TDK has adapted hard disk drive technology to deliver data transmission speeds 10x faster than competing solutions for AI infrastructure.

Japanese manufacturer TDK has achieved a major breakthrough in addressing one of artificial intelligence's most pressing challenges: data transfer bottlenecks. The company's innovation, called the Spin Photo Detector, combines magnetic, optical, and electronic technologies to move data at speeds up to ten times faster than existing semiconductor-based photodetectors. The device reaches response speeds of just 20 picoseconds—or 20 trillionths of a second—using light at a wavelength of 800 nanometers. According to Hideaki Fukuzawa, senior manager of the company's next-generation product development center, "This data transfer is the biggest bottleneck for AI rather than the semiconductor GPU performance."

To achieve these speeds, TDK re-engineered its magnetic tunnel junction (MTJ) technology, previously used in billions of hard disk drive heads. The MTJ structure is only 200 nanometers wide and integrates a single CoFeB free layer that responds to laser pulses, allowing switching from parallel to antiparallel magnetic states with 100% reliability under defined conditions. Unlike semiconductors that rely on carrier generation and face limitations at short wavelengths, Spin Photo Detectors use electron heating, enabling faster speeds and broader spectral coverage. The MTJ elements are also resistant to cosmic rays, making them ideal for aerospace applications.

The technology demonstrated effective operation across a wide range of wavelengths, from 300 nanometers to 1.6 micrometers. Its ultrafast analog light detection capability, energy efficiency, and compact size open potential applications in AR/VR smart glasses, high-speed imaging, space systems, and addressing power consumption issues in AI deployment. Arata Tsukamoto, a professor of electrical engineering at Nihon University who collaborated with TDK on validation, stated that "The Spin Photo Detector holds remarkable promise, both from a scientific and technological perspective."

TDK expects to supply samples by March 2026 and aims for mass production within five years. Fukuzawa noted that "since we can break through many of the current bottlenecks, we think this technology will be a game-changer for the AI and data center industry." Thanks to the few processing steps required, TDK expects to maintain a cost advantage and challenge major chipmakers like TSMC and Nvidia, who are investing heavily in silicon photonics.