Chen Gen: Super chip, 2 billion images can be processed per second

Wen | Chen Gen

Human brain has billions of neurons, and there are thousands of connections between each neuron and other neurons. Many studies hope Improve human brain to improve information processing capabilities. In recent years, with the development of science and technology, the development of deep neural networks has gained a qualitative leap. In the fields of computer vision, voice recognition, medical diagnosis, deep neural network provides a lot of support.

At present, the consumer -grade image classification technology on digital chips can be performed by billions of calculations per second, which can meet most applications. But more complicated images, such as identifying sports objects, 3D objects, or human cells, still face many obstacles. In addition, in the field of optics, although photon computing has made good progress, the lack of scalable optical chips and the loss of photonic devices have limited the advancement of optical deep networks.

Recently, researchers at the University of Pennsylvania have developed a powerful new optical chip, which is the first deep neural network chip that is fully achieved on integrated photon device in an integrated photon device. The size is only 9.3 mm (0.01 inches), and nearly 2 billion images can be classified per second.

Traditional electronic equipment usually uses a digital clock -based platform, such as graphic processing unit (GPU) to achieve data processing, where the calculation speed is limited by the clock frequency and memory access time. In addition, the original image data usually needs to be converted to a digital electronic signal, which not only takes a long time, but also requires the big memory unit to store images and videos, which may cause potential privacy issues.

Different from the traditional electronic system, the chip is processed in the form of light. It uses optical cables as neurons and stacked in multi -layers. Each layer is specifically used in specific types of classification. In the test, the chip classification of image sets containing two types of characters can reach 93.8%accuracy, and the accuracy of the four types of character set recognition can reach 89.8%.

In addition, the chip can classify each character within 0.57 nano, so that it can process 1.75 billion images per second. At the same time, it does not need to store information that is being processed, which greatly saves time and guarantees data security.

At present, related research results have been published in the journal of Nature; in the future, this new chip can promote face recognition, automatic face recognition, automatic automatic, automatic The development of driving and other fields.

- END -