AI computing sees advancement with ecofriendly light-based computer chip

(CN) — A silicon-photonic chip that leverages light rather than electricity would alter the speed and sustainability of computing technologies essential for training artificial intelligence, researchers with the University of Pennsylvania’s School of Engineering and Applied Science hope.

The inception of this innovation arose from a discussion over a cup of tea between Penn professors Nader Engheta — known for his expertise in nanoscale light manipulation — and Firooz Aflatouni, who specializes in silicon-based microdevices. The chip, they say in a study published Friday in Nature Photonics, may lead to more efficient and environmentally-friendly data processing methods.

They aimed to tackle the challenge of efficient vector-matrix multiplication, a key operation in training neural networks.

“We’re on the brink of a paradigm shift towards optical computing, focusing on creating systems that are both flexible and programmable,” Engheta said in an interview.

The core innovation of their silicon-photonic chip lies in its use of photons, or light particles, for data processing, which promises significant advancements in computing speeds and a drastic reduction in energy consumption.

“Transitioning from electron to photon-based computation not only accelerates data processing but also substantially lowers energy demands, paving the way for greener computing solutions,” Aflatouni said.

Jiaqi (Jackey) Gong, an artificial intelligence expert from the University of Alabama, also pointed out the significant environmental benefits of this innovation, especially in reducing the energy consumption associated with AI’s training processes.

“By utilizing light for computations, we can achieve a substantial reduction in the energy required for training AI models,” Gong said.

While this technology holds vast potential across various domains — including artificial intelligence system training and general computational tasks — it also faces hurdles in mass production and seamless integration with existing systems. Gong also cautioned about the challenges ahead, particularly in demonstrating the technology’s reliability and scalability to encourage industry adoption.

“Ensuring the reliability and scalability of this technology is crucial for its integration into current artificial intelligence infrastructures and its acceptance in the broader industry,” he said.

Aflatouni and Engheta echoed this, describing the complex journey from concept to realization, which required the duo to meet the rigorous standards of commercial chip production while striving to maintain the integrity of their computational objectives.

“One of the most challenging aspects was aligning our innovative goals with the manufacturing constraints without compromising on performance,” Aflatouni said in an interview.

The pair also said the new chip technology could allay privacy concerns — without the need to store information in a computer’s memory since the chip can perform many computations at once, silicon-photonic chip computers would be less vulnerable virtual attacks.

“No one can hack into a non-existing memory to access your information,” Aflatouni said in a statement.

The collaborative effort between Engheta, Aflatouni, and their team reflects a comprehensive approach to technological innovation, intertwining innovative research with a commitment to sustainability.

“Our work is not just about advancing computational capabilities; it’s about fostering a sustainable future through technological innovation,” said Engheta.

Follow @MarkHebert100