As electronic chips hit physical limits, scientists are turning to an unexpected solution: light. Optical computing, which replaces electrical signals with photons to perform complex operations, promises unprecedented speed and energy efficiency. Among the most promising techniques? Optical diffraction devices— ultra-thin, plate-like components that manipulate light to carry out mathematical functions. These systems can analyze massive streams of data simultaneously while consuming minimal power. Yet one major roadblock remains: sustaining stable, coherent light waves at speeds exceeding 10 GHz has proven extraordinarily challenging.

A Leap Forward with OFE2

A research group led by Professor Hongwei Chen at Tsinghua University in China has now unveiled a revolutionary device: the Optical Feature Extraction Engine 2 (OFE2). Their innovation, detailed in Advanced Photonics Nexus, introduces a novel method for executing high-velocity optical data analysis, with potential across numerous practical fields.

Smarter Data Input, Faster Processing

One of OFE2’s core breakthroughs lies in how it prepares optical information. Delivering rapid, parallel light signals to the device’s computational heart without disrupting their phase coherence has long been a major hurdle. Traditional fiber-optic setups often cause disruptive phase shifts when splitting or delaying beams. To resolve this, Chen’s team engineered a fully integrated chip-based solution featuring dynamic power dividers and finely tuned delay pathways. This configuration transforms sequential data inputs into multiple perfectly synchronized optical streams. An embedded phase-control array further enables the system to adapt on the fly for diverse computational tasks.

Once formatted, the light signals move through a specialized diffraction component that carries out the actual feature extraction. Functionally akin to matrix-vector multiplication, this stage guides interacting light waves to form intense focal points — or “bright spots” — at designated output locations. By adjusting the input light’s phase, researchers can steer these bright zones toward selected outputs, allowing the system to detect nuanced patterns and temporal changes in incoming data.

Unprecedented Speed and Real-World Impact

Clocking in at an astounding 12.5 GHz, OFE2 completes a single matrix-vector operation in just 250.5 picoseconds — setting a new speed record for this class of optical computation. “This achievement establishes a critical milestone for pushing integrated photonic diffraction technology beyond the 10 GHz threshold in practical deployments,” Chen asserts.

The team put OFE2 through its paces across various domains. In image analysis, it effectively isolated edge details from visual inputs, producing enhanced “relief and engraving” visual maps that boosted classification accuracy — particularly in medical imaging tasks like organ identification in CT scans. Compared to conventional AI architectures, systems incorporating OFE2 required significantly fewer electronic elements, highlighting how optical preprocessing can streamline hybrid AI models for greater speed and efficiency.

In another compelling application, the researchers deployed OFE2 in high-frequency trading. By analyzing real-time market data streams, the system translated price fluctuations directly into actionable buy/sell signals. Once trained on refined trading strategies, it consistently delivered profits, leveraging its light-speed processing to capitalize on fleeting market opportunities with minimal latency.

Bright Prospects for AI and Beyond

Collectively, these advancements mark a pivotal moment in computing. By shifting the most computationally intensive aspects of AI from traditional, energy-demanding silicon chips to ultra-fast photonic platforms, innovations like OFE2 could pave the way for real-time, low-power artificial intelligence. “Our work elevates the performance of integrated diffraction optics, laying the groundwork for demanding applications in image recognition, healthcare diagnostics, and financial technology,” Chen concludes. “We’re eager to partner with organizations facing intensive data processing challenges.”

With its record-breaking speed and versatile real-world applications, OFE2 illuminates a compelling path toward the future of intelligent computing.