News
Recently, an engineer team from Columbia University, Cornell University, and other institutions has successfully developed a novel three-dimensional (3D) optoelectronic chip by deeply integrating photonic technology with advanced complementary metal-oxide-semiconductor (CMOS) electronics.
This new chip achieves unprecedented energy efficiency and bandwidth density for data transmission, laying a solid foundation for the development of next-generation artificial intelligence (AI) hardware.
Measuring just 0.3 square millimeters, the newly developed chip integrates 80 high-density photonic transmitters and receivers, delivering a remarkable data transmission bandwidth of 800 gigabytes per second while consuming only 120 femtojoules per bit—an outstanding level of energy efficiency.
In addition, the chip boasts a bandwidth density of 5.3 terabytes per second per square millimeter, far surpassing current industry benchmarks. Its architectural design is also highly compatible with existing semiconductor manufacturing lines, making it promising for large-scale production.
Light, as a communication medium, can transmit vast amounts of data with minimal energy loss. This property has not only driven the rise of fiber-optic communication in the internet era but also holds the potential to significantly enhance computing capabilities. If more efficient data communication between computing nodes can be achieved, the development landscape of AI technologies could be fundamentally transformed.
By integrating photonic technologies, the new chip provides ultra-energy-efficient, high-bandwidth data communication links. These innovations may eliminate bandwidth bottlenecks between spatially distributed computing nodes, thereby accelerating the development of next-generation AI hardware and opening new avenues for faster, more efficient AI systems. Distributed AI architectures—previously limited by high energy consumption and latency—may later become fully viable.
