Carding 4 Carders
Professional
- Messages
- 2,724
- Reaction score
- 1,588
- Points
- 113
The new generation optical chip has the ability to self-configure.
Scientists have created an optical chip that can be independently configured to perform various functions. This development opens up new possibilities for use in optical neural networks used for processing large amounts of data, such as image classification, gesture interpretation, and speech recognition.
In the past, photonic integrated circuits have been developed that can be reconfigured after production. However, setting them up is often difficult due to the need to understand the internal design of the chip. The new chip, developed by a team from Huazhong University of Science and Technology in China, allows the user to change its functions without a deep understanding of its structure. "Users can perceive our chip as a black box. It is enough to set a training goal, and the chip will automatically configure itself to achieve the desired functionality," said Jiangji Dong, head of the research group.
The new chip is based on a network of optical components based on waveguides, called Mach-Zehnder interferometers. The researchers demonstrated that the chip can self-tune to perform optical routing, efficient low-loss light energy separation, and matrix calculations needed to create neural networks.
It was also demonstrated that the chip can efficiently route optical signals between equipment in data centers, which reduces latency and power consumption when processing a large number of signals. In addition, the chip was used to divide optical power with minimal losses. Analysis of the results showed that the energy loss during separation remained below 1.16 dB.
At the moment, researchers are working on improving the chip to expand its matrix capabilities and apply it to other matrix computing applications, surpassing the capabilities of optical neural networks.
Scientists have created an optical chip that can be independently configured to perform various functions. This development opens up new possibilities for use in optical neural networks used for processing large amounts of data, such as image classification, gesture interpretation, and speech recognition.
In the past, photonic integrated circuits have been developed that can be reconfigured after production. However, setting them up is often difficult due to the need to understand the internal design of the chip. The new chip, developed by a team from Huazhong University of Science and Technology in China, allows the user to change its functions without a deep understanding of its structure. "Users can perceive our chip as a black box. It is enough to set a training goal, and the chip will automatically configure itself to achieve the desired functionality," said Jiangji Dong, head of the research group.
The new chip is based on a network of optical components based on waveguides, called Mach-Zehnder interferometers. The researchers demonstrated that the chip can self-tune to perform optical routing, efficient low-loss light energy separation, and matrix calculations needed to create neural networks.
It was also demonstrated that the chip can efficiently route optical signals between equipment in data centers, which reduces latency and power consumption when processing a large number of signals. In addition, the chip was used to divide optical power with minimal losses. Analysis of the results showed that the energy loss during separation remained below 1.16 dB.
At the moment, researchers are working on improving the chip to expand its matrix capabilities and apply it to other matrix computing applications, surpassing the capabilities of optical neural networks.
