As AI systems drive surging energy demand, researchers are exploring gallium nitride microLED arrays that process neural signals using light instead of electricity.
Artificial intelligence systems already consume vast amounts of energy. A single ChatGPT request can require roughly ten times more power than a Google search, and the demand adds up quickly. Estimates suggest the popular chatbot could consume around 40 million kilowatt-hours of electricity per day.
To put that into perspective, that’s enough energy to charge roughly eight million smartphones—nearly every phone in New York City. And this is only the beginning. As AI tools become more powerful, complex, and personalized, their energy appetite is expected to grow rapidly. Data from Lawrence Berkeley National Laboratory suggests that by 2028, more than half of the electricity used by data centers could be driven by AI workloads.of all US households combined. With data center power demand rising by roughly 15 percent each year, scientists are searching for ways to make AI significantly more energy-efficient.) project. By pairing tiny gallium nitride microLEDs with silicon electronics, the researchers aim to develop a new type of optical neuromorphic hardware that processes information using light rather than electricity.. It belongs to the family of wide-bandgap semiconductors and is characterized by high electron mobility and excellent thermal stability.Although GaN is now central to modern electronics, it was once considered unsuitable as a semiconductor because of its imperfect crystal structure. Early GaN crystals contained more than By comparison, typical LED semiconductor materials contained fewer than 100,000 defects per cm², while silicon crystals could have as few as 100.That changed in the early 1990s when researchers Isamu Akasaki and Hiroshi Amano of Nagoya University, along with Shuji Nakamura of the University of California, Santa Barbara, used GaN to create high-brightness blue LEDs. Their breakthrough earned them the According to Andreas Waag, PhD, managing director of the Institute of Semiconductor Technology at TU Braunschweig and one of the scientists involved in the BRIGHT project, GaN gradually paved the way for modern LED technologies. “Step by step, gallium nitride became a wide-bandgap semiconductor capable of producing blue light emitters,” Waag said in an exclusive interview with, also known as neuromorphic engineering, mimics how the human brain processes information. The approach allows artificial neural networks to operate far more efficiently than conventional computing architectures, potentially reducing the high energy costs associated with modern AI systems. While current AI models require enormous amounts of energy for training and operation, neuromorphic chips are inspired by the brain’s ability to run on roughly 20 watts of power. These systems rely on sparse, event-driven processing that activates circuits only when needed. As a result, neuromorphic processors can use only a fraction of the energy required by conventional CPUs or GPUs while still performing complex AI tasks. However, when neuromorphic systems rely on conventional silicon electronics, building the massive number of connections required between artificial neurons becomes increasingly difficult. Maximilian Müller, a PhD student; Optimizing the parameter settings in an optical neuromorphic processing system. Credit: Jan Hosan/TU Braunschweig According to Waag, each of these connections must be physically wired, which, in turn, quickly limits how large and efficient these systems can become. “This is why we use microLEDs,” he said. “When anBernhard Wicht, PhD, a co-speaker of the BRIGHT initiative, as well as head of the Department for Mixed-Signal IC Design at Leibniz University Hannover, said the project is designed to tackle this challenge. “In the BRIGHT consortium, we combine the complementary expertise of leading microelectronics groups,” Wicht pointed out. “We are committed to leading this collaboration to develop a uniqueas a potential solution to the increasing energy demands of AI and to achieve major strides in this field.”into an optical one. “Gallium nitride has a unique property among semiconductor materials,” he stated. “It allows us to build extremely small LEDs, micrometer or even nanometer scale, that still emit light efficiently.” This makes it possible to build optical systems with extremely high information density, comparable to modern microelectronics, but operating on light rather than electrical signals. “The key advantage is that light naturally enables massively parallel interconnectivity,” he added. microelectronics consortium and supported with €15 million from the State of Lower Saxony and the Volkswagen Foundation, aims to combine gallium nitride microLEDs with silicon-based integrated circuits .each of them separately controllable at very high frequencies. Credit: Jan Hosan/TU Braunschweigserve as the system’s light sources, while silicon electronics handle signal control and integration. Together, they create a platform capable of performing neuromorphic