Preface:

In the narrative of information technology, few technologies have experienced prolonged [neglect] like silicon photonics, only to suddenly emerge as an indispensable foundational capability at a critical moment.

Silicon photonics has undergone nearly forty years of gradual accumulation: from conceptualization and physical validation to engineering breakthroughs and industrial implementation. This trajectory exemplifies a classic path of [long-term technological commitment].

Chapter 1: From Theoretical Frontier to Industrial Pinnacle

Every key breakthrough in the history of silicon photonics technology has been inseparable from the continuous maturation of the semiconductor industry ecosystem.

① 1985-1999: Theoretical Foundations – Solitary Exploration in the Wilderness

At the end of the 20th century, the information industry was characterized by a starkly divided binary landscape.

Silicon-based semiconductor processes dominated the computing world, while materials systems centered on III-V compounds like indium phosphide and gallium arsenide monopolized the telecommunications sector.

These two domains operated like parallel lines that never intersected, each advancing along its own trajectory.

From the perspective of engineering at the time, using silicon—an indirect bandgap material inherently unsuited for light emission—to process photonic signals seemed like a utopian idea that defied the material's fundamental nature.

It was against this industry consensus that Richard Soref initiated the pioneering work in silicon photonics.

In 1985, he systematically demonstrated for the first time the feasibility of single-crystal silicon as an optical waveguide material, outlining the initial theoretical framework for silicon-based integrated photonics.

In 1987, he published another landmark paper that quantitatively derived the relationship between carrier concentration variations and silicon's refractive index and absorption coefficient.

This discovery provided the [fundamental principles] for humans to manipulate photons in silicon through electrical means, transforming silicon photonics devices from theoretical concepts into engineering realities.

The academic torch continued to pass. Graham Reed's team at the University of Surrey developed the first low-loss silicon waveguides, validating the feasibility of implementing basic optical circuits on silicon wafers.

In 1988, Reed's student Andrew Rickman founded Bookham Technology, the world's first silicon photonics company. This venture introduced standardized semiconductor manufacturing thinking into the field of optical devices, attempting to mass-produce optical components using CMOS process lines.

In 1990, Leigh Canham's discovery of room-temperature photoluminescence in porous silicon shattered the conventional wisdom that [silicon cannot emit light], sparking a global academic surge in silicon-based photonics research.

In 1992, the silicon-on-insulator (SOI) process achieved low-loss waveguides, firmly confining photons within submicron-scale cores.

This made possible the [dimensional collapse] of photonic devices from millimeter to submicron scales, finally establishing a suitable material platform for silicon photonics technology.

However, at this stage, silicon photonics remained a [premature infant]. These advancements failed to generate industrial momentum, as the world had no pressing need for light within chips.

Copper wires were sufficiently inexpensive, and bandwidth demands were far from reaching their limits. The technology existed, but the market did not.

② 2000-2009: Technological Breakthroughs – The Turning Point of Industry Giants' Entry

Entering the 21st century, Moore's Law encountered the impenetrable [power wall].

Processor clock speeds stalled around 3 GHz as electronic signal transmission losses and heat accumulation in copper wires became the primary bottlenecks restricting chip performance improvement.

Simultaneously, the advent of the Web 2.0 era triggered explosive exponential growth in data traffic. Hyperscale data centers began to emerge, and the interconnectivity demands between servers quietly approached the physical limits of electrical signals.

In 2004, Mario Paniccia, director of Intel's Photonics Technology Lab, led his team to publish the world's first silicon-based optical modulator with bandwidth exceeding 1 Gbps in the journal Nature.

The core value of this achievement lay in its utilization of mature CMOS processes to achieve high-speed optical wave modulation through electric field-induced carrier accumulation.

This proved that silicon could not only carry optical signals but also precisely and rapidly modulate light, functioning like an electronic switch.

From this point onward, silicon photonics transitioned from an academic outlier to a strategic priority for global chip giants.

The final critical piece of the silicon photonics puzzle—[the light source challenge]—was resolved two years later.

