Power/Performance Bits: July 31

Training optical neural networks
Researchers from Stanford University used an optical chip to train an artificial neural network, a step that could lead to faster, more efficient AI tasks.

Although optical neural networks have been recently demonstrated, the training step was performed using a model on a traditional digital computer and the final settings were then imported into the optical circuit. In contrast, the new method trains networks directly in the device by implementing an optical analog of the ‘backpropagation’ algorithm, which is the standard way to train conventional neural networks.

“Using a physical device rather than a computer model for training makes the process more accurate,” said Tyler W. Hughes, a PhD student at Stanford. “Also, because the training step is a very computationally expensive part of the implementation of the neural network, performing this step optically is key to improving the computational efficiency, speed and power consumption of artificial networks.”

The new training protocol operates on optical circuits with tunable beam splitters that are adjusted by changing the settings of optical phase shifters. Laser beams encoding information to be processed are fired into the optical circuit and carried by optical waveguides through the beam splitters, which are adjusted like knobs to train the neural network algorithms.

“Our method not only helps predict which direction to turn the knobs but also how much you should turn each knob to get you closer to the desired performance,” said Hughes. “Our approach speeds up training significantly, especially for large networks, because we get information about each knob in parallel.”

Researchers have shown a neural network can be trained using an optical circuit (blue rectangle in the illustration). In the full network there would be several of these linked together. The laser inputs (green) encode information that is carried through the chip by optical waveguides (black). The chip performs operations crucial to the artificial neural network using tunable beam splitters, which are represented by the curved sections in the waveguides. These sections couple two adjacent waveguides together and are tuned by adjusting the settings of optical phase shifters (red and blue glowing objects), which act like ‘knobs’ that can be adjusted during training to perform a given task. (Source: Tyler W. Hughes, Stanford University)

In the new training protocol, the laser is first fed through the optical circuit. Upon exiting the device, the difference from the expected outcome is calculated. This information is then used to generate a new light signal, which is sent back through the optical network in the opposite direction. By measuring the optical intensity around each beam splitter during this process, the researchers showed how to detect, in parallel, how the neural network performance will change with respect to each beam splitter’s setting. The phase shifter settings can be changed based on this information, and the process may be repeated until the neural network produces the desired outcome.

The researchers tested their training technique with optical simulations by teaching an algorithm to perform complicated functions, such as picking out complex features within a set of points. They found that the optical implementation performed similarly to a conventional computer.

“Our work demonstrates that you can use the laws of physics to implement computer science algorithms,” said Shanhui Fan, a professor at Stanford. “By training these networks in the optical domain, it shows that optical neural network systems could be built to carry out certain functionalities using optics alone.”

The researchers plan to further optimize the system and want to use it to implement a practical application of a neural network task. The general approach they designed could be used with various neural network architectures and for other applications such as reconfigurable optics.

Thermal material
Engineers at UCLA developed a semiconductor material, boron arsenide free from defects, which they say is more effective at drawing and dissipating waste heat than any other known semiconductor or metal materials.

Heat in electronics has become a major issue, as thermal issues increase along with gate density and lead to problems including unpredictable reliability, mechanical stress, increased leakage power, and premature aging. While there are a number of ways designers approach thermal issues, such as using temperature monitoring to reduce performance when necessary, the research team focused on drawing heat away from hotspots.

The defect-free boron arsenide has a record-high thermal conductivity, 1300 W/mK, making it more than three times faster at conducting heat than the commonly used materials silicon carbide and copper.

Illustration showing a schematic of a computer chip with a hotspot (bottom); an electron microscope image of defect-free boron arsenide (middle); and an image showing electron diffraction patterns in boron arsenide. (Source: Hu Research Lab/UCLA Samueli)

“This material could help greatly improve performance and reduce energy demand in all kinds of electronics, from small devices to the most advanced computer data center equipment,” said Yongjie Hu, UCLA assistant professor of mechanical and aerospace engineering. “It has excellent potential to be integrated into current manufacturing processes because of its semiconductor properties and the demonstrated capability to scale-up this technology.”

Water into hydrogen
Researchers at Rutgers University–New Brunswick developed an improved way to split water into hydrogen for solar energy storage using star-shaped gold nanoparticles.

“Instead of using ultraviolet light, which is the standard practice, we leveraged the energy of visible and infrared light to excite electrons in gold nanoparticles,” said Laura Fabris, associate professor in the Department of Materials Science and Engineering at Rutgers. “Excited electrons in the metal can be transferred more efficiently into the semiconductor, which catalyzes the reaction.”

The team says their use of visible and infrared light allowed gold nanoparticles to absorb it more quickly and then transfer some of the electrons generated as a result of the light absorption to nearby materials like titanium dioxide.

The engineers coated gold nanoparticles with titanium dioxide and exposed the material to UV, visible, and infrared light and studied how electrons jump from gold to the material. The researchers found that the electrons, which trigger reactions, produced hydrogen from water over four times more efficiently than previous efforts demonstrated.

“This was our first foray,” Fabris added, “but once we understand the material and how it operates, we can design materials for applications in different fields, such as semiconductors, the solar or chemical industries or converting carbon dioxide into something we can use. In the future, we could greatly broaden the ways we take advantage of sunlight.”