In 2023, data centers accounted for 4.4% of electricity consumption in the U.S., with projections indicating this could rise to as much as 12% by 2028. Most of this energy is utilized for data transfer between chips. Hyperlume, a company aiming to enhance energy efficiency while accelerating this process, is leading the charge.
Based in Ottawa, Canada, Hyperlume has developed a type of microLEDs that facilitate faster data transmission compared to the traditional copper connections typically found in data centers. These microLEDs also consume less energy while transferring data than their copper counterparts.
Mohsen Asad, co-founder and CEO of Hyperlume, explained to TechCrunch that the startup represents a “logical extension” of their previous work. With a background in electrical engineering, Asad focused on enhancing data transfer between chips and racks, while co-founder Hossein Fariborzi specializes in low-power electrical circuit design.
“I was involved in microLEDs and data transmission, and with the rapid rise of AI, the demands for chip communication and power efficiency naturally aligned,” Asad noted. “We identified a significant market opportunity.”
According to Asad, energy usage and latency have historically posed challenges for chip-to-chip communication in data centers, problems that have intensified with the explosive growth of AI technology. Addressing latency issues can not only enhance the speed of existing chip connections but also unlock previously inaccessible chip capacities hindered by these delays.
“If we can effectively tackle the latency issue, we enable chips to collaborate more efficiently,” Asad stated. “With large language models, the need for chips to communicate with minimal latency is crucial.”
When Asad and Fariborzi launched Hyperlume in 2022, they focused on leveraging current technologies to mitigate data center latency. Although silicon offered a potential solution for chip connectivity, its high cost impeded scalability. Similarly, using lasers was considered too expensive.
As a solution, Hyperlume opted to adapt affordable microLEDs to enable rapid data transfer between chips, achieving results akin to fiber optic connections without the associated costs.
“Our unique advantage lies in ultra-fast microLEDs combined with a low-power ASIC that facilitates communication among chips,” Asad highlighted.
Currently, Hyperlume is collaborating with a select group of early customers, primarily in North America, while refining its product. Asad mentioned that there has been significant interest, particularly from hyperscalers, cable manufacturers, and other industries that could gain from their technology.
“Our initial phase involves partnering with these early adopters. Once our technology has been validated within their data centers, we will have the opportunity to scale and meet the broader market’s needs,” Asad remarked. “The demand is rapidly increasing each year.”
Hyperlume has recently secured $12.5 million in seed funding, spearheaded by BDC Capital’s Deep Tech Venture Fund and ArcTern Ventures, with contributions from MUUS Climate Partners, Intel Capital, and SOSV, among others.
This funding will be utilized to hire additional engineers and accelerate the development of Hyperlume’s technology, aiming for quicker market entry. The company envisions expanding its bandwidth capabilities to be ready for the next generation of sophisticated data centers.
“Presently, we’re focused on creating optical connections for chip and board interconnectivity, but we ultimately aspire to evolve into a leading AI connectivity solution provider,” Asad concluded.
Compiled by Techarena.au.
Fanpage: TechArena.au
Watch more about AI – Artificial Intelligence
