Two new photonic computer chip components developed in Singapore and the United States could be the answer to growing computing demands as artificial intelligence (AI) technology advances.
The new photonic computing chips, created by two different teams, are described in two papers published in the journal Nature that analyze the performance of these integrated silicon electronics.
One study, led by Singapore-based Lightelligence, presents a photonic accelerator, the other, led by US-based Lightmatter, describes a photonic processor that can effectively run AI models with high accuracy.
Photonic computer chips combine the use of light and electricity to increase computing performance and reduce energy consumption, and these
The increasing complexity of artificial intelligence and deep learning models is pushing conventional electronic computing to its limits and increasing energy demand.
Multiplication and accumulation, which are fundamental computational operations for artificial intelligence, may find a possible solution in photonic computing, which uses photons instead of electrons, the magazine says.
Photonic computing has been studied for decades, but these demonstrations could mean we are finally "on the cusp of harnessing the power of light to build more powerful and energy-efficient computing systems," according to engineer Anthony Rizzo of Dartmouth University in the United States.
Rizzo, who was not involved in any of the studies, published an article in the journal Nature in which he explains and assesses the importance of this research.
Both studies describe electronic-photonic computing systems "with performance metrics comparable to, and in some cases superior to, purely electronic processors in real-world applications", which constitutes "a significant leap forward towards finally capitalizing on the promise of photonic computing".
The first team developed a photonic accelerator called PACE, which enables very low latency (delay) computing, a measure of computational speed important for real-time processing.
The accelerator, consisting of more than 16,000 photonic components on a die, enables high-speed computing (up to 1 GHz) and demonstrates up to a 500-fold reduction in minimum latency compared to other smaller-scale circuits.
PACE solves difficult computational problems known as Ising problems, demonstrating the system's viability for real-world applications, Nature summarizes.
The second study describes a four-die photonic processor that can effectively run high-precision AI models, such as the BERT natural language processing model and the ResNet neural network (used for image processing), with accuracies similar to those of conventional electronic processors.
The authors show that their photonic processor has a variety of applications, such as generating texts similar to William Shakespeare, accurately classifying movie reviews, and playing classic Atari computer games such as Pac-Man.
Both teams indicate that their systems are scalable, although further optimization is needed.
No comments:
Post a Comment