Join top executives in San Francisco on July 11-12, to hear how leaders are integrating and optimizing AI investments for success. Learn More
As AI models become more computationally demanding, engineers are looking to new types of materials and hardware to speed up the model development process. One category of components with promise is photonic chips, which leverage light to send signals as opposed to the electricity that conventional processors use. In theory, photonic chips could lead to higher performance because light produces less heat than electricity, can travel faster, and is less susceptible to changes in temperature and electromagnetic fields.
But photonic chips have drawbacks that must be addressed if the technology is to reach the mainstream. They’re physically larger than their electronic counterparts and difficult to mass-produce, for one, owing to the immaturity of photonic chip fabrication plants. Moreover, photonic architectures still largely rely on electronic control circuits, which can create bottlenecks.
The hurdles haven’t discouraged the companies currently pursuing photonics, which include Lightmatter and LightOn as well as tech giants like Intel and Japan-based NTT. Joining the growing segment is Celestial AI, which launched in 2019 to develop a hardware and software platform for AI chipsets that use light to move data across chips. Celestial today announced that it raised $56 million in funding at a $150 million post-money valuation, bringing its total capital raised to over $64 million.
Light-based AI processors
Celestial (previously Inorganic Intelligence) is a portfolio company of The Engine, the venture capital firm spun out of MIT in 2016. Phil Winterbottom, David Lazovsky, Michelle Tomasko, and Preet Virk launched the startup with the goal of creating light-based products that “highly efficiently” map data and compute without the need for complex optimizations. Winterbottom previously spent time at Bell Labs, Nokia’s distinguished R&D division, prior to stints at fiber access startup Entrisphere (which was acquired by Ericsson) and cable networking company Gainspeed (which was acquired by Nokia). Tomasko, who leads Celestial’s software effort, worked in various engineering roles at Nvidia and Google.
Join us in San Francisco on July 11-12, where top executives will share how they have integrated and optimized AI investments for success and avoided common pitfalls.
Celestial claims to have developed a photonic-based architecture that scales across multichip systems, enabling data transfer via light both within chips and chip-to-chip. The company’s hardware product integrates photonics into an AI accelerator, making both memory and compute available for AI workloads.
“Celestial was founded to leverage advancements in silicon photonics, driven by data communications, in addition to advanced [chipset design] to decouple system performance from the limitations of electronic-only systems,” CEO David Lazovsky told VentureBeat via email. “Celestial’s mission is to fundamentally transform computing efficiency with [our] technology platform that uses light for data movement both within chip and between chips.”
In traditional hardware, transistors control the flow of electrons through a semiconductor, performing operations (and running software) by reducing information to a series of ones and zeros. By contrast, Celestial’s hardware — like all photonics hardware — calculates by splitting and mixing beams of light within nanometer-wide channels. Photonics chips’ calculations are analog as opposed to digital, meaning that they’re inherently less accurate. But photonics chips can perform these calculations — including the calculations involved in training AI models — quickly and in parallel.
“We have developed a highly flexible, general-purpose AI accelerator that has broad applicability to various machine learning workloads spanning computer vision, recommendation and personalization, natural language processing, and more,” Lazovsky said. “[The] new money is being used to commercialize our Orion AI accelerator products, including our design services engagement with Broadcom for the 5nm ASIC manufactured at TSMC. Orion AI accelerators will be offered in two form factors: a standard PCIe card (75W TDP) and a multi-chip, optically interconnected server configuration.”
The details of Celestial’s technology remain vague — we’ve reached out with questions — but regardless, the startup is up against well-established competitors in the photonics field. Lightmatter already sells server blades containing photonic chips that fit into conventional datacenters. LightOn announced late last year that one of its photonic “co-processors” would be integrated into France’s Jean Zay supercomputer. And in December, Lightelligence claimed to have solved a challenging math problem 100 times faster than a typical GPU with its photonic hardware.
On the other hand, while the competition is indeed fierce, Verified Market Research predicts that the global market for AI chips could be worth $109.83 billion by 2028 — making even a small slice potentially very lucrative. The photonics industry trade group The Photonics Public Private Partnership optimistically estimates that only 20% of the power and benefits of light technologies have been unlocked to date.
“The AI industry is experiencing unprecedented growth, which is the product of exponentially increasing datasets from ever-increasing sources coupled with increasing complexity in AI workloads,” Lazovsky added. “Celestial’s technology platform brings an entirely new set of tools to the equation to allow our customers to address this problem. We are leveraging the complementary strengths of electronics for high-performance, high-precision computing and photonics for high-speed, low-power, high-bandwidth data movement … The machine learning application benefits extend beyond performance and low power to latency, user-friendly software, and low total cost of ownership.”
KDT led Celestial’s latest funding round with participation from Temasek’s Xora Fund, The Engine, Tyche Partners, M-Ventures, IMEC XPand, and Fitz Gate. The startup — which is pre-revenue and has only focused on “a small number of strategic customer collaborative engagements” for now — says that the new capital will be used to expand its global engineering team, product development, and strategic supplier engagements, including semiconductor manufacturer Broadcom.
VentureBeat's mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Discover our Briefings.