Two studies show significant progress in photonic microchips. (Image credit: 3dartists via Shutterstock)
Electronic microchips are the backbone of the modern world. They are present in our laptops, mobile phones, cars and home appliances. Over the years, manufacturers have strived to make them more powerful and efficient, which improves the performance of our electronic devices.
However, this trend is currently waning due to increased costs and complexity in chip manufacturing, as well as performance limitations imposed by the laws of physics. This comes at a time when there is a need for more computing power due to the growing interest in artificial intelligence (AI).
An alternative to traditional electronic microchips are photonic chips. They use light instead of electricity to achieve greater performance. However, photonic chips have not yet become widespread due to a number of obstacles. Now, two papers published in the journal Nature discuss some of these obstacles and suggest the necessary steps to achieve the computing power required for complex AI systems.
By using light (photons) instead of electricity (electrons) to transmit and process information, photonic computing promises faster speeds and greater throughput with increased efficiency. This is because it is not subject to electrical current loss due to a phenomenon known as resistance, nor to unwanted heat loss from electrical components.
Photonic computing is also particularly well suited to performing so-called matrix multiplications—mathematical operations that are fundamental to AI.
Here are some of their advantages. However, the challenges are not trivial. Previously, the performance of photonic chips was usually studied in isolation. However, due to the prevalence of electronics in modern technology, photonic equipment will need to be integrated with these electronic systems.
However, converting photons into electrical signals can slow down processing time because light operates at higher speeds. In addition, photonic computing relies on analog operations rather than digital ones. This can reduce accuracy and limit the types of computational tasks that can be performed.
It is also difficult to scale them up from small prototypes, as large-scale photonic circuits cannot currently be manufactured with sufficient precision. Photonic computing will require its own software and algorithms, complicating issues of integration and compatibility with other technologies.
Two new papers in Nature discuss many of these obstacles. Bo Peng of Singapore-based Lightelligence and his colleagues present a new type of processor for photonic computing called the Photonic Arithmetic Computing Engine (Pace). This processor has low latency, meaning that there is minimal delay between an input or command and the corresponding response or action by the computer.
The large-scale Pace processor, with more than 16,000 photonic components, is capable of solving complex computational problems, demonstrating the feasibility of the system for real-world applications. The processor illustrates how issues of photonic and electronic hardware integration, precision, and the need for different software and algorithms can be addressed. It also shows that the technology can be scaled.
This represents a significant step forward, despite some speed limitations of current hardware.
In a separate paper, Nicholas Harris of Lightmatter, a California-based company, and his colleagues describe a photonic processor that can control two AI systems with precision comparable to that of traditional electronic processors. The authors demonstrated the effectiveness of their photonic processor by generating Shakespearean text, accurately categorizing movie reviews, and playing classic Atari computer games like Pac-Man.
The platform is also potentially scalable, although in this case the limitations of the materials and technologies used limited one of the parameters of the processor speed and its overall computing capabilities.
Both teams envision that their photonic systems could be part of next-generation, scalable hardware capable of supporting AI. This would eventually make photonics viable, although further improvements would be needed. These would include the use of more efficient materials or designs.
This edited article is republished from The Conversation under a Creative Commons license.
Sourse: www.livescience.com
