Beyond Moore’s Law: How Light-Based Computing is Redefining the Future of AI

Lightmatter’s breakthrough in photonic computing overcomes the limits of Moore’s Law by using light to enable faster, more efficient data processing, paving the way for scalable AI and the future of artificial general intelligence (AGI).

For decades, Moore’s Law was the cornerstone of progress in computing. The idea that the number of transistors on a microchip doubles every two years led to exponential growth in computing power. This relentless miniaturisation fuelled the digital revolution, making everything from smartphones to global internet connectivity possible. But that path of silicon-based electronics has hit a fundamental limit: transistors are now approaching atomic scales, where quantum uncertainty and thermodynamic challenges prevent further shrinking.

The world’s answer? Make computers bigger. As ambitions for artificial intelligence (AI) soar, the need for computational power is outpacing what traditional silicon-based technologies can deliver. Enter photonics, a field poised to revolutionise computing by harnessing the speed of light.

Leading this cutting-edge movement is Lightmatter, a company at the forefront of integrating photonics into computer chips. Their approach promises to unlock unprecedented computational power, potentially bringing the dream of artificial general intelligence (AGI) within reach.

The Limits of Traditional Computing

Since the 1960s, fiber optics — ultra-thin strands of glass designed to guide and trap light — have fundamentally transformed how we communicate. By enabling high-speed data transfer over immense distances with minimal signal loss, fiber optics laid the foundation for the modern internet, which relies on an intricate global network of undersea cables spanning continents and oceans. This infrastructure supports everything from video streaming to real-time communication, showcasing our ability to transmit vast amounts of data using light.

However, while we’ve mastered using light for communication, harnessing light for computation has proven far more challenging. Unlike electrical signals, which can be easily controlled within the confined pathways of microchips, photons (light particles) have a natural tendency to scatter, spread out, and resist making sharp turns. These fundamental properties of light make it difficult to guide photons through the intricate and compact circuits required for computational tasks.

For decades, optical computing — the idea of using light instead of electrons to process data — has been an enticing vision due to light’s inherent advantages. Photons travel at vastly higher speeds, generate less heat, and consume less energy compared to electrons. In theory, optical computers could perform calculations faster and more efficiently than their electronic counterparts. Yet the reality of manipulating photons within the microscopic confines of a chip presented immense technical obstacles. Precisely controlling light in such tiny spaces — forcing it to follow exact paths and interact predictably — seemed nearly impossible with the technology available at the time.

Now, thanks to advancements in silicon photonics and innovative companies like Lightmatter, these longstanding barriers are being overcome. By integrating photonic components directly into chips, scientists are finally unlocking the potential of light for computation, heralding a new era where data can be processed and transmitted at the speed of light. This breakthrough offers the possibility of unprecedented computing power, efficiency, and scalability — essential for tackling the demands of artificial intelligence and beyond.

The Photonics Revolution

Lightmatter has made photonics commercially viable by integrating photonic components directly into computer chips. Unlike traditional chips that rely on electrons, Lightmatter’s technology uses waveguides—microscopic pathways that guide light through a chip—to transmit data. This breakthrough enables data to travel faster, over longer distances, and with far less energy loss compared to conventional electronic circuits.

Nick Harris, CEO and co-founder of Lightmatter, explains the impact of this innovation: “At Lightmatter, we’re building AGI supercomputers that can think at the caliber of the best human minds. By integrating photonics into chips, we’re redefining what’s possible in computing.”

How Lightmatter’s Technology Works

Lightmatter’s approach centers on two groundbreaking products: Passage and Envise.

Passage: A Superhighway for Data

Passage, developed in 2019, is a photonic interconnect platform. It consists of a 200mm by 200mm chip with 48 interconnected tiles using optical waveguides. These waveguides allow GPUs, CPUs, and switches to communicate using light, dramatically improving speed and efficiency. The latest iteration of Passage supports over 100 terabits per second of data transfer and can connect 256 optical fibers—the highest bandwidth and fiber count in the world.

This technology solves the “shoreline problem” faced by traditional chips. As chip designs grow, there’s limited physical space to plug in electrical wires for data transfer. Photonics sidesteps this bottleneck by enabling data to escape through waveguides embedded within the chip, massively increasing potential bandwidth.

Envise: Photonic Computing at the Speed of Light

Envise is Lightmatter’s all-in-one photonic computing chip. It merges traditional electronics with photonics to perform computations at light speed. Unlike conventional chips constrained by electrical resistance and capacitance, Envise leverages light’s properties to achieve near-instant data processing.

This means data can flow through the chip using different colors of light, allowing simultaneous processing of multiple tasks. The result is faster, more efficient computation with far less power consumption—critical for scaling AI models that demand massive computational resources.

Sim Janowski, Lightmatter’s CFO and a former Nvidia executive, emphasises the transformative potential: “The combination of our technology and team made it clear that Lightmatter is set to revolutionise computing. The energy efficiency and speed of photonic chips are game-changers.”

Why the Future of AI Depends on Photonics

AI development, particularly in large language models and potential AGI, relies on supercomputers with thousands of interconnected chips. The sheer volume of data transfer needed for training these models creates a networking bottleneck. Traditional electronic interconnects struggle to keep up, causing delays and inefficiencies.

Lightmatter’s photonic solutions enable supercomputers to function as a single massive chip. This seamless integration reduces latency, increases bandwidth, and allows for faster training of complex AI models.

Steve Klinger, VP of Product at Lightmatter, explains: “The biggest challenge for supercomputers today is the connectivity between GPUs. We’re solving that by allowing bandwidth to escape anywhere on the chip, creating dramatic improvements in communication speed and efficiency.”

Scaling Beyond Moore’s Law

As transistors hit physical limits, Lightmatter’s innovations offer a way to continue scaling computational power. By integrating photonics into existing systems, Lightmatter provides a bridge to the future without requiring a complete overhaul of current infrastructure. This approach ensures rapid adoption and ongoing progress in computing capabilities.

Ritesh Jain, Senior VP at Lightmatter, highlights the company’s integrated approach: “Our end-to-end capabilities, from silicon engineering to system validation, allow us to rapidly prototype and innovate. This vertical integration is key to pushing photonics to new heights.”

A New Era of Computing

Moore’s Law may have reached its end, but the future of computing is far from bleak. Photonics opens a new frontier where data moves at the speed of light, energy efficiency soars, and the potential for AGI becomes increasingly tangible.

As Lightmatter continues to push the boundaries of what’s possible, the dawn of a new computing age is here—one where progress is limited not by the size of transistors but by the speed of light itself.

The death of Moore’s Law doesn’t mark the end of innovation. It marks the beginning of a brighter, faster, and infinitely scalable future.