Chinese scientists have developed a reconfigurable integrated photonic chip capable of running diverse neural networks, including those for image and speech processing, with high efficiency.
Key Details
- 1The chip integrates microring resonators and Mach-Zehnder interferometers, powered by a soliton microcomb light source.
- 2Supports fully-connected, convolutional, and recurrent neural networks within a single hardware architecture.
- 3Area efficiency reaches up to 2.45 TOPS/mm² at 10 GHz frequency.
- 4Demonstrated on tasks: image classification (MNIST 92.93% accuracy, CIFAR-10 56.57%), sentiment analysis (IMDB 80.81%), and speech recognition.
- 5Device enables dual-path computation per resonator, doubling throughput versus traditional schemes.
Why It Matters
This advance shows that photonic computing chips can support the flexible and scalable deployment of multi-model neural networks for imaging and multimodal AI tasks on efficient, high-throughput hardware. Such technology may lead to breakthroughs in hardware acceleration for radiology AI and similar domains.
Source
EurekAlert
Related News
•EurekAlert
AI Predicts Risks for Outpatient Stem Cell Therapy in Myeloma
Researchers use machine learning to predict adverse events during stem cell therapy for multiple myeloma, improving outpatient safety.
•EurekAlert
USC Unveils Joint Biomedical Engineering Department Bridging Medicine, Engineering, and Imaging
USC's medical and engineering schools launch a joint biomedical engineering department to accelerate interdisciplinary research and innovation, including imaging and AI.
•EurekAlert
AI-Enhanced CT Heart Fat Measurement Boosts Cardiovascular Risk Prediction
AI-derived measurement of heart fat from CT scans significantly improves long-term cardiovascular disease risk prediction.