Apr 26, 2025

Optical chips revolutionize artificial intelligence: energy saving with light 17 A German startup currently creates processors that perform calculation operations with light. Optical processors have the potential to enormously accelerate the AI’s calculations. Lasert test of a director of the bright waves of the new chips in a q.ant laboratory in Stuttgart PD Light accelerates everyday life. A glass fiber brings movie in the living room without delay. A laser scanning to the cash register every half. The light could soon bring the profit of the next rhythm: chip to the computer that are expected to be bright signals instead of electronic. These chips should make the main calculators more efficient that reach their limits under the load of applications to high -intensity data. Autonomous cars could also use optical processors one day to protect the battery. Optimizes browser settings Regulate the settings. The researchers have worked on optical computer chips for a long time, now companies want to test the first applications. Stuttgart’s startup Q.ant collaborates with the Leibniz data center in Garching near Munich and the Jülich research center. Together they want to find out for which arithmetic tasks The optical chips are suitable. For this, Q.ant builds a production system in Stuttgart, which will be due to record series production in 2028. Technology makes some artificial intelligence calculations up to 50 times faster, the company promises with only thirteen energy consumption. The training of models to the eats energy Artificial intelligence algorithms can be optimized to process data in a more efficient way and save energy, as shown by the model of the Chinese Deepseek. But the hardware remains a problem: the processing of computer chips in order that makes constant storage stores necessary. Since arithmetic and storage are separated, push the energy costs back and forth. But it could also work differently: with the new basic hardware, which requires in particular the activities that consume time to a conventional computer. The idea: simple but enormous processing phases such as multiplication or addition with light. An optical chip leads light through the waves as on a marshalling courtyard. Along the wave conductors, the modulators thus called in particular removes the light, which creates multiplication. In other places, the tracks cross and bright signals overlap. This therefore matures mathematically to an addition. Multiplications and additions are enormous in today’s artificial intelligence applications, for example in the detection of images or in vocal models. By outsourcing these tasks and making an optical chip, the vision can be saved a lot of energy. Optical chips work in a very different way than electronics. However, there are similarities. Both can be made of silicon. The chip industry has a lot of experience. However, other materials can also be processed with similar methods, for example lithium niobat. The fotonic chip of q.ant accelerates artificial intelligence applications. Q.ant This material is suitable for optical chips due to its interesting optical properties. Its refraction index can be changed by creating an electrical voltage. With this effect, Q.ant controls the optical components on his chip. “This can be done with very low tensions, therefore very efficient from an energy point of view,” explains Michael Förtsch, CEO of Q.ant. A disadvantage of optical chips is that they cannot be made small as electronic, since their components must have the size of the light wavelength (a few hundred nanometers): the transistors on the classic chips are about a hundred times smaller. Optical chips elaborate information in parallel But the light can develop the information at the same time: many wavelengths, that is, colors, can be traveled through the chip and manipulated at the same time. Parallelism is almost unlimited, says Michael Förtsch. “It is possible to obtain extreme speed advantages,” he says. Since the light “calculates” while traveling through the optical chip, it is not necessary to archive data, time and energy costs. In addition to Q.ant, other companies try to market these advantages, Information about Lumai from Oxford . According to the company, its technology uses “millions of light rays” simultaneously and perform some artificial intelligence calculations 1000 times faster than classic chips. The Californian Lightmatter company has also developed an optical processor especially for artificial intelligence calculations. According to the company’s website It must be used in daily applications such as chatbots, e-commerce or autonomous vehicles. The processor It is described in detail in the specialist magazine “Natura”. Contains four optical chips and electronic connections to be able to incorporate it into a conventional IT environment. This is the production of chips with conventional methods of the “Lasse semiconductors industry, they expect that the photonic arithmetic accelerators will find their way into the real systems in the near future”, writes Anthony Rizzo of the Dartmouth College of Hanover in a “nature” accompanying comment. So replace Classic for quick and efficient optical computers from an energy point of view? It is not so simple. Optical chips do not simply exchange electricity with light, but break the digital arithmetic itself. While digital computers represent everything through 0 and 1 and therefore in principle they get high precision, the optical chips expect analog signals, which are shown by physical dimensions such as the intensity of light. These dimensions can only be measured to a limited extent; They are susceptible to noise and float due to inaccuracies in the production of chips to the chip. Q.ant still wants to have reached precision as a 16 -bit digital calculator from 3 to 4 decimal seats. To make a comparison: a typical processor in a smartphone can develop data with precision at 64 bit, or 15-16 decimal places. Since a deeper precision is sufficient for some artificial intelligence activities such as images detection, the Q.ant chip stops well in the tests. The hardware recognized the handwritten figures with precision of 95 percent. The success rate of the lightmatter processor for hand -written characters is even 99.3 percent. They are therefore only 0.1 percentage points behind digital computers. In other applications, the processor also approached the services of an electronic computer. This includes, for example, an artificial intelligence algorithm that learns Atari games such as Pac-Man or a linguistic model that classifies cinematographic reviews as positive or negative. The light to be expected is not better for all applications Förtsch admits that optical processors will not bring advantages for all applications. As a result, potential users approach optical arithmetic. “This is an exciting research project for us,” says Stefan Krieg of the Jülich Research Center on cooperation with Q.ant. “We want to understand the possibilities of hardware,” says the physique, which uses high -performance calculators for physical simulations and also uses methods. Its team wants to compare optical processors with other new types of computers, such as quantum computers. “Applications will not be performed completely on optical chips,” adds Josef Weidendorfer, head of the future calculation at the Leibniz data center. As a rule, optical chips must be integrated into conventional computers. But this makes things complicated: information must often be converted between optical and electronic signals. Since there is still no optical memory, the information must also be electronically stored. “The energy requirement of these conversions has yet to be studied,” says Weidanderfer. The tests that Q.ant will take with the partners of science and industry in the next three years should answer these questions. Only then will you know if the optical chips will actually bring advantages compared to conventional chips in practice. An article by “”