Current processors used for machine learning are limited by the power required to process data when performing complex operations. Generally speaking, the smarter the task, the more complex the data, and the greater the demand for power. Such networks are limited by the slow transmission of electronic data between the processor and the memory. Researchers at George Washington University in the United States found that the use of photons in a neural network (tensor) processor (TPU) can overcome these limitations and create more powerful and energy-efficient artificial intelligence.
A paper recently published in the scientific journal Applied Physics Reviews describes this research and shows that their photonic TPU performance is 2-3 orders of magnitude higher than electronic TPU.
Note: The tensor processing unit (TPU) is a dedicated chip (ASIC) customized by Google for machine learning. It is designed for Google's deep learning framework TensorFlow.
After a breakthrough in artificial intelligence, machines can learn at the speed of light without supervision.
Mario Miscuglio, one of the authors of the paper, said that they found that integrated photonic platforms integrated with high-efficiency optical memory can achieve the same operations as tensor processors, but they only consume a small amount of power and have higher throughput. With proper training, (these platforms) can be used to interfere at the speed of light.
Dr. Miscuglio said that photonic dedicated processors can save a lot of energy, improve response time and reduce data center traffic. Potential commercial applications of this innovative processor include 5G and 6G networks, as well as data centers for massive data processing.
Note: The tensor processing unit (TPU) is a dedicated chip (ASIC) customized by Google for machine learning. It is designed for Google's deep learning framework TensorFlow.
After a breakthrough in artificial intelligence, machines can learn at the speed of light without supervision.
Mario Miscuglio, one of the authors of the paper, said that they found that integrated photonic platforms integrated with high-efficiency optical memory can achieve the same operations as tensor processors, but they only consume a small amount of power and have higher throughput. With proper training, (these platforms) can be used to interfere at the speed of light.
Dr. Miscuglio said that photonic dedicated processors can save a lot of energy, improve response time and reduce data center traffic. Potential commercial applications of this innovative processor include 5G and 6G networks, as well as data centers for massive data processing.
