Technology seems to have become one of the factors for a country's comprehensive strength. In recent years, the competition in the field of science and technology has been particularly fierce. There are fierce competitions in many new technology fields such as 5G/artificial intelligence technology, AR, and VR. In recent years, my country has also been carrying out research and development of related technologies, striving to make up for technical shortcomings and achieve leadership in some aspects.
With the development of AI technology in the future, people's demand for storage will further increase, which will also bring more challenges to integrated circuit chips. However, due to the influence of Moore's Law, it will become very difficult to increase computing power by scaling the integrated circuit process. In addition, the traditional Von Neumann architecture will also bring new problems to computing and storage, resulting in a large amount of power consumption during the process of moving large amounts of data and additional delays. How to solve the problem of storage integration has also become a problem that needs to be solved now.
Recently, the team of Professors Qian He and Wu Huaqiang from Tsinghua University worked together with their collaborators to solve this problem and developed a memory-computing integrated system with multiple memristor arrays. The energy efficiency of processing convolutional neural areas is two orders of magnitude higher than that of GPUs, which greatly improves the computing power of computing devices. In the future, more complex computing problems can be solved with smaller power consumption and lower hardware costs. It can also increase the recognition accuracy rate on the handwritten digit set to 96.19, which is comparable to software recognition.
At the same time, with the blessing of the memristor array, it is also possible to realize physically scheduled parallel computing, integrate storage and computing, break the "von Neumann bottleneck" limitation on computing power, and meet people's higher demand for artificial intelligence. At present, the relevant research results have been published in "Nature".
Although the rescue team of Tsinghua University has now realized a complete hardware based on the memristor array, it still faces severe challenges. The lack of reliable memristors and the lack of a stable and reliable architecture will affect the research and development system of storage integration, thereby affecting the final computing efficiency.
All in all, it is still difficult to realize the integration of storage and computing. Now the Tsinghua University team has taken a key step to integrate multiple memristor arrays in the system, which greatly improves the performance of convolutional neural network algorithms and realizes image recognition. This also shows that the integrated storage and computing architecture is achievable. As a top university in China, Tsinghua University also has a strong strength in scientific research. It is believed that when the technology matures in the future, it will be able to achieve greater breakthroughs, and make the integrated storage and computing architecture mature as soon as possible, so as to solve people's troubles in the era of intelligent AI.
