor technology is used in a cell phone, the power consumption of a chip will be drastically reduced,” said Qian He, a professor at the Institute of Microelectronics at Tsinghua University, explaining changes that the memristor technology can bring to ordinary people’s lives. “A cell phone will only need to be charged every two days.”

As a key project in the field of nanometer science and technology supported by the National Key Research and Development Program, Qian He led the research into three-dimensional integration of nanometer memory to provide China’s memory chip industry with independent intellectual property and prototype technology. His work also supports and contribute toward the rapid development of the memory chip industry in China.

In addition, there is another unique function of electronic synapses. It can both be used for data storage and data processing. This integration of computer architecture can be described as disruptive technology and the latest research has found that memristors are particularly suitable for the integration of both functions, making it the focus of research and development of many scholars internationally.

Qian He added that besides memory, memristors in the field of neuromorphic computing chip signal an even more important potential. Neuromorphic computing is a new way of calculating computational power and energy efficiency by mimicking the structure of the human brain.

The difference between neuromorphic computing and traditional computational methods is that neuromorphic computing fuses together the units responsible for data storage and processing, eliminating the significant energy overhead caused by the frequent movement of data between the memory and the central processor, directing the data to be stored and processed within the same unit.

The memristor is suitable for neuromorphic calculation, and the memristor itself can be used to store the value of the calculation. External voltage can be applied and the memristor can complete the function of multiplication and the electric current of multiple memristors can be added together to achieve the calculation Function “By combining multiplication and addition, memristors can accomplish most of the computational tasks in a fraction of the time,” Qian said.

In the first half of 2017, a team led by Qian He completed the world’s first face recognition system that integrates thousands of bidirectional continuous resistive memristor units into the real-time response and online learning like human brain. The energy consumption per iteration of the system is less than one-thousandth of the Intel Xeon Phi processor. Relevant results have been published in the May issue of Nature Communications.

“It is foreseeable that once the memristor-based neuromorphic computing chip technology is mature, making a ‘super artificial brain’ that resembles or even exceeds human brain intelligence and energy efficiency will become a reality,” Qian said.



    Institute of Microelectronics



    Spotlight Group



    Alexis See Tho



    Ren Zuoli



    Lu Xiaobing, Cheng Xi, Zhang Geming, Zhang Li



    Cheng Xi



    Zhao Shujing