Scientists demonstrate reinforcement learning on photonic chips that compute using optical spikes instead of electronic signals.
Researchers have built new photonic computing chips that allow neural networks to learn using light instead of electronics. The technology could improve autonomous vehicles and help robots learn directly from real-world interactions.
The chips run a type of artificial intelligence called a photonic spiking neural system.These systems mimic how biological neurons communicate using rapid pulses. In this case, those signals travel as light.Scientists say the new system removes a major bottleneck that has slowed photonic AI development. Earlier systems still needed electronics for key steps in learning. The new design keeps those operations entirely in the optical domain.Photonic neural computingPhotonic spiking neural systems use short bursts of light to represent neural signals. These optical spikes move through photonic circuits much faster than electrical signals.However, previous photonic systems could only process linear computations using light. The nonlinear operations required for learning still relied on electronic hardware.“Photonic spiking neural systems use brief optical pulses, or spikes, to emulate neural signaling, but they can typically only process the linear parts of computation using light,” said research team leader Shuiying Xiang from Xidian University in China.Xiang explained the limitation in earlier systems.“Previously, the nonlinear steps that make learning and decision making possible required the signal to be converted back into electronic signals. This adds delay and undercuts the speed and energy advantages of photonics.”The new system removes that limitation. It performs both linear and nonlinear neural computations directly with light.The researchers designed a programmable photonic neuromorphic platform to demonstrate the concept. The system uses two chips working together.One chip contains a 16-channel photonic neuromorphic processor. It includes 272 trainable parameters and processes multiple optical signals simultaneously.The second chip features a distributed feedback laser array with a saturable absorber. This component enables low-threshold nonlinear optical spiking.The researchers tested the system using reinforcement learning. This AI approach trains systems through trial and error.“We used this system to demonstrate reinforcement learning, supported by a hardware and software collaborative framework that trains and runs the neural network,” said Xiang.“The system was able to learn quickly through trial and error, showing potential as a fast, low-latency solution that could be used for applications such as autonomous driving and embodied intelligence.”Engineers trained the neural model in software first. The chips then performed hardware training and execution. Researchers later fine-tuned the results in software to account for small hardware variations.The team evaluated the system using two standard control problems. One involved balancing a pole on a moving cart, known as the CartPole task. The other required stabilizing an inverted pendulum.Hardware decisions closely matched the software model. Accuracy dropped only 1.5 percent for CartPole and 2 percent for the pendulum test.The system also delivered strong computing performance. Photonic linear processing reached 1.39 tera operations per second per watt. Nonlinear computation achieved nearly 988 giga operations per second per watt.On-chip computing latency measured just 320 picoseconds.Toward photonic AI systemsResearchers believe the technology could support future AI systems that require fast learning and low energy use. Potential applications include autonomous driving and robotic systems that adapt in real environments.The current prototype uses 16 optical channels. Future designs may scale to larger architectures.The team plans to develop a 128-channel photonic spiking neural chip. That upgrade could support more complex reinforcement learning tasks.Researchers also aim to build compact hybrid photonic systems suitable for edge computing.If successful, photonic AI hardware could offer an alternative to electronic processors in future intelligent machines.The study is published in the journal Optica.
Neuromorphic Computing Optical Computing Photonic Chips Photonic Computing Robotics AI
United States Latest News, United States Headlines
Similar News:You can also read news stories similar to this one that we have collected from other news sources.
Scientists Discover New Evidence a Common Virus Helps Trigger MSThe Best in Science News and Amazing Breakthroughs
Read more »
Marvel Officially Announces New Leader for the Avengers' New Appearanceiron man and captain america with the avengers logo and various comic covers 2
Read more »
UNIQLO to open three new NYC stores, launches partnership with New York Public LibraryThe popular apparel retailer UNIQLO is opening three new stores across the city and launching a new partnership with the New York Public Library (NYPL).
Read more »
BenQ’s new Mac monitor could be a cheaper alternative to Apple’s new Studio DisplayBenQ’s new 27-inch MA270S display is designed for Mac users with Thunderbolt 4 ports, a sleek design, and a 5K resolution.
Read more »
Deep Ellum’s New Deal: The Struggle, The Shift and a New Restaurant RowFor Restaurants, Deep Ellum is a punishing obstacle course but a little help might be on the way.
Read more »
New Number, New Coach, New Opportunity for Gators' Jayden WoodsAfter entering the NCAA Transfer Portal, and after being heavily recruited by Texas, Jayden Woods ultimately decided to return to the Florida Gators.
Read more »