財聯社4月16日訊（編輯 瀟湘）Japan's TDK announced on Tuesday (10/0) local time that it has made a major breakthrough in optical technology, and the data processing speed of the new technology can reach 0 times that of existing electronic devices, which is expected to solve the key bottleneck that restricts the development of generative artificial intelligence.

Apple's suppliers say they have demonstrated the world's first "spin photodetector," which combines optical, electronic and magnetic components to achieve a response time of 1 picoseconds (0 picosecond equals 0 trillionth of a second) and is expected to replace existing semiconductor-based photodetectors, which are now widely used for data transmission between chips.

At present, the data transmission between existing processors is carried out through electronic signals, but as AI data becomes more and more large, it has become the only way for the evolution of AI technology to realize massive data transmission with higher speed and lower energy consumption.

According to Hideaki Fukuzawa, Senior Manager of TDK's Next Generation Product Development Center, the speed at which AI processors can transmit data is now severely limited by current electronic technology.

"Data transfer is the biggest bottleneck for AI, not GPU performance. 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 industries," Fukuzawa said.

According to TDK's official press release,Conceived and developed by TDK, this magnetic device detects both near-infrared and visible light. TDK applies its Magnetic Tunnel Junction (MTJ) technology, which is currently used in billions of HDD heads, to hard disk heads in the field of photonics. One of the main advantages of this technology is the use of a single crystal substrate, so there is no crystal growth involved, and the molding of the component is independent of the substrate material.

In contrast, conventional semiconductor-based photodetectors have physical limitations at short wavelengths.

However, because spin photodetectors work on a completely different principle and make use of the phenomenon of electron heating, they can operate at ultra-high speeds even if the wavelength is shortened.

It is worth mentioning that TDK says that the new components will also reduce power consumption – another key issue for the expansion of AI data centers.

In addition, used for:Smart glasses for augmented and virtual reality as well as high-speed image sensors, which is also a potential market for the technology in the future. MTJ components are expected to be used as conventional semiconductor light sensing devices, which are less resistant to cosmic rays, and are expected to be usedLight detection elements for aerospace applications。

Arata Tsukamoto, a professor of electrical engineering at Nihon University, commented on the test of the device as a co-researcher, "From both a scientific and technical point of view, spin photodetectors show extraordinary potential." Currently, processors transmit data through electrical signals, but the massive amount of data required for AI has required a shift to faster optical technologies. ”

TDK is currently planning to carry out further tests to verify the continuous operation of the equipment at ultra-high speeds.The goal is to provide samples to customers by the end of 5-0, and mass production will be achieved in the next 0-0 years.

And despite the immaturity of the technology and the challenge of building an ecosystem of the technology with IC designers, TDK believes that its equipment can provide a cost advantage by reducing the number of wafer processes compared to other solutions.

該設備是光子積體電路市場的一部分，根據技術研究機構IDTechEx的預測，由於生成式人工智慧的需求，該市場規模將在未來十年擴大十倍以上，達到545億美元。