The Germans have developed a processor that works with electric
Hardware

The Germans have developed a processor that works with electric fields rather than currents

Scaling computing resources for artificial intelligence tasks comes with an impressive increase in consumption and hardware costs. German startup Semron offers Reduce dependence on both factors. The company’s founders introduced a new neural network control device they called meme capacitors. It works with electric fields rather than currents.

    AI generation “Chinese quantum computer”, “anime” style.  Image source: AI Generation Kandinsky 3.0/3DNews

Image source: AI Generation Kandinsky 3.0/3DNews

The founders of Semron were graduates of the Technical University of Dresden Kai-Uwe Demasius and Aron Kirschen. They received a patent for the “Memcapacitor” control element back in 2016. According to their ideas, abandoning the operation of neural networks on classic chips with transistors controlled by electrical currents would enable the development of energy-saving and cost-effective neural processors.

“Due to the expected scarcity of computing resources [для работы] In the face of artificial intelligence, many companies whose business model is based on access to such resources are endangering their existence. These are, for example, large startups that train their own models, – said Kirshen in an interview with one of the site’s authors TechCrunch. “The unique features of our technology will allow us to match the price levels of today’s consumer electronics chips, even though our chips have advanced artificial intelligence that others do not offer.”

Put simply, Semron is talking about the ability to produce cheap chips like those used in smartphones, headsets and similar wearable devices that can still power powerful neural networks.

    Image source: Nature

Conceptual structure of a storage capacitor. Image source: Nature

Semron chips are a multi-layered organization that enables significantly scalable solutions. An elementary memory cell or a storage capacitor-based computing element must contain a dielectric with a storage effect (charge or capacity). It could be ferroelectric, for example. Depending on the amount of charge, one or another weight coefficient is written into the cell, which is used for the calculations. This storage dielectric in turn separates two electrodes that induce mutual electric fields. The “coupling” of these fields depends on the coefficient written into the cell – how much the interlayer shields the electromagnetic field, which serves as a value for calculations.

Let’s hope for a development in the area of ​​silicon. Founder of the company published a series of papers in renowned scientific journals and they promise to soon show how storage capacitors work in practice.

RELATED TOPICS

About the author

Dylan Harris

Dylan Harris is fascinated by tests and reviews of computer hardware.

Add Comment

Click here to post a comment