Shulou(Shulou.com)11/24 Report--

CTOnews.com October 10 news, Recently, Tsinghua University School of Integrated Circuits Professor Wu Huaqiang, Associate Professor Gao Bin based on the storage and calculation of the integrated computing paradigm, In support of on-chip learning memristor storage and calculation integrated chip field has made a major breakthrough, Research results published in Science.

According to Tsinghua University, Memristor is the fourth basic circuit element after resistance, capacitance and inductance. It can still "remember" the charge that passed through after power failure, and is regarded as a new type of nanoelectronic synaptic device.

In 2012, Qian He and Wu Huaqiang began to study the use of memristors for storage. However, due to the immature material device optimization and integration process of memristors, the team could only rely on their own exploration in the laboratory to explore and improve the consistency and yield of devices in failed experiments again and again.

Two years later, Tsinghua University cooperated with Institute of Microelectronics of Chinese Academy of Sciences, Peking University and other units to optimize the device technology of memristor and prepare high-performance memristor array, which became an important foundation for China to take the lead in realizing large-scale integration of memristor array.

▲ Schematic diagram of cooperative work of multiple memristor array chips, source: Tsinghua University official micro-blog. The same as 2020, Qian He and Wu Huaqiang team built a complete storage and calculation system composed of all hardware based on multi-array memristors. Convolutional neural network algorithm was efficiently run on this system, successfully verifying the image recognition function, which was two orders of magnitude higher than the energy efficiency of graphics processor chips, and greatly improved the computing power of computing equipment. Complex calculations are accomplished with less power consumption and lower hardware costs.

▲ Storage and computing integrated system architecture Tsinghua University said that the storage and computing integrated architecture, just like the new working mode of "working from home," completely eliminates the energy consumption of commuting, avoids the time delay caused by commuting, and greatly saves the operating cost of office space. It has wide application prospects in edge computing and cloud computing.

Qian He and Wu Huaqiang led the team to innovate and design a new general algorithm and architecture (STELLAR) suitable for efficient on-chip learning of memristor storage and calculation, and developed the world's first system-wide integrated memristor storage and calculation integrated chip supporting efficient on-chip learning.

▲ Under the same task of memristor storage and calculation integrated learning chip and test system, the energy consumption of this chip to realize on-chip learning is only 1/35 of that of ASIC system under advanced technology, and it is expected to achieve 75 times energy efficiency improvement.

▲ General algorithm and architecture for efficient on-chip learning based on memristor storage and calculation "On-chip learning can effectively protect user privacy and data while achieving lower latency and lower energy consumption. "According to Yao Peng, the chip can realize fast" on-chip training "and" on-chip recognition "of different tasks with reference to bionic brain processing mode, effectively complete incremental learning tasks under edge computing scenes, adapt to new scenes with extremely low power consumption and learn new knowledge to meet the personalized needs of users.

CTOnews.com with links to related papers:

https://www.science.org/doi/full/10.1126/science.ade3483

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.