Next Unicorns: Unlocking the power of photonic computing with Lightmatter CEO Nick Harris | E1787 | Summary and Q&A

TL;DR

Light Matter, a pioneer in photonic computing, is set to revolutionize the tech industry with its innovative optical products. The company aims to build the leading photonics company, offering groundbreaking solutions that leverage light-based technologies.

Key Insights

• 🙂 Light Matter is pioneering the field of photonic computing, which uses light for computation and data transfer instead of electrical signals.
• 💿 The company's optical components enable deep learning computations and offer a breakthrough in neural network mathematics.
• 🔐 By bringing optical engines closer to processors and developing efficient data transfer solutions, Light Matter addresses key challenges in traditional computing, such as heat generation and power consumption.
• 💗 The demand for computing power, especially in the field of AI, is rapidly growing, driving the need for innovative solutions like photonic computing.
• ✋ Light Matter's technology has the potential to transform industries that rely heavily on computing, such as data centers, AI, and high-performance computing.
• 👯 The company's focus on people, product development, predictability, and go-to-market strategy has been instrumental in its success in achieving rapid growth.
• 🍉 While fundraising and capital are important, focusing on unit economics, customer feedback, and sustainable growth is crucial for long-term success.

Transcript

I think what you're going to see from light matter over the next year and a half is Extreme volume on groundbreaking Optical products and you know what I'm really aiming to do with light matter ultimately is to build the photonics company so if you think of the word photonics I want you to think of light matter ultimately and the way that's going t... Read More

Questions & Answers

Q: What is photonic computing, and why is it important?

Photonic computing uses light instead of electrical signals for computation and communication. It offers significant advantages in terms of speed, energy efficiency, and scalability. By leveraging optical components, photonic computing overcomes the limitations of traditional computing and enables groundbreaking advancements in various fields.

Q: How does Light Matter's technology contribute to the future of computing?

Light Matter's technology plays a crucial role in advancing computing capabilities. Their optical components enable deep learning computations using light, revolutionizing the field of artificial intelligence. Additionally, their interconnect product improves data transfer between chips, increasing overall speed and efficiency. By pushing the boundaries of what is possible with light-based solutions, Light Matter is shaping the future of computing.

Q: What challenges does photonic computing address?

Photonic computing addresses several challenges in traditional computing. It overcomes the limitations of electrical signals, such as heat generation and energy inefficiency. By using light instead, photonic computing can achieve much higher data transfer rates and reduce power consumption significantly. This breakthrough technology enables faster, more energy-efficient computing systems that can drive advancements in AI, data centers, and other computing-intensive applications.

Q: How does Light Matter's technology compare to other computing solutions like GPUs and TPUs?

Light Matter's technology offers unique advantages compared to GPUs and TPUs. While GPUs and TPUs have played a significant role in accelerating computing performance, they still rely on electrical signals for computations and communication. Light Matter's photonic computing leverages the speed and efficiency of light, enabling even more significant advancements in computing power and energy efficiency. The company's optical components and interconnect solutions provide a new paradigm for next-generation computing systems.

Summary

In this video, Jason interviews Nick Harris, the CEO of Light Matter, a company at the forefront of photonic computing. They discuss the concept of photonic computing, its potential applications, and how it compares to traditional computing. Nick explains the technology developed by Light Matter, including their optical compute engines and interconnect product line. They also discuss the impact of photonic computing on AI models, the challenges of heat dissipation in traditional computing, and the future of supercomputers and quantum computing. They touch on the recent room temperature superconductor breakthrough and the excitement around replicating the results. Overall, they explore the potential of photonic computing and its role in shaping the future of AI and technology.

Questions & Answers

Q: What is photonic computing and how does it differ from traditional computing?

Photonic computing refers to the use of light to perform computations and transmit data, in contrast to traditional computing that relies on electrical signals. This new approach offers several advantages in terms of speed, energy efficiency, and scalability. By using light instead of electricity, photonic computing enables faster data transmission, reduced energy consumption, and the potential for more compact and powerful computational systems.

Q: How did the trend of using electrical signals in computing break in 2005?

The trend of using electrical signals in computing broke in 2005 when the concept of "Moore's Law" and "Dennard Scaling" reached their limits. Moore's Law suggested that the number of transistors on a chip would double approximately every two years, leading to increased computational power. Dennard Scaling referred to the idea that the energy usage per device would decrease as transistors became smaller. However, as chip designs became more complex, the energy density and heat dissipation challenges became increasingly difficult to overcome. This marked a turning point in the development of new computing technologies.

