Fudan University's National Key Laboratory of Integrated Circuits and Systems, along with the Institute of Frontier Technologies in Chips and Systems, has recently released a semiconductor charge storage technology capable of performing 2.5 billion operations per second. This is currently the fastest semiconductor charge storage technology in the world. The research results have been published in the renowned academic journal Nature.
The "Dawning" can perform 2.5 billion operations per second.
The Fudan University research team named this memory "Dawning." It is only one square centimeter in size and looks no different from ordinary memory devices. However, through innovative physical mechanisms, it achieves an erase or write speed of 400 picoseconds, with one picosecond equivalent to one trillionth of a second. "Dawning" operates at a speed that can perform 2.5 billion operations per second, which is one million times faster than traditional flash memory.
Researcher Chunsen Liu from Fudan University's National Key Laboratory of Integrated Circuits and Systems: This speed improvement is highly groundbreaking and completely breaks through the theoretical bottlenecks of existing storage technology frameworks.
According to reports, the team of researchers proposed a completely new approach to speed up the process. This innovation not only achieved a significant breakthrough in speed but also ensured that data would not be lost when power was cut off for this new type of memory.
After commercial application
It is expected to promote the development of artificial intelligence computing
The main person in charge of the research team introduced that this entirely new achievement is not far from being converted into a commercial application. Currently, the team has collaborated with manufacturers for tape-out verification. In the future, this technology is expected to play a huge role in areas such as artificial intelligence computing.
Experts stated that the calculations of large artificial intelligence models mainly depend on GPU chips. The mainstream commercial GPU chips can currently achieve 33.5 trillion floating-point operations per second. However, the write or erase speed of the accompanying memory remains at the microsecond level.
Fudan University's "Dawning" memory is perfectly suited to meet the high-speed computing needs of GPU chips. To ensure that this breakthrough technology can be applied commercially as soon as possible, the Fudan University research team collaborated deeply with production enterprises during the R&D process. It is understood that tape-out verification has already been conducted and initial results have been achieved.
Researcher Chunsen Liu from Fudan University's National Key Laboratory of Integrated Circuits and Systems: We have already been able to achieve small-scale full-function chip tape-outs. Next, we can begin trying to integrate it into existing phones and computers. In this way, our phones and computers will no longer encounter bottlenecks such as lagging and overheating caused by existing storage technologies when deploying local models.
(Reported by CCTV reporter Liang Zhengzheng, Dai Qinfu, Wang Dianjia)
(Source: CCTV News Client)
For more wonderful information, please download the "Jimu News" client from the app market. Unauthorized reproduction is prohibited. Welcome to provide news leads, and rewards will be given upon acceptance. 24-hour hotline: 027-86777777.
Original article: https://www.toutiao.com/article/7494464431951249983/
Disclaimer: The article solely represents the views of the author. Please express your opinions by clicking the "like/dislike" buttons below.