Xscape Photonics Develops Multicolor Lasers to Boost AI Datacenters

Silicon Valley-based Xscape Photonics is developing multicolor laser technology to revolutionize data transmission between GPUs, AI chips, and memory hardware in datacenters. Traditional interconnects, which use metal wires to transmit data, are constrained by bandwidth limitations, high power consumption, and heat generation. Xscape aims to address these challenges by leveraging silicon photonics, a cutting-edge technology that uses light to transmit data more efficiently.

The Power of Silicon Photonics

Unlike traditional interconnects, silicon photonics allows data to be transmitted with minimal power and heat, while enabling higher bandwidth. In datacenters that use fiber-optic links between components, electrical data must be converted to optical signals and then back again, which adds latency to the process.

Vivek Raghunathan, CEO and co-founder of Xscape, explained that the company’s technology offers a solution to the limited bandwidth and power issues in today’s datacenters. The use of optics-on-chip through silicon photonics can eliminate the need for electrical data conversion, reducing latency and improving overall performance.

Xscape’s first product, a programmable laser for datacenter fiber-optic interconnects, can transmit multiple data streams simultaneously by using different wavelengths of light, or colors. This multicolor approach prevents interference between data streams, allowing for smoother communication among GPUs, AI chips, and memory systems.

Silicon Photonics’ Roots and Xscape’s Origins

Xscape originated in a Columbia University lab, where professors Alexander Gaeta, Keren Bergman, and Michal Lipson developed the underlying technique for using light to transmit terabytes of data. The company was officially spun off in 2022 and soon attracted talent from industry giants like Broadcom and Intel, including Raghunathan and Yoshitomo Okawachi, a laser engineer.

Xscape’s lasers offer a unique advantage over traditional systems, using the same fabrication processes as existing microelectronics, which simplifies scaling. While the first-generation laser can emit between 4 and 16 colors, Xscape is already planning an improved version capable of transmitting up to 128 colors.

Competing in the Silicon Photonics Market

Xscape faces stiff competition in the multi-billion-dollar silicon photonics market, which is dominated by players like Intel, Ayar Labs, and Celestial AI. Intel reports that it has shipped more than 8 billion photonics chips and 3.2 million on-chip lasers since 2016. However, Xscape’s approach has attracted interest from major industry names like Cisco and Nvidia, both of which participated in Xscape’s recent $44 million Series A funding round. Although these investments aren’t strategic, Cisco’s role as a leader in optical networking components signals trust in Xscape’s potential.

The funding will help Xscape scale up production of its lasers and expand its 24-person team. The startup is currently working with ten potential customers, including vendors and hyperscalers, to deploy its technology and address the bandwidth limitations in modern datacenters.

What This Means for the AI Industry

As AI workloads become more complex, the demand for faster and more efficient datacenter interconnects continues to grow. Xscape’s multicolor lasers represent a major leap in addressing the bandwidth bottlenecks that limit AI training performance. By enhancing data transfer between chips and minimizing power consumption, Xscape’s silicon photonics technology could significantly improve the performance of AI infrastructure, paving the way for next-generation AI applications.