As an indirect bandgap material, silicon exhibits extremely low spontaneous emission efficiency, making efficient light emission inherently difficult for pure silicon.

In 2006, Professor John Bowers at the University of California, Santa Barbara, collaborated with Intel to develop a hybrid integrated silicon-based laser using low-temperature plasma-assisted wafer bonding technology. This approach tightly integrated indium phosphide material with an SOI substrate at the atomic scale.

This [heterogeneous integration] solution assigned light emission to III-V materials while leveraging the silicon platform for optical transmission.

It perfectly complemented the strengths of both material systems, marking silicon photonics' complete transition from passive to active optical emission capabilities.

On the sidelines of technological breakthroughs, the spark of commercialization was quietly ignited.

Founded in 2001, Luxtera pioneered the concept of [integrated optoelectronics on silicon], attempting to manufacture optical modulators, photodetectors, and CMOS driver circuits on the same SOI chip to achieve monolithic integration of photonic and electronic components.

Silicon photonics began shifting from a [scientific challenge] to an [engineering problem]. The industry reached a consensus: the core advantage of silicon photonics lay not in performance limits but in manufacturing system compatibility.

Photons could be [manufactured] like transistors, laying the foundation for subsequent large-scale production.

③ 2010-2019: Cloud Computing Boom – The Golden Era of Silicon Photonics

By 2010, silicon photonics technology finally encountered the [perfect storm] it had awaited in the wilderness for two decades: the comprehensive rise of hyperscale data centers.

The aggressive expansion of Amazon AWS, Microsoft Azure, and Google Cloud fundamentally transformed internet traffic patterns.

In data centers housing hundreds of thousands of servers, traditional copper wires were limited to transmission distances of just a few meters at 100G speeds.

Meanwhile, conventional III-V discrete optical modules proved prohibitively expensive for mass deployment due to their manual assembly processes.

Around 2016, Intel introduced its 100G PSM4 silicon photonics module, developed over more than a decade. By fully integrating hybrid silicon-based lasers, high-speed modulators, and CMOS driver circuits, this solution achieved dramatic cost reductions through economies of scale.

This breakthrough directly shattered the price barrier for widespread optical interconnect adoption in data centers.

Within a few years, Intel shipped millions of silicon photonics modules, establishing itself as the undisputed leader in driving the global 100G transition in data centers.

During this period, the mature [foundry + fabless] model from the electronic semiconductor industry was successfully adapted to silicon photonics.

Photonic chip designers no longer needed to build billion-dollar wafer fabs. Instead, they could leverage standard component libraries to transform designs into physical chips.

This [technological democratization] movement spawned numerous silicon photonics innovators like Acacia Communications.

These companies combined silicon photonics with coherent communication technologies, shrinking refrigerator-sized coherent optical transponders to pluggable module form factors. This breakthrough completely opened up the long-haul transmission market for silicon photonics.

Traditional network equipment giants also embarked on aggressive merger and acquisition strategies during this era.

The industry's underlying consensus became clear: as switch chip throughput climbed toward 12.8T and 25.6T, the panel density and power consumption bottlenecks of traditional pluggable optical modules would prove insurmountable.

This meant silicon photonics wasn't merely superior—it was more scalable, cost-effective, and suitable for large-scale deployment.

④ 2020-Present: AI Era – Evolution from Pluggable to Co-Packaged Optics

Training and inference for large AI models require interconnectivity among GPU clusters with tens of thousands of cards, driving bandwidth demands rapidly from 400G toward 800G, 1.6T, and even 3.2T.

While traditional electrically absorbed modulated lasers (EMLs) saw diminishing returns in power efficiency and cost as speeds increased, the comprehensive advantages of silicon photonics became fully apparent.

Compared to traditional EML solutions, silicon photonics modules reduce costs by 20-30%, cut power consumption by nearly 40%, and shrink volume by over 30%.

These benefits perfectly matched AI computing clusters' core requirements for high-density, low-power, high-bandwidth interconnectivity.

During this phase, Chinese manufacturers achieved a leap from followers to leaders, with companies like InnoLight, Accelink, and YOFC securing critical breakthroughs.

