In 2024 research, our Tech Scouting and Intelligence team discussed how photonics is “an attractive new approach to data transfer—both on- and off-chip—given its added benefits of speed, bandwidth, and low heat generation.” Since that original research was published, the SiPh market has grown rapidly and has shown the potential to transform a myriad of technologies as the AI industry demands increased compute performance at lower cost and higher efficiency.

SiPh goes beyond Moore's Law to break the efficiency wall that limits all-electronic devices.

SiPh enables the integration of optical components onto a single silicon (or silicon-based material like silicon nitride or silicon-on-insulator) chip leveraging standard complementary metal oxide semiconductor processes. It leverages existing semiconductor manufacturing techniques to simplify this evolution.  

The resulting devices use light (photons) instead of electrical signals, which overcomes many limitations of traditional copper- and electron-based devices. These photonic integrated circuits (PIC) are highly versatile and can efficiently generate, manipulate, and transmit light at high speeds and low power, making them valuable across many applications.

Energy Efficiency

Heat matters: The more heat, the more wasted energy, cooling costs, stress on components—and the less space there is to add processors and memory without the risk of melting. SiPh devices can be more energy-efficient since using light, rather than electrical currents, reduces resistive heating and power loss. Additionally, photonic interconnects can achieve high-speed data transfer with lower energy consumption by minimizing the need for power-hungry repeaters and buffers.

Signal Integrity / Interconnect Reach

Electrical interconnects lose their signal strength over increasingly long distances and suffer crosstalk and interference. On the other hand, photons don’t suffer from resistive losses, parasitic effects, or electromagnetic interference, maintaining signal integrity over long distances. Optical interconnects also reduce signal propagation delays, as photons experience less dispersion compared with electrical signals. 

Latency

Electrical interconnects are constrained by resistive-capacitive delays and require signals to be constantly boosted, converted, and cleaned up—all of which accumulates latency as data moves across a system. Photonic links avoid these bottlenecks by transmitting information optically, where waveguides experience negligible dispersion and do not require frequent regeneration. While the speed is still shaped by how fast the light signals can be turned on and off, SiPh reduces the overhead introduced by electronic signal processing, enabling lower end-to-end delay and faster communication across increasingly complex systems.

Data Rates

SiPh greatly expands the amount of information that can be moved at once, making it possible to handle the growing demands of modern computing and communications. It enables much higher data rates with wavelength division multiplexing—a technique where multiple signals are transmitted simultaneously on a single waveguide by using different wavelengths (i.e., colors) of light. As a result, while high-speed electrical interconnects have plateaued at data rates of hundreds of gigabits per second, SiPh is on track to scale up to tens of terabits per second by the end of this decade.

This technology can advance compute-intensive use cases across a number of domains.  One of the most compelling is using SiPh to accelerate the performance of AI systems.

Challenge

Today’s AI computations are fundamentally limited by the “memory wall,” which is the gap between the amount of memory access needed and the amount available. This gap is growing exponentially as models become larger and more complex, but memory bandwidth is stagnating. While data center applications can address these limits through scale-up architectures, edge devices face strict limits on size, weight, and power (SWaP), meaning they cannot simply add more chips or energy-hungry cooling systems to keep up. These combined bottlenecks create mounting pressure for new approaches to move data faster and more efficiently within AI hardware.

Application

The bandwidth, signal integrity, and power advantages of SiPh directly address the memory wall by easing the flow of data between compute and memory, reducing one of the biggest bottlenecks in AI systems. By integrating on-chip optical interconnects, photonics also makes it possible to bring more compute power closer to edge sensors, enabling real-time AI inference in devices like drones, autonomous vehicles, and industrial robots.

Beyond raw bandwidth, the ability of photonics to deliver deterministic, low-latency communication supports tight synchronization across distributed sensor networks, which is critical for autonomy and situational awareness. In this setup, data is still processed electrically but transmitted optically, with PICs embedded alongside electronic integrated circuits to maximize reach, network bandwidth, and efficiency. This co-location not only supports memory disaggregation and composable AI architectures but also scale-out and scale-up systems that break the limits of traditional copper interconnects.

Shortening Distance of Optical Interconnections Leads to Greater Information Processing Capacity

Export restrictions on lithography and advanced electronics have heavily contributed to China’s SiPh indigenous development drive. Moving forward, China and the United States will compete for supremacy in the geopolitical microelectronic and SiPh race. China is aggressively investing in SiPh as part of its broader effort to secure self-sufficiency in advanced compute, sensing, and communications technologies.

China’s supply chain dominance in germanium, lithium niobate, and other rare and critical metals strengthens their hold over the silicon wafer (SW) market, needed for SiPh, and accounts for 50% of global SW production. The top five global SW producers are all Chinese, including the Shanghai Xinkehui New Material Co. 

In 2024, private investment in SiPh grew rapidly after a year of protracted funding. The large decrease in capital allocation and stagnant deal volume in 2023 was likely due to macroeconomic factors and longer runways created from the previous two years of large, later-stage deals.

The U.S. government still maintains a majority investment in electronic-only technologies. But there are significant efforts to accelerate development of SiPh, given its heightening role in national security priorities. The Defense Advanced Research Projects Agency (DARPA) and the Defense Sciences Office are the major drivers of Department of Defense (DOD) SiPh funding, with DARPA being a top catalyst for SiPh research and development over the last two decades.

It’s no accident that SiPh has reached this inflection point. Decades of research and development by both the private and public sectors, including more than 20 years of DARPA investment, have strategically accelerated SiPh applications. Already dominating data communications in the data center market, SiPh is now on the cusp of disrupting several other key fields in the next two to five years.

Over the coming years, the SWaP-C metrics of SiPh will improve sufficiently to accelerate edge AI computations and other critical edge applications.

For long-term dominance in advanced computing, communications, space, and other technology verticals, the United States must prioritize the adoption of photonics and optoelectronics. They are the next generation of capabilities to break the limits of all-electronic solutions.

In the coming years, the federal government can leverage SiPh for:

• Secure, high-bandwidth communications that are both compact and resistant to electromagnetic interference.
• Tactical and strategic networks where low SWaP is critical, such as space environments.
• The foundational platform to manipulate quantum states of light, making them key enablers for scalable quantum systems.
• Support of next-generation sensors used in intelligence, surveillance, and reconnaissance (ISR) and electronic warfare. Applications include LiDAR, hyperspectral imaging, and quantum radio frequency sensors, among many others.

Silicon photonics will enable and transform a diverse array of industries by providing a scalable platform for manufacturing advanced devices. By marrying light waves with the ubiquitous semiconductor manufacturing used for modern electronics, SiPh technology overcomes traditional barriers in cost, performance, and scalability that have historically limited photonics adoption.

For continued progress, it will be crucial to:

• Prioritize domestic R&D and manufacturing investment into photonic and optoelectronic technologies to ensure competitive positioning in SiPh for the United States and our allies.
• Explore supply chain alternatives for key SiPh components and critical mineral resources currently controlled by China.
• Keep U.S. capital and SiPh intellectual property from China, as with other advanced semiconductor technologies crucial to the Tech Cold War.

Today’s leaders in emerging software solutions like AI must take hardware more seriously than before. Key innovations in SiPh design, materials, integration, packaging, and system-level design will unleash exponential growth in computing, sensing, quantum information technologies, telecommunications, and many more applications—some of which we have yet to discover.