stub Computing at the Speed of Light with Silicon Photonics -
Connect with us


Computing at the Speed of Light with Silicon Photonics

Updated on is not an investment adviser, and this does not constitute investment advice, financial advice, or trading advice. does not recommend that any security should be bought, sold, or held by you. Conduct your own due diligence and consult a financial adviser before making any investment decisions.

Silicon Photonics

Engineers from the University of Pennsylvania have developed a chip that utilizes light waves instead of electricity to perform intricate math necessary to train artificial intelligence (AI). This innovation could significantly speed up processing and reduce the energy use of devices.

The study, published in Nature Photonics, shows that this is an “inverse-designed low-index-contrast structure” on a silicon photonics (SiPh) platform, potentially enabling large-scale wave-based analog computing platforms.

Silicon photonics uses silicon, an abundantly available and inexpensive element used in the mass production of computer chips, integrating components like photodetectors, optical switches, optical waveguides, and optical modulators on a silicon substrate.

The silicon photonic (SiPh) chip in this study manipulates materials at the nanoscale to perform mathematical computations using light. This method of interacting light waves with matter promises to develop computers that exceed the current limitations of today's chips.

“We decided to join forces,” said H. Nedwill Ramsey Professor Nader Engheta, pointing to the development of nanoscale silicon devices by the research group of Firooz Aflatouni, who's an Associate Professor in Electrical and Systems Engineering.

The goal has been to develop a platform to perform vector-matrix multiplication (VMM), which is utilized in the development and function of neural networks powering current AI tools. 

According to the study, while inverse-designed SiPh metastructures efficiently perform analog computations with electromagnetic waves, scaling them up to manage a large number of data channels presents a challenge. To address this, the team adopted a 2D inverse-design approach to create compact amorphous lens systems that are typically feed-forward and low-resonance. The study successfully demonstrated a vector–matrix product for 2 × 2 and 3 × 3 matrices and also designed a 10 × 10 matrix.

Instead of using a silicon wafer of uniform height, the team selectively thinned the silicon in specific areas. These height variations enable control over light transmission through the chip.

By distributing these variations, the chip scatters light in specific patterns, allowing it to conduct mathematical calculations at the speed of light, the fastest possible communication method.

According to Aflatouni, this design is already set for commercial applications due to the constraints imposed by the commercial foundry that produced the chips. Also, the design can potentially be adapted for use in graphics processing units (GPUs), a specialized electronic circuit currently in huge demand in line with the AI fervor. By integrating the Silicon Photonics platform as an add-on, Aflatouni mentions, one could accelerate the training and classification processes.

However, the benefits extend beyond speed and energy efficiency, as the chip also enhances privacy. So, by allowing many computations to happen at the same time, there's no need to store sensitive information in the working memory of your computer. This makes a computer powered by such technology essentially unhackable. Aflatouni remarked:

“No one can hack into a non-existing memory to access your information.” 

Funded in part by a grant from the US Air Force Office of Scientific Research's Multidisciplinary University Research Initiative and another from the US Office of Naval Research, this study aims to overcome the limitations of chips that are being used today, which operate under the principles that have been in place for the last many decades. But by utilizing the power of light, this new approach can pave the way for the new generation of AI development. 

The Vast Potential of Silicon Photonics

For the last few decades, research and development in this material have continued. Recently, however, Silicon Photonics (SiPh) has gained attention due to the rising demand for fast and efficient data processing.

This growing interest has the global market size of silicon photonics valued at $1.29 billion in 2022 and is projected to grow at a CAGR of 25.8% by the end of this decade, as per Grand View Research. This growth is due to the need for higher data transfer rates and bandwidth-intensive applications.

SiPh is the perfect platform here due to its economic efficiency and high integration density. Also, given that SiPh is compatible with electronic fabrication, SiPh Photonic integrated circuits (PICs) can be manufactured using established foundry infrastructure. SiPh further has the potential to integrate hundreds to thousands of devices into complex PICs with a design and fabrication scalability similar to CMOS, opening up new applications at the intersection of photonics and computing.

Hence, through its high-speed transmission, high integration density, excellent optical properties, lower power consumption, and relatively inexpensive manufacturing, Silicon Photonics has become a valuable tech in a variety of fields. 

For instance, research has been ongoing in the application of silicon photonics in LiDAR for autonomous driving and industrial automation. LiDAR uses light reflected on surfaces rather than radio frequency (RF) signals to analyze and deliver critical information about the surroundings.

Moreover, silicon photonics can be used for sensing (i.e., optical sensing), where the transmission of a signal and the receipt of the transmitted optical signal can help determine the properties of the surrounding environment. This can be beneficial for health applications and consumer health wearable applications. 

Besides autonomous vehicles and sensing, the usage of silicon photonics has also been explored in telecommunications, quantum communications, biomedical, aerospace, astronomy, and AR/VR. Silicon photonics also shows promise for complete integration and large-scale optical quantum information processing.

Then there is AI, which requires high-performance computing. With AI mania reaching new highs and set to grow further, the chip industry faces a pressing need for innovation. It is diligently working to place more transistors on a single chip to significantly enhance processing power and energy efficiency. Such improvements are crucial for training and operating AI algorithms more accurately, quickly, and cost-effectively.

In an attempt to win the semiconductor race, even China is building a photonic chip production line due to its calculation speed being faster and its information capacity larger, which will be significantly higher than the existing silicon-based chips.

A Game Changer for AI

AI mania is showing no signs of slowing down. This new wave of technological advancement has emerged as a powerful force that will revolutionize many industries and transform the future. With AI fast becoming an integral part of our daily lives and data-intensive applications growing in complexity, everyone from companies and governments to institutions and scientists is looking at ways to make it more efficient. 

