Artificial intelligence is rapidly moving beyond cloud computing into the physical world. Future intelligent systems including autonomous vehicles, robots, smart sensors and human-machine collaborative platforms must not only process information but also perceive, learn, make decisions and interact with dynamic environments in real time. However, conventional electronic computing architectures face increasing challenges in supporting such applications. The separation of memory and computation results in intensive data movement, creating bottlenecks in latency, bandwidth and energy efficiency. These limitations are becoming particularly critical for edge intelligence and embodied intelligence, where fast response, low power consumption and local autonomy are essential.
To address these challenges, researchers from the University of Shanghai for Science and Technology (USST), led by Min Gu, Xinyuan Fang and Yuetian Jia propose a new framework called BOC. The Perspective published in National Science Review outlines the development, opportunities and future directions of this emerging field.
Inspired by biological nervous systems, brain-like intelligence aims to emulate the principles of perception, learning, memory and decision-making found in the brain to build more adaptive and autonomous intelligent systems. Within this framework, brain-like perception emphasizes real-time and event-driven sensing, while brain-like computing focuses on integrating memory, learning and decision-making directly at the hardware level.
In this Perspective, the BOC framework brings together optical neural networks (ONNs), optical spiking neural networks (OSNNs), memory-engram neural networks (ENNs) and optical memristive devices within a unified photonic architecture. ONNs provide rapid perception and feature extraction, OSNNs enable event-driven processing and ENNs together with optical memristors furnish memory and learning capabilities. By tightly integrating sensing, memory and computation in a single photonic platform, the framework reduces the need for frequent data transfers and opens new opportunities for low-latency, energy-efficient intelligent systems operating at the physical edge.
The authors argue that BOC represents a transition in optical computing research—from a computation-oriented paradigm focused primarily on accelerating mathematical operations to an intelligence-oriented paradigm capable of supporting adaptive and autonomous behavior. By integrating event-driven processing, memory and learning capabilities directly into photonic platforms, the framework aims to provide a foundation for future intelligent systems operating at the physical edge.
Potential applications span a wide range of emerging technologies. In edge intelligence, BOC could support real-time processing for autonomous driving, holographic communications and intelligent healthcare devices. In embodied intelligence, it could enable low-latency perception, decision-making and control for autonomous robots, remote operation systems and advanced human-machine collaboration.
According to the researchers, continued advances in photonics, materials science, neuroscience and artificial intelligence are expected to accelerate the development of brain-like optical computing. The framework may ultimately provide a key technological foundation for the transition from digital intelligence to physical intelligence, helping future intelligent machines perceive, learn and act more efficiently in the real world.
National Science Review
Commentary/editorial