BEIJING, Feb. 28 (Xinhua) -- Chinese researchers have developed a type of new computing system with high performance and accurate image recognition ability, according to a recent research article published in the journal Nature.
Memristors are leading candidates for future data storage and neuromorphic computing, with the advantages of high speed and small size.
Convolutional neural network, an important model for image recognition, has not yet been fully hardware-implemented using memristor crossbars.
The researchers at Tsinghua University reported the fabrication of high-yield, high-performance and uniform memristor crossbar arrays for the implementation of the convolutional neural networks, which integrate eight 2,048-cell memristor arrays to improve computing efficiency, said the article titled "Fully hardware-implemented memristor convolutional neural network."
Moreover, they proposed an effective training method to adapt to device imperfections and improve the overall system performance. A five-layer memristor-based convolutional neural network was built to achieve a high accuracy of more than 96 percent in handwritten digit recognition.
The results are expected to be applied in the smartphone chips to facilitate the artificial intelligence apps operation.