This is driving people to silicon photonics, which is one of the most promising technologies to tackle complex and expensive calculations performed by deep neural networks, a subset of machine learning algorithms that allows a model's performance to be more accurate. Deep networks consist of layers that contain mathematical relationships.

With such complexities involved, silicon photonics can help improve performance and cost efficiency, which would improve the function of AI and machine learning applications. The world of AI/ML needs the data to be exchanged quickly while consuming as little energy as possible and, at the same time, must maintain high computational density. 

Here, silicon photonics allows for better communication between computing units. The material further allows the use of short-range optical interconnects to transfer data efficiently over relatively short distances within AI/ML applications. Quick transmission of data is essential for real-time decision-making. 

This way, silicon photonics contributes to the overall effectiveness and performance of AI systems. By leveraging this material, companies can also unlock greater computational capabilities and have more accurate and responsive results.

Silicon photonics is particularly suited to computing due to such circuits' ability to be faster than traditional electronic circuits. Moreover, their optical processing is inherently parallel, which makes it possible to perform multiple actions at the same time. 

Silicon photonics also allows fundamental components to be brought together in numerous combinations to build very complex circuits, enabling the creation of advanced systems tailored to specific applications.

The future of silicon photonics in AI, as we are seeing, is bright, given its potential to transform AI algorithms and further AI systems' capabilities. It's an interesting time for silicon photonics, for sure.

Click here to learn all about investing in artificial intelligence. 

A Look at Popular Chip Manufacturers

Now, let's take a look at a couple of prominent names that are in the chip manufacturing business:

#1. NVIDIA Corporation

The leader in the chip industry, Nvidia, is currently the US stock market's third-most valuable company. After all, it controls about 80% of the AI chip market. With its shares trading at $793.50, the company has achieved a market cap of $1.95 trillion. 

finviz dynamic chart for  NVDA

Nvidia's shares have been soaring like crazy and are already up 58.6% YTD. With that, the company has an EPS (TTM) of 11.93, P/E (TTM) of 65.84, and a ROE (TTM) of 69.17%. It also pays a dividend yield of 0.02%.

As demand surges worldwide across industries and nations, Nvidia reported its fourth-quarter results, with revenue more than tripling to $22.1 billion. According to CEO and co-founder Jensen Huang:

“Accelerated computing and generative AI have hit the tipping point.”

The rising demand for its chips has the company forecast a 233% growth in Q1 revenue. The company's H100 data center chip is what is helping the company lead the AI space. It is optimized to process huge amounts of data and computation at high speeds, making it a perfect solution for the power-intensive task of training AI models. 

Click here to learn all about investing in NVIDIA Corporation (NVDA).

#2. Intel Corporation

The US-based chip maker is making a comeback as it expands its foundry business, which manufactures chip designs for other companies. Microsoft has chosen the company to make its high-end semiconductors and “rebuild western manufacturing at scale.” 

The chip will be designed to use Intel's 18A node, a manufacturing process that makes semiconductors smaller and more energy-efficient. “Intel is the country's champion chip company,” said US Commerce Secretary Gina Raimondo as she noted that Google, OpenAI, and others building LLMs would require a “mind-boggling” volume of semiconductors in the coming years.

finviz dynamic chart for  INTC

As of writing, Intel shares are trading at $43.12, down 14.47% YTD, which puts the company's market cap at $181.7 bln. It has an EPS (TTM) of 0.38, P/E (TTM) of 113.46, and a ROE (TTM) of 1.63%. It also pays a dividend yield of 1.16%. According to Intel CEO Pat Gelsinger:

“The overall demand (for AI chips) appears to be insatiable for several years into the future.”

#3. Samsung

The South Korea-based tech giant is planning to release its 2nm chip technology to gain a lead over other chip manufacturers. As per Samsung's Foundry Forum (SFF) plan, the company will start producing the 2nm process at a large scale in 2025 for mobile apps and will go for high-performance computing applications in the year after that and will then move to the auto industry. A year after that, Samsung is expected to start the 1.4nm process.

The company has a market cap of $373 bln with its shares trading at $1,373. Samsung has a PE Ratio (TTM) of 14.25, an EPS (TTM) of 96.44, and pays a dividend yield of 1.98%. During its Q4 2023 financial report, Samsung noted that its foundry division has secured a deal for its 2nm AI chips from Japanese AI startup Preferred Networks (PFN), which has been previously working with Taiwan Semiconductor Manufacturing Co (TSMC). 

The chipmaker is also collaborating with Arm to optimize the Cortex-X core on its most advanced chip-making process tech, GAA. Late last year, Samsung also signed on with Tenstorrent, which aims to challenge Nvidia as its customer. 


As advances in AI drive the demand for increased computing power, silicon photonics have emerged as a promising technology, which has the potential to reduce latency while increasing efficiency by enabling the fabrication of photonic components on silicon using standard semiconductor manufacturing processes. 

While silicon photonics has many advantages, they won't replace electronic chips anytime soon. This is because the capabilities of silicon photonics remain narrowly focused while having technical barriers in terms of software development to optimize their capabilities. So, it is going to take some time before silicon photonics usage becomes widespread, but then this technology is just getting started, and given AI technology's rate of advancement, this can certainly be accelerated.

Click here to learn about the current state of quantum computing. 

Gaurav started trading cryptocurrencies in 2017 and has fallen in love with the crypto space ever since. His interest in everything crypto turned him into a writer specializing in cryptocurrencies and blockchain. Soon he found himself working with crypto companies and media outlets. He is also a big-time Batman fan.