Machine learning architectures based on convolutional neural networks (CNNs) have proved to be highly valuable for a wide range of applications, ranging from computer vision to the analysis of images and the processing or generation of human language. To tackle more advanced tasks, however, these architectures are becoming increasingly complex and computationally demanding.
In recent years, many electronics engineers worldwide have thus been trying to develop devices that can support the storage and computationally load of complex CNN-based architectures. This includes denser memory devices that can support large amounts of weights (i.e., the trainable and non-trainable parameters considered by the different layers of CNNs).
Researchers at the Chinese Academy of Sciences, Beijing Institute of Technology, and other Universities in China have recently developed a new computing-in-memory system that could help to run more complex CNN-based models more effectively. Their memory component, introduced in a paper published in Nature Electronics, is based on non-volatile computing-in-memory macros made of 3D memristor arrays.
To read more, click here.