[Editor-in-Chief’s Circle]
Science and Technology Daily, Beijing, February 3 (Reporter Zhang Mengran) Running a generative artificial intelligence (AI) system not only has high hardware costs, but also causes staggering energy consumption. According to the latest report from the technology website TechCrunch, Semron, a startup company headquartered in Germany, has recently developed an innovative AI chip design method and is the first to use a new neural network control device-memory container to power its 3D chip. This has the potential to revolutionize energy-efficient computing technology and make advanced AI capabilities more accessible to consumer electronics devices.
Unlike transistors in processors, Semron's chips use electric fields instead of electric currents. These memory containers made of traditional semiconductor materials can store energy and control electric fields, which not only improves energy efficiency but also reduces manufacturing costs and makes it easier for consumer electronics to run advanced AI models.
The Semron chip is a multi-layer structure. The core principle is charge shielding. The electric field between the top electrode and the bottom electrode is controlled through the shielding layer. The shielding layer is managed by the chip memory and can store various "weights" of the AI model. Weights are essentially like knobs in a model, manipulating and fine-tuning its performance while training and processing data.
The electric field approach minimizes the movement of electrons within the chip, reducing energy use and heat. Semron aims to harness the cooling properties of electric fields to greatly increase computing power by placing hundreds of layers of capacitors on a single chip.
In a recent study published in the journal Nature Electronics, Semron chips demonstrated significant energy efficiency improvements, achieving an excellent energy efficiency of more than 3500TOPS/W (tera operations per watt per second), surpassing today's There are technologies 35 times to 300 times. This indicator indicates that energy consumption during AI model training will be significantly reduced.
Although still in its early stages, Semron has already attracted the attention of prominent venture capital firms, which could have a major impact on the future of computing resources.
We often suffer from "charging anxiety" when using electronic devices. On the one hand, this is related to insufficient battery life; on the other hand, it is also related to the high energy consumption of the chip. Today, ordinary silicon-based chips encounter the "ceiling" of Moore's Law in terms of computing performance and energy consumption. With the continuous updating and iteration of new generation electronic products and various artificial intelligence devices, there is an urgent need to develop chips using new materials and new design methods to provide consumers with electronic products with more powerful computing performance and more energy saving.