Celestial AI raises $100 million to transfer data using light-based interconnects

Image credits: Jian Fan/Getty Images

David Lazovsky and Preet Virk, technologists with backgrounds in semiconductor engineering and photonics, came to the joint realization several years ago that TO THE and machine learning workloads would quickly run into a data movement problem. Increasingly, they predicted, it would become difficult to move data to and from computing hardware as AI models scaled beyond what could be stored on the die of a single memory chip.

Their solution designed by Phil Winterbottom, formerly a researcher at the distinguished Bell Labs, was an optical interconnect technology for data transmission from compute to compute, memory and on-chip computation. Together with Winterbottom, Lazovsky and Virk founded a startup, Celestial AI, to commercialize the technology. And now, that startup is attracting huge backers.

Celestial AI announced today that it has raised $100 million in a Series B round led by IAG Capital Partners, Koch Disruptive Technologies and Temaseks Xora Innovation fund. The tranche, which brings Celestial AI’s total raised to more than $165 million, will be used to support production of the Celestial photonics platform by expanding the company’s engineering, sales and technical marketing departments, according to CEO Lazovsky.

Celestial currently has about 100 employees, a number Lazovsky expects to grow to 130 by the end of the year.

Today, computation and memory are closely linked. The only way to add more high-bandwidth memory is to add more compute, whether or not the additional compute is required, Lazovsky told TechCrunch via email. Celestials technology enables memory disaggregation.

In a data center, memory is often one of the most expensive resources, in part because it isn’t always used efficiently. Because memory is tied to compute, it is challenging and sometimes impossible due to bandwidth limitations and ultra-high latency for operators to disaggregate and pool memory across hardware within the data center.

According to an internal Microsoft study, up to 25% of memory in Azure is either frozen or remains after server cores have been leased to virtual machines. Reducing this locked memory could reduce data center costs by 4% to 5%, the company estimates potentially significant savings in the context of a multibillion-dollar operation.

Celestial, which started as a portfolio company of The Engine, the VC firm born out of MIT in 2016 has developed an apparent solution in its photonics-based architecture, which fits into multi-chip systems. Using light to transfer data, Celestials technology can transmit information both within chips and from chip to chip, making both memory and computation available for artificial intelligence and other workloads.

Image credits: Heavenly

Celestial also claims its technology can reduce the amount of electricity required for data movement, indirectly boosting a chip’s performance. Typically, chips dedicate some of the electricity they draw to moving data between their circuits, which takes away any electricity the chip can direct to computing tasks. Celestial photonics reduces the power required for data movement, allowing a chip at least in theory to increase its computing power.

Celestials photonics technology, which is compatible with most industry interconnect standards (e.g., CXL, PCIe), delivers 25X higher bandwidth and 10X lower latency and power consumption than optical alternatives , says the company.

With the growth in TO THE , especially large language models (LLM) and recommendation engine workloads, there is a shift towards accelerated computation, Lazovsky said. The key issue going forward is memory capacity, memory bandwidth, and data movement, i.e. chip-to-chip interconnectivity, which is what we’re addressing with the Celestials photon fabric.

Celestial offers its interconnect product through a licensing program and says it is engaged with several tier-one customers, including hyperscalers and processor and memory companies.

The interconnect product appears to be the number one priority for Celestial. Celestial sells its own AI accelerator chip, dubbed the Orion, built on the company’s photonics architecture. But as investors told TechCrunch in a recent piece for TC+, AI photonics chips still need to overcome the engineering challenges that would make them practical at scale. Unless Celestial has stumbled upon breakthroughs in the areas of data-to-analog conversion and signal regeneration, the major roadblocks for today’s photonics chips, Orion is unlikely to be far ahead of the competition.

Chips aside, Celestial has a number of competitors in a photonics IC market that could be worth $26.42 billion by 2027.

Ayar Labs, which makes chip solutions based on optical network principles, has raised more than $200 million in venture capital since its founding in 2015. Ravonus, another rival, recently secured a $73.9 million investment.

However, there may be some consolidation in sight in the broader optical interconnect space. About three years ago, Marvell bought Inphi, an optical networking specialist, for $10 billion. After a quiet period, last year Microsoft acquired Lumenisity, a startup that develops high-speed optical cables for data centers and transportation networks.

Both Inphi and Luminensity were targeting different use cases with their technology. But Big Tech’s enthusiasm for optics and photonics is worth noting.

Our photonics technology is truly differentiated and is unique with superior characteristics, said Lazovsky. Given the growth of AI workloads due to LLMs and the pressures it places on current data center architectures, the demand for optical connectivity is rapidly increasing to support the transition from general IT data center infrastructure to accelerating computing.

Samsung Catalyst, Smart Global Holdings, Porsche Automobil Holding SE, The Engine Fund, imec.xpand, M Ventures and Tyche Partners also participated in Celestials Series B.



#Celestial #raises #million #transfer #data #lightbased #interconnects
Image Source : techcrunch.com

Leave a Comment