IBM has introduced a fibre-optic module that allows computer chips to communicate using light, replacing traditional electrical signals. This innovation increases data transfer speeds 80 times over standard methods and is set to reshape how artificial intelligence systems are trained.
Currently, chips mostly rely on copper wires to send data, which can slow down AI training processes. With fibre optics, chips can exchange much larger volumes of information quickly. The new module uses an optical waveguide, packing 51 fibres per millimetre, a design that avoids signal interference and maximises efficiency.
This design is particularly useful as it enables six times more optical connections than existing methods. As a result, chips can handle larger AI models without long delays or high energy consumption.
How Will This Technology Affect AI Training?
The new fibre-optic module has the potential to reduce the training time for large AI models from three months to three weeks. This is not only faster but also much more energy-efficient, lowering the energy costs of data centres up to fivefold.
Stress tests showed the module works under extreme conditions, including temperatures from -40°C to 125°C. Its resilience makes it a strong candidate for widespread use in industries requiring high-speed computing.
Leading semiconductor companies are expected to take an interest in this technology. Its a good option when it comes to improving the performance of AI applications across different sectors, from healthcare to autonomous vehicles. “We’re really in the early days of all this, but it’s the hottest area in semiconductor technology right now in terms of high-performance computing and AI technology,” said Dan Hutcheson from TechInsights.
How Does MIT’s Photonic Chip Advance AI Computing?
Meanwhile at MIT, researchers have created a photonic chip that processes AI tasks using light rather than electricity. This chip completes tasks at ultra-fast speeds, achieving accuracy above 92%, which matches current electronic hardware but with greater speed and efficiency.
Unlike older designs, this chip handles both linear and nonlinear computations internally. It eliminates the need for external electronic components, which previously slowed down processing.
What Makes It Work So Well?
The chip uses Nonlinear Optical Function Units to carry out complex tasks directly on the device. Data is processed as light, passing through programmable beamsplitters and NOFUs to handle tasks seamlessly. This design avoids converting optical signals into electrical ones, which maintains its speed and uses very little energy.
Built using commercial techniques that are always available, the chip can be manufactured at scale and easily integrated into existing systems. Its design keeps energy use low while achieving high accuracy and rapid processing times.
Where Could This Chip Be Used?
This technology is well-suited for applications like lidar, real-time navigation, and telecommunications, where speed is needed. It also opens doors for scientific research, such as astronomy or particle physics, where large amounts of data need to be processed quickly.
The design prioritises both speed and practicality, showing promise for industries that need reliable and energy-conscious solutions. Researchers are already considering how to further scale this technology and adapt it for even more specialised uses.