From 2025 to 2026, the silicon photonics industry reached a new inflection point as co-packaged optics (CPO) technology transitioned from research to commercial scale.

Compared to traditional pluggable solutions, CPO systems reduce power consumption by over 70% while increasing bandwidth density eightfold.

At GTC 2026, NVIDIA announced full-scale production of its Spectrum-X switch based on CPO technology, marking silicon photonics' complete transformation from [scientific speculation] to [computing infrastructure].

Its development journey has been deeply intertwined with the maturation of semiconductor processes, internet traffic growth, and AI computing explosions.

Chapter 2: Silicon Photonics' Trillion-Dollar Application Landscape

While data centers and AI computing clusters currently represent the core application market for silicon photonics, they are far from its final destination.

① Core Foundation: AI Data Centers and Supercomputing Interconnects

Silicon photonics has emerged as the cornerstone interconnect solution for next-generation exascale and zettascale supercomputers.

The world's top supercomputing centers are exploring silicon photonics to achieve low-latency, high-bandwidth connections between computing nodes, addressing the persistent [memory wall] and [communication wall] challenges in supercomputing systems.

U.S. Department of Energy laboratories are already conducting research on all-optical interconnect supercomputing architectures based on silicon photonics. By 2030, over 60% of the world's Top500 supercomputers are projected to adopt silicon photonics interconnects.

② Second Growth Engine: Telecom Networks and 6G Communications

Silicon photonics coherent technology has achieved widespread commercialization. Silicon-based coherent optical modules from vendors like Acacia and Cisco have miniaturized rack-mounted coherent transmission equipment into pluggable modules, reducing costs by over 50% and power consumption by 40%.

For 6G networks, silicon photonics has become a core enabling technology.

6G demands terahertz-class transmission rates and sub-millisecond end-to-end latency—requirements unattainable with traditional electrical interconnects.

With its high speed, low power consumption, and high integration capabilities, silicon photonics will serve as the foundational technology for 6G base stations, core networks, and terahertz communication systems.

Yole predicts that by 2030, the telecom market will account for 30% of the total silicon photonics market, becoming one of the industry's most important growth engines.

③ Emerging Explosive Application: Automotive LiDAR and Autonomous Driving

Silicon photonics enables all-solid-state FMCW LiDAR solutions that integrate all functions—emission, reception, scanning, and signal processing—onto a single millimeter-scale silicon photonic chip.

With no moving parts, reliability improves dramatically while service life extends more than tenfold compared to traditional mechanical LiDAR.

Mass production through CMOS processes can reduce costs to less than 10% of conventional LiDAR solutions.

Currently, overseas vendors like Aeva and Voyant Photonics have introduced all-solid-state 4D LiDAR based on silicon photonics, while domestic companies like Lumotive and Moore Light have achieved volume production of silicon photonics FMCW LiDAR.

Yole projects the silicon photonics market for automotive LiDAR to exceed $1.5 billion by 2030.

④ Frontier Applications: Medical Sensing, Quantum Computing, and Optical Computing

Silicon photonics-based biosensors enable non-invasive, real-time monitoring of physiological indicators like blood glucose, lipids, and heart rate.

Consumer electronics giants like Apple are developing non-invasive blood glucose monitoring solutions based on silicon photonics, with potential integration into wearable devices like smartwatches.

Silicon photonics has become the cornerstone for scalable optical quantum computing.

The technology can integrate all components—single-photon sources, optical quantum gates, and photon detectors—onto a single silicon chip, enabling miniaturization, scalability, and mass production of optical quantum systems.

Companies like PsiQuantum and Hefei Silicon Quantum have developed quantifiable optical quantum computing platforms based on silicon photonics.

Looking further ahead, silicon photonics will drive the development of optical computing.

Photonic computing offers inherent advantages over electronic computing—higher parallelism, lower power consumption, and faster speeds—with transformative potential for AI inference, cryptography, and scientific computing.

The maturation of silicon photonics makes large-scale photonic integrated circuit production feasible, paving the way for [integrated transmission and computing] all-optical chips that could revolutionize existing computing architectures.

