Science & Technology (Commonwealth Union) – Artificial Intelligence (AI) data centers are set to become more and more significant in the years ahead. As the relationship between data and AI are extremely significant, the proper management of AI centers which are economical are key. The usage of energy and its impact on the environment is also a key concern.
Researchers from the Karlsruhe Institute of Technology (KIT) in Germany and École Polytechnique Fédérale de Lausanne (EPFL) in Switzerland have introduced a new component that enables extremely fast, cost-effective, and dependable data transmission using an advanced production method. Their electro-optical modulator sends data efficiently through fiber-optic cables and can be produced cheaply in large volumes on conventional semiconductor wafers. This development is significant because the rapid growth of AI applications and rising data traffic are pushing data centers and fiber-optic networks close to their operational limits. The team published their results in Nature Communications.
Like today’s computer chips, the modulator can be manufactured using well-established semiconductor fabrication techniques. The researchers combine lithium tantalate—a material known for guiding light very efficiently and forming the core of the modulator—with a widely used microelectronics chip-manufacturing process. Until now, these two technologies had not been integrated. Their combination makes reliable large-scale production possible for the first time.
Researchers of the study indicated that mass-production methods were proven on millions of chips. Light modulators transform electrical signals into pulses of light, forming the foundation of high-speed internet and making them well suited for handling massive data flows, such as those used in artificial intelligence training. However, the production approach behind this new device is different.
Professor Christian Koos, head of KIT’s Institute of Photonics and Quantum Electronics (IPQ) indicated that the key innovation lies in the copper electrodes and the method they make use of, to produce them. Copper carries signals more efficiently than the gold that has traditionally been used. In addition, copper allows the creation of extremely smooth surfaces, which improves efficiency because less energy is lost during transmission.
These copper electrodes are produced using a manufacturing process that has already been applied millions of times in the fabrication of electronic computer chips. Compared with earlier techniques, this process produces a nearly mirror-like surface, allowing optical microchips to connect more easily with electronic chips. As a result, modulators become simpler to produce and can also be integrated more effectively into existing electronic systems.
Researcher further indicated that record data speeds with reliable performance were demonstrated. Testing by the KIT research team shows that the device can achieve extremely high data speeds while maintaining stable performance.
Alexander Kotz from IPQ pointed out that the modulator delivers very high data rates and, most importantly, operates steadily without the need for continuous adjustments. This is an important improvement because constant recalibration during operation increases costs, complicates transmission systems, and consumes additional energy. When considering the millions of such components used in data centres and AI clusters, this stability becomes especially significant.
The modulator is capable of reaching data speeds exceeding 400 gigabits per second. This is roughly equivalent to transmitting about 80,000 HD video streams at once (assuming 5 megabits per second per stream) or sending eight full HD movies simultaneously.
“We are working at the limits of what is technically possible today. With more powerful control electronics, we could even increase the data rates,” explained Kotz. “Fast, economical, reliable, and manufacturable on an industrial scale — this combination makes the technology attractive, especially for data centers and AI clusters that are already suffering from bottlenecks in data exchange between processors,” added Koos.
