Artificial intelligence is everywhere these days, becoming unavoidable even for those who have no interest in it and people who don’t want AI on their devices. No matter how you feel about the technology, two things are undeniable: AI has led to a chip shortage that continues to drive up tech prices, and it is incredibly bad for the environment. The latter is a result of the processing power AI requires. AI data centers consume a huge amount of electricity and water to the point that simply being polite to chatbots has cost ChatGPT tens of millions of dollars.
One new development has the potential to improve AI’s energy efficiency, though: photonic chips. Unlike the current chips, which are powered by electrons, photonic chips use photons (particles of light) for computing. In other words, they perform AI calculations using light instead of electricity, consuming fewer natural resources.
Photonic chips aren’t just better for the environment; they’re also more powerful, working even faster than existing GPUs. As our increasingly online and AI-powered lives cause data centers to use more energy, photonic chips could lessen the technology’s negative impact on the environment while also being an overall improvement.
Photonic chips make AI more energy efficient
Before diving further into what makes photonic chips better for the environment, we should first explain how the traditional silicon ones work. A tiny microchip contains billions of transistors to turn an electrical current on or off, essentially pushing electrons through circuits. That causes friction, which generates heat and requires data centers to have complex cooling systems that use a lot of water.
Photonic chips replace those transistors with photonic neurons that guide light through the chip for power. Granted, they aren’t entirely free from electricity, as they work within electronic devices. What they do is convert electrical signals to light, do the necessary calculations, then turn it back to an electrical system that returns to the computer.
Additionally, one path of light can hold multiple channels (something electrical currents cannot do), meaning these chips are capable of simultaneously processing values to use less energy than individual processing would. That uses less energy than individual processing. On top of that, light doesn’t face resistance the same way electricity does, so photonic chips can perform the same tasks without generating lots of heat, in turn using less water.
Light-powered chips are faster and smarter
In addition to being more energy efficient, photonic chips are more powerful than ones currently used. The Chinese scientists behind LightGen shared in the journal Science that their prototype was “more than two orders of magnitude greater than those of state-of-the-art electronic chips” in terms of computing speed and energy efficiency, and German company Q.ANT claims its NPU Gen 2 is up to 30 times more energy efficient and 50 times faster than the current tech used in data centers. That’s impressive, but it shouldn’t be too surprising that light-powered chips work better since electrical friction has long been a roadblock to advances in chip technology.
Additionally, one path of light can hold multiple channels (something electrical currents cannot do), meaning these chips can simultaneously process values to use less energy than individual processing would. Even better, they’re accurate, with the University of Sydney’s prototype correctly identifying images 90–99% of the time.
Love it or hate it, AI isn’t going away anytime soon. Advancements in chip technology like protonic processing are encouraging signs that we can make AI technology and other data usage less devastating to the environment, at least, and the fact that it does so while also working more efficiently is a huge plus.