Chapter 3: Global Silicon Photonics Industry Landscape and Competition

The immense potential of silicon photonics has attracted aggressive layout (strategic positioning) from global tech giants, creating an industry landscape where [international leaders set the pace and Chinese vendors rapidly catch up].

Different companies have adopted varying technical routes and strategies based on their resources, engaging in fierce global competition.

Intel acquired Acacia Communications to strengthen its coherent communication capabilities and solidify its market leadership.

Its core strategy leverages silicon photonics' full-stack capabilities to provide [computing + interconnect] integrated solutions for data center customers, creating differentiation against NVIDIA.

AMD rapidly closed technical gaps through acquisitions of silicon photonics startups like Enosemi.

Its silicon photonics solutions primarily target MI-series GPUs and EPYC processors for supercomputing and AI cluster markets.

Cisco acquired both Luxtera and Acacia, gaining core technologies for monolithic silicon photonics integration and coherent communication. This created a complete [chip-module-device-system] solution.

Cisco's strategy deeply integrates its Silicon One switching chips with Acacia's silicon photonics technology to build an open AI networking stack.

In the AI networking era, this approach prevents marginalization by white-box switches and independent optical module vendors while regaining pricing power and market influence.

As AI computing demand exploded, NVIDIA became the world's largest purchaser of high-speed optical modules while aggressively deploying silicon photonics and CPO solutions.

At GTC 2026, NVIDIA announced volume production of its Spectrum-X CPO switch.

The company is also collaborating with TSMC on 3D stacking packaging that integrates GPUs with silicon photonics engines, potentially bringing optical interconnects directly into GPU chips.

By defining AI cluster interconnect standards, NVIDIA directly shapes silicon photonics' evolutionary path, establishing itself as one of the industry's most influential players.

By late 2025, GlobalFoundries completed its acquisition of Singapore-based silicon photonics firm AMF, rapidly upgrading AMF's 200mm production line to 300mm.

Combining this with its existing Fotonix silicon photonics platform, GF created the industry's first fully integrated silicon photonics manufacturing platform, establishing unique technical barriers in RF-silicon photonics integration.

Broadcom and Marvell focus on chip-level solutions, leveraging their market dominance in switching chips to deeply deploy CPO technology.

Broadcom has introduced a 51.2Tbps CPO switch chip and collaborates closely with cloud providers like Google and Meta to drive CPO adoption.

These vendors' core strategy tightly binds silicon photonics engines with switching chips to provide chip-level optoelectronic integration solutions, gaining first-mover advantage in next-generation interconnect technology competition.

Chinese manufacturers are rapidly making breakthroughs in the upstream sectors of the industrial chain, such as chip design and wafer fabrication, forming a development pattern of [tiered competition and comprehensive catch-up].

Innolight collaborates with manufacturers such as Broadcom and NVIDIA to develop next-generation optoelectronic integration solutions, continuously consolidating its technological leadership.

Eoptolink Technology has quickly enhanced its self-research capabilities in silicon photonics chips by acquiring Alpine Optoelectronics, a U.S.-based silicon photonics startup, forming a full-chain layout of [silicon photonics chips + optical engines + optical modules].

Cambridge Industries Group has achieved rapid growth in the high-speed silicon photonics module market through its partnerships with Microsoft and Meta.

Conclusion:

Looking back over the past forty years, the breakthrough of silicon photonics technology from 0 to 1 holds its true significance not in replacing copper wires but in transforming the computing paradigm.

When light enters the chip, the way information flows is rewritten, and this is the true meaning of [optoelectronic integration on a single chip].

Partial references: Smart Sensor Network's 'Silicon Photonics Technology (Part 1): Technological Evolution and Industrial Foundation,' Huang Danian Chasiwu's 'GTC2026 Observations | Standing on the Eve of CPO's Breakthrough, Reviewing the Forty-Year History of Silicon Photonics Technology Development,' and XinLianHui's 'IMEC: The Bottleneck of Silicon Photonics Is No Longer About [Whether It Can Be Made].'