Q: How does Light Matter's technology contribute to photonic computing?

Light Matter has developed two key technologies for photonic computing. Firstly, they have created optical compute engines that utilize tiny optical components to perform computations for deep learning and neural network mathematics. These optical compute engines offer increased speed, energy efficiency, and scale potential compared to traditional computing methods. Secondly, Light Matter has developed an interconnect product line called "Passage" that facilitates high-speed data transmission between chips and servers, allowing for efficient utilization and workload management.

Q: Are Light Matter's products already available in the market?

Light Matter has made significant progress in its development. They have raised $300 million in funding and have already built several generations of chips. They currently have six customers for their compute product line and are planning to launch their interconnect product line in high volume next year. While the products are not widely available yet, Light Matter is well on its way to commercializing their photonic computing technologies.

Q: Who are Light Matter's customers and how do they compete against companies like Nvidia?

Light Matter's current customers are cloud infrastructure providers, such as the major cloud computing platforms. For their compute product line, they do compete with companies like Nvidia, as both provide hardware for running AI models. However, on the interconnect side, Light Matter's technology can be used by various companies, including AMD, Intel, and even Amazon's internal product teams. Light Matter's interconnect technology allows these companies to extend the capabilities of their chips and achieve better performance by integrating optical interconnects within their systems.

Q: How does photonic computing address the energy and heat dissipation challenges faced by traditional computing?

Traditional computing systems generate a significant amount of heat during data transmission and computations, which requires complex cooling systems. Photonic computing addresses this issue by using light instead of electrical signals for data transmission. Light-based interconnects and compute engines produce less heat, enabling higher energy efficiency and reducing the need for extensive cooling systems. This results in better overall system performance, reduced energy consumption, and, potentially, the ability to fit more compute power into a smaller space.

Q: Why is underutilization of hardware a problem in AI models?

AI models, especially large language models like GPT-3, often suffer from underutilization of hardware resources. This means that a considerable portion of the available computational power goes unused in performing calculations and data transmission. Most of the time, these models spend a significant portion of their runtime transferring data between memory and processors or communicating between different cores within the same or different boxes. This inefficiency can be addressed by improving data transfer rates and developing interconnect technologies, such as Light Matter's "Passage," which increases the utilization and overall efficiency of the hardware.

Q: Will companies catch up with the increasing demand for AI models and computing power?

The demand for AI models and computing power is growing rapidly, and it's hard for companies to keep up with this demand. While significant progress can be made through innovations in hardware and software, it is unlikely that companies will fully catch up in the near future. The growth of AI and computing power is accelerating, making it challenging to meet the increasing demand. Efforts are being made to scale chip production, improve process technology, and develop new technologies, but it will take time to address the growing need for computational resources.

Q: Is there a limit to the extent of progress that can be made in traditional computing?

As traditional computing approaches encounter physical limitations, such as heat dissipation issues and the increasing complexity of chip designs, there is a limit to the extent of progress that can be made using conventional methods. To continue making advancements, new technologies and approaches are needed. This includes leveraging photonic computing with its potential for superior energy efficiency, faster data transmission, and scalability. By adopting new technologies like photonic computing, progress can be made even as traditional computing approaches reach their limitations.

Q: What is the potential of photonic computing in relation to general AI?

Photonic computing has the potential to play a significant role in the development of general AI. With its faster data transmission, energy efficiency, and scalability, photonic computing can enhance the performance and capabilities of AI models. AI models, such as deep neural networks, can benefit from the high-speed optical compute engines developed by companies like Light Matter. These advancements in computing technologies can contribute to the further progress of general AI and enable the development and training of more powerful and sophisticated models.

Q: How does light matter contribute to the AI field with their optical technologies?

Light Matter's optical technologies provide key innovations for the AI field. Their optical compute engines allow for deep learning computations using tiny optical components, providing faster and more energy-efficient processing compared to traditional electrical-based methods. These optical compute engines are crucial for AI models that deal with large language models and neural network mathematics. Additionally, Light Matter's interconnect product line, "Passage," improves data transfer rates, addressing the underutilization of hardware resources. By leveraging these optical technologies, AI models can perform at higher speeds and achieve better overall system efficiency.

Summary & Key Takeaways

• Light Matter is at the forefront of photonic computing, which uses light instead of electrical signals for computation and communication.

• The company has developed two types of technology: optical components for deep learning computations and an interconnect product that improves data transfer between processors and servers.

• By bringing optical engines closer to processors and developing more energy-efficient solutions, Light Matter is addressing the limitations of traditional computing and paving the way for increased computing power and efficiency.