Researchers from Tsinghua University have developed a groundbreaking optical processing system that leverages light instead of electricity to enhance artificial intelligence (AI) capabilities. This development addresses a significant challenge faced by AI models: the conversion of raw data from cameras, sensors, and market feeds into usable features, which is often inefficient on conventional chips.
Current electronic systems consume substantial time and power, leading to excessive energy waste. The new device, termed the Optical Feature Extraction Engine (OFE2), aims to rectify this by performing computations optically, which can potentially accelerate AI processing speed to a practical level outside of laboratory settings.
The OFE2 operates by converting a serial data stream into multiple distinct optical channels on a chip. This design allows mathematical computations to be conducted as light travels through a specifically patterned section of the engine, facilitating precise interference of light waves. As photons are not subjected to electrical resistance, the system boasts low latency and minimal energy consumption.
Project lead and researcher Hongwei Chen emphasized the significance of their work, stating, “We firmly believe this work provides a significant benchmark for advancing integrated optical diffraction computing to exceed a 10 GHz rate in real-world applications.” In initial testing, OFE2 demonstrated a processing speed of 12.5 GHz, completing individual operations in around 250 picoseconds, a time interval that is crucial for applications such as real-time medical imaging and financial trading.
The implications of this technology are substantial. In a demonstration related to medical imaging, the OFE2 effectively extracted crisp edges from CT images, which enabled subsequent models to categorize these images with greater accuracy while requiring less power and fewer electronic adjustments. Similarly, in a trading application, the optical engine processed live price feeds, generating buy and sell signals with lower latency than conventional electronic systems.
However, the researchers caution that while the optical engine excels in handling straightforward mathematical tasks, it does not eliminate the need for traditional CPUs or GPUs, which are still necessary for more complex calculations later in the process. The integration of optical components also requires meticulous design to ensure clean data handling, but the system’s ability to be retuned dynamically allows for task flexibility without the need for hardware modifications. This adaptability is particularly beneficial in environments such as clinics, factories, and trading floors, where workload demands can vary significantly.
The findings from this research were published in the October 2025 issue of Advanced Photonics Nexus. As optical computing continues to evolve, this innovation could pave the way for more efficient and faster AI systems, marking a notable step forward in the integration of optical technologies into mainstream computing applications.
See also
Nvidia Stock Plummets Over 3% as Meta Considers $100B Google AI Chip Deal
Elon Musk Claims His AI Can Defeat League of Legends Champion Faker in Upcoming Match
Broadway’s Marjorie Prime Sparks Urgent Debate on AI’s Ethical Implications in Grief
Bachchan Lawsuit Against Google Highlights Urgent Need for AI Personality Rights Reform
MIT’s BoltzGen AI Model Transforms Drug Discovery with Novel Protein Design for Undruggable Diseases
