The integrated storage and computing chip we have developed showcases high adaptability, high energy efficiency, high usability, and high accuracy, effectively enhancing the learning adaptability of intelligent devices in practical application scenarios. "On the 10th, Gao Bin, associate professor of the School of Integrated Circuits at Tsinghua University, said in an interview with reporters, This chip reveals a new paradigm of edge learning in the era of artificial intelligence, providing an innovative development path to break through the energy efficiency and computing power bottlenecks under the traditional computing architecture of von Neumann. Recently, Professor Wu Huaqiang and Associate Professor Gao Bin from the School of Integrated Circuits at Tsinghua University developed the world's first full system integration, based on the integrated computing paradigm of storage and computing A memristor memory computing integrated chip that supports efficient on-chip learning (machine learning can be directly completed on the hardware side). The relevant results are published online in the latest issue of Science. Under the same task, the energy consumption of on-chip learning achieved by this chip is only 3% of that of specialized integrated circuit systems under advanced technology, demonstrating excellent energy efficiency advantages and has the potential to meet the high computing power requirements of the artificial intelligence era. Relevant achievements can be applied to smart terminal devices such as mobile phones, and can also be applied to edge computing scenarios, such as cars, robots, etc. It is reported that currently, international research is still focused on demonstrating the learning function at the level of memristor arrays. The fully integrated memristor on-chip learning chip has not yet been implemented, mainly due to the poor compatibility between the high-precision weight update method required by traditional backpropagation training algorithms and the actual characteristics of memristors, resulting in bottlenecks in system accuracy and energy consumption under large-scale integration. What are the challenges to overcome in breaking through the constraints of traditional memory computing separation architecture on computing power improvement and achieving a low energy consumption and high-precision integrated memory resistor memory computing on-chip learning chip for the entire system? This is a breakthrough that the research team has accumulated for many years. The main difficulties in the research and development process are as follows: firstly, we need to solve the large-scale integration problem of memristors. We not only optimized the device materials and architecture, improved the device characteristics, but also developed large-scale integration processes. Secondly, we need to solve the non ideal characteristics of the underlying hardware at multiple physical scales to improve accuracy, such as device nonlinearity and asymmetry, array parasitism, and circuits Noise, etc; Finally, to achieve efficient hardware systems, collaborative optimization of algorithms architecture circuits and devices is required Yao Peng, a team member and postdoctoral fellow at the School of Integrated Circuits at Tsinghua University, said. (New News Agency)