Photo credit: www.sciencenews.org
It’s an exciting time in the field of computing, as advancements in laser technology are poised to revolutionize how processing is done. Two tech firms, Lightelligence from Boston and Lightmatter located in Mountain View, California, have introduced groundbreaking computer components that harness laser light for data processing. These cutting-edge processors promise to tackle certain real-world challenges more rapidly and with less energy consumption compared to conventional electronic computers. Their respective findings were detailed in separate publications on April 9 in the journal Nature, marking a significant milestone in the evolution of computing methodologies.
Anthony Rizzo, a photonics engineer at Dartmouth College, commented on the implications of these innovations, asserting that photonic components can outperform traditional electronic chips in meaningful ways. “They can do things that we care about, and that they can do them better than electronic chips that we already have,” he noted.
Photonics technology is already integral to global data transmission via fiber optic networks, and it plays a key role in transferring data in sophisticated data center environments. For instance, NVIDIA, a major player in tech, recently showcased technology that employs light for inter-device communication. However, as Rizzo points out, in traditional computer systems, light signals are merely converted into slower electronic signals (the familiar binary 1s and 0s), which travel through tiny transistors.
Lightmatter and Lightelligence’s devices stand in stark contrast, as they actually utilize light to perform calculations such as matrix multiplication—a critical operation in artificial intelligence (AI) processing and beyond. While these devices use light for specific calculations, they also incorporate electronic components for other tasks.
The timing of these announcements is especially significant as AI models continue to expand in size and complexity, while advancements in traditional chip technologies appear to be stalling. The historic trend known as Moore’s law, which suggested that the number of transistors on a chip would double approximately every two years, is reaching its limits. Nick Harris, CEO of Lightmatter, acknowledges that the fundamental physics governing the flow of electricity through transistors limits their further miniaturization, suggesting that traditional electronic computing might not improve significantly moving forward. Photonic computing is positioned as a transformative alternative.
Both companies’ innovations feature an array of photonic chips capable of performing calculations with light. For example, Lightmatter’s system manipulates over 200,000 components to handle 512 light beams, enabling efficient matrix multiplication more quickly than typical electronic chips.
On the other hand, Lightelligence’s device, named PACE, synergizes both photonic and electronic chips to optimize computations for various sectors including finance, manufacturing, and logistics. In contrast, Lightmatter’s more adaptable processor integrates four light-based chips with two electronic ones, showcasing its capability to run mainstream AI applications, from large language models like those informing ChatGPT to deep learning algorithms that even play classic Atari games.
Harris noted the significance of these developments, stating, “That’s never been done using any sort of alternative computer processing technology.” Despite the progress, past attempts to create photonic processors struggled with real-world applications and accuracy. Light signals inherently possess a diverse range of values, posing a risk that minor transmission errors could cascade into significant miscalculations.
Both companies have addressed these challenges in their designs. Lightelligence’s technology leverages randomness in optimization problems to enhance solution exploration, while Lightmatter employs electronic chips to manage data flow and minimize errors more effectively.
“We’re not presenting just a lab prototype,” Harris asserts. “This is a new type of computer. And it’s here.” The photonic components that both companies utilize can be produced using existing manufacturing processes, indicating a ready path for scaling up these technologies. Experts believe we could see these innovations integrated into practical systems, such as data centers, in the near future, potentially within the next five years.
Source
www.sciencenews.org