The First Photonic Neural Morphology Chip with Laser Power

The research team at Princeton University in the US has developed the world's first photonic neural morphology chip and demonstrated that it can be calculated at ultra-fast speeds, according to the MIT's Technical Review website. The chip is expected to open a new photon computing industry.
Princeton University Alexandre Tate team's new achievement is the use of photons to solve the neural network circuit speed limitation of this problem. Neural network circuit has been in the field of computing storm. Scientists hope to create more powerful neural network circuit, the key is to create a circuit that can work as neurons, or nerve morphology chip, but the main problem is to increase the speed of such circuits. Photon computation is the "star of tomorrow" in the field of computational science. Compared with electronics, photons have more bandwidth and can process more data quickly. However, photonics data processing systems have a high manufacturing cost and have not been widely adopted.

The core of the photon neural network, developed by the team, is an optical device in which each node has neuron-like response characteristics. These nodes are in the form of miniature circular waveguides etched into a silicon substrate in which light can circulate. When light is input, the output of the 100mw laser pointeroperating at the threshold is then adjusted, in which small changes in incident light can have a significant effect on the output of the laser.
The principle of this optical device is that each node in the system uses a certain wavelength of light, a technique known as wavelength division multiplexing. The light from each node is fed into the laser, and the 2000mw laser pointer output is fed back to the node, creating a feedback circuit with non-linear characteristics. The extent to which this nonlinear behavior simulates neural behavior has been shown to be mathematically equivalent to a device known as the "continuous-time recurrent neural network (CTRNN)", which suggests that CTRNN's programming tools can be applied to more Large silicon photon neural networks.
The Tate team used a 49-node silicon photon neural network to model the mathematical problem of a differential equation and compare it to an ordinary central processing unit. The results show that the speed of the photon neural network is improved by three orders of magnitude in this task. The researchers said that this will open a new photon computing industry. "Silicon photonic neural networks are likely to be the 'vanguard of a larger, scalable silicon-based photonic system," Tate said.