computations in the optical domain. Such an optical neuromorphic processor would transmit and process information using photons rather than electrons. Because light can travel faster and generate less heat than electrical currents, these systems could significantly reduce the energy required to run AI workloads.Waag believes this hybrid integration could benefit not only neuromorphic computers but also a range of emerging microelectronic applications.power electronics If millions of microLEDs can be driven in parallel by a silicon chip and operated at gigahertz frequencies, the result would be extremely high-bandwidth data streams. “We began exploring applications for this about seven or eight years ago,” Waag said. Daniel Prades, PhD, a professor at TU Braunschweig; inspecting the emission of microLED arrays by optical microscopy and intensity measurements.The integration of light, microelectronics, and neuromorphic computing could pave the way for powerful AI systems that consume far less energy, making them more sustainable—particularly for large-scale data centers. “Today, large AI and high-performance computing centers are often built near nuclear power plants simply because of their enormous electricity demands,” the researcher added. “Some estimates suggest that by 2035, AI could account for 30 to 40 percent of global electricity use.”, especially if AI continues to grow exponentially. “This is why energy efficiency must become a central priority when building future AI infrastructure,” Waag continued.The current prototype contains about 1,000 neurons and utilizes microLED arrays originally built for other uses. The team’s goal is to create a compact design built specifically for energy efficiency. “Simulations suggest it can be at least 100 times more efficient than current AI processors.” Waag said. He added that the team has already begun discussions with data centers about the technology. “We also launched a startup,, which is now beginning to explore these opportunities,” he said. “The systems will need to be integrated into a plug-and-play environment before they can be adopted.” Waag stated that GaN is crucial for the continent’s strategic advantage. “While silicon CMOS is largely dominated by Asia and the United States, Germany and Europe have a strong position in GaN technology,” he said. “Supporting nitride technologies could therefore strengthen Europe’s technological position.”He believes that the technology will become a natural part of modern processors in the future. “Today processors are entirely silicon, but in the next decade gallium nitride could be integrated into more advanced chips with greater capabilities and far better energy efficiency,” Waag said. Helene Kuhn, PhD, a researcher at TU Braunschweig and scientific manager of the NTC BRIGHT project, said that the initiative highlights the growing importance of gallium nitride in future computing systems.Based in Skopje, North Macedonia. Her work has appeared in Daily Mail, Mirror, Daily Star, Yahoo, NationalWorld, Newsweek, Press Gazette and others. She covers stories on batteries, wind energy, sustainable shipping and new discoveries. When she's not chasing the next big science story, she's traveling, exploring new cultures, or enjoying good food with even better wine.Science
Energy &Amp Environment Gallium Nitride Germany Inventions And Machines Microleds Neuromorphic Computing Optical Computing Science Semiconductors
United States Latest News, United States Headlines
Similar News:You can also read news stories similar to this one that we have collected from other news sources.
Bethenny Frankel's 'beyond comfortable' tank tops are just $7 eachSLIM FIT: Close fit that hugs the body. COTTON MODAL MICRO RIB: Super soft cotton modal micro rib with added stretch offers all-day ease and comfort, while retaining its shape for the perfect slim fit. Fabric you want to live in year round.
Read more »
The microneedling-inspired treatment that sold out in 3 months is backDr. Brandt Skincare Needles No More Dermal Micro-Points is a microneedling-inspired nighttime treatment that uses dissolving sponge spicules to create microchannels in the epidermis, enhancing ingredient penetration and stimulating collagen production.
Read more »
A college within a college: How Sacramento State’s Black Honors College centers Black studentsBlack Honors College (BHC) is a “micro-university” nestled within the campus of California State University, Sacramento (Sac State).
Read more »
Home-building robots could help fix the housing crisisUK tech company AUAR makes portable micro-factories that it says can produce the wooden frame of a house in a day.
Read more »
Another Bose Deal Beyond Headphones, SoundLink Micro Portable Bluetooth Speaker (2nd Gen) Is Now Below Black Friday PricingPump up the music with ease with this amazing speaker.
Read more »
Labor Pains Raise Chances of Fed Rate Cuts as Payrolls DeclineMarket Analysis by covering: Crude Oil WTI Futures, Crude Oil WTI Spot US Dollar, Brent Spot US Dollar, Micro WTI Crude Oil Futures. Read 's Market Analysis on Investing.com
Read more »