Obstacles overcome... 'Light speed' for artificial intelligence

At this point, AI applications began to require huge processing power. To overcome this problem, scientists are working on photonic chips as an alternative to traditional electronics. These chips process data using light (photons) instead of electricity, and thus can offer higher speed and bandwidth.

But photonic chips have long faced serious obstacles to realization. Many of these hurdles are beginning to be overcome with two new studies published in the journal Nature. Both studies point to significant advances that could enable photonic systems to be integrated into real-world applications.

Photonic computing eliminates efficiency problems such as resistance and heat loss in electrical systems. It can also perform complex mathematical operations such as matrix multiplications, one of the basic building blocks of artificial intelligence, with high efficiency.

But photonic systems are still analog-based, which means less precision than digital electronics. The process of converting light into electricity can be slow. In addition, large-scale photonic circuits cannot be manufactured with sufficient precision and a lack of compatible software complicates the process.

To meet these challenges, a team led by Bo Peng from Singapore-based Lightelligence has developed a new photonic processor called “Pace.” Pace contains more than 16,000 photonic components and operates with low latency. Able to successfully perform complex tasks, this processor has the capacity to solve fundamental problems such as photonic-electronic integration and software compatibility.

Another important development came from Nicholas Harris and his team at the US-based Lightmatter company. This team managed to run two different artificial intelligence systems with the photonic processor they developed. Tasks such as producing Shakespearean-style text, correctly classifying movie reviews and playing Atari games such as Pac-Man were successfully completed.

Both studies show that photonic chips are scalable and could become core hardware for AI in the future, but experts stress that better materials and designs are needed to make these technologies more efficient.

These developments signal that photonic technologies are now moving beyond theory into usable hardware for AI-enabled real-world applications.