Faster Chip-to-Chip Communication | Hyperlume

Data Center Energy Consumption and Hyperlume's Innovation

In 2023, data centers accounted for 4.4% of the total electricity used in the United States, with projections indicating this figure could rise to as much as 12% by 2028. A significant portion of this energy is dedicated to facilitating data transfer between chips within these facilities.

Hyperlume, a company based in Ottawa, Canada, is developing a solution aimed at enhancing the energy efficiency and accelerating the speed of this crucial data transfer process.

MicroLED Technology for Enhanced Data Transfer

Hyperlume has engineered a novel version of microLEDs capable of transmitting information at a faster rate than traditional copper-based connections commonly utilized between server racks in data centers.

Furthermore, these microLEDs demonstrate reduced energy requirements for data transmission compared to conventional copper wiring.

The Genesis of Hyperlume

Mohsen Asad, co-founder and CEO of Hyperlume, explained to TechCrunch that the company’s formation was a natural progression from the prior work undertaken by himself and his co-founder, Hossein Fariborzi.

Asad’s expertise in electrical engineering focused on optimizing data transfer methods both within and between chips, while Fariborzi possesses specialized knowledge in the design of low-power electrical circuits.

“The convergence of my work on microLEDs, the increasing demand for data transfer, and concerns regarding power consumption all aligned perfectly,” Asad stated. “We identified a substantial market opportunity.”

Addressing Latency and Energy Challenges

Asad highlighted that both energy consumption and latency have historically posed challenges to chip-to-chip communication within data centers.

These issues have been amplified by the rapid growth and accelerating demands of artificial intelligence (AI).

Resolving the latency problem – minimizing data delay – will not only improve the performance of existing chip connections but also unlock previously inaccessible chip capacity hindered by these bottlenecks, Asad added.

“Successfully mitigating latency allows chips to collaborate more efficiently,” Asad explained. “For large language models, near-zero latency communication between chips is essential.”

Development and Technological Approach

When Hyperlume was founded in 2022, Asad and Fariborzi initially explored utilizing existing technologies to address data center latency.

Silicon was considered, but its cost proved prohibitive for large-scale implementation. Similarly, lasers presented significant cost barriers.

Consequently, Hyperlume opted to enhance readily available microLEDs to facilitate rapid data transfer, effectively replicating the functionality of fiber optic connections without the associated expenses.

“Our core innovation lies in the utilization of ultra-fast microLEDs, coupled with a low-power ASIC that manages communication with other chips,” Asad clarified.

Current Status and Future Plans

Hyperlume is currently collaborating with a select group of early customers, primarily located in North America, to refine its product.

The company has garnered considerable interest, particularly from hyperscale data center operators, as well as cable manufacturers and other industries that could benefit from this technology.

“Our initial focus is on working closely with these early adopters,” Asad said. “Successful implementation within their data centers will enable us to expand our reach to the broader market.”

“Demand is substantial and continues to grow annually.”

Recent Funding and Investment

Hyperlume recently secured $12.5 million in seed funding, led by BDC Capital’s Deep Tech Venture Fund and ArcTern Ventures.

Additional investors included MUUS Climate Partners, Intel Capital, and SOSV.

These funds will be allocated to expanding the engineering team and furthering the development of Hyperlume’s technology, with the goal of accelerating its availability to a wider customer base.

Looking ahead, the company aims to increase its bandwidth capacity to prepare for the next generation of high-performance data centers.

“Our current focus is on optical connections for chip and board communication, but our vision is to evolve into a comprehensive AI connectivity solution provider,” Asad concluded.