According to the Nikkan Kogyo Shimbun, the team of Associate Professor Masaharu Kobayashi from the Institute of Production Technology at the University of Tokyo has successfully developed a new artificial intelligence chip.
For a long time, deep learning systems have been composed of multi-layer neural networks and learned through large amounts of data. However, since the efficiency of deep learning is limited by the ability to transmit data between the processor and the memory, people have been looking forward to the development of In-Memory Computing (In-Memory Computing) memory hardware. However, the two-dimensional structure of the memory array has shortcomings in terms of computing speed and power consumption, making the efficiency of parallel computing unable to improve.
The research team integrated ultra-thin indium gallium zinc oxide semiconductor (IGZO) transistors and resistance-switching nonvolatile memory in three dimensions, and successfully formed a multi-layered nerve that can complete the learning function and simulate the brain structure on a chip The internet. The temperature requirements of its manufacturing process are the same as those of ordinary integrated circuits. This kind of chip can complete deep learning calculations with extremely high efficiency, not only in cloud space, but also advanced AI calculations on terminals such as mobile phones.
The results were published in the International Conference on Semiconductor Technology and Circuits "VLSI Technology Symposium".
Household Electric Screwdriver
Household Electric Screwdriver,Mini Power Screwdriver,Mini Tool Kit,Screw Driver Power Tool
SUZHOU CREATION SPACE INTELLIGENT TECHNOLOGY CO.,LTD , https://www.pkanglegrinder.com