Discussing the future of high-capacity photonic chips

The recent explosion of artificial intelligence (AI) has driven a related demand for faster processors and greater data bandwidth. This is especially relevant at the interconnects between chips and between racks inside the large data centers where AI models are trained. The data processed by computing chips is rapidly outpacing the speed of traditional electrical interconnects, requiring ever more interconnects with greater power and cost requirements.

As an alternative to electrical interconnects, optical systems can achieve faster transmission rates with less power usage. Su et al. explored the future of photonic chips, estimating when high-capacity input/output (I/O) chips in the 100 terabits per second (TBps) and petabit per second (Pbps) range will likely be developed and available.

Given the capabilities of existing photonic chips, along with their current growth rates, the authors projected that commercial optical I/O chips will likely reach 16 Tbps in a decade. However, with accelerated development to keep pace with advances in computing speed, research-stage optical I/O chips could reach over 400 Tbps in 10 years, while ground-breaking research could reach 1 Pbps.

The authors highlighted several challenges researchers would need to overcome to reach these capacities.

“Channel count, chip size, and power consumption are three major challenges for achieving petabit capacities, and some may be approaching the theoretical limits in 10 years or so,” said author Yikai Su.

However, they are positive that improved materials and designs can overcome these hurdles.

“Hybrid integration looks needed in the near term while heterogeneous or monolithic integration is promising for long term development; single material systems cannot solve all these challenges,” said Su. “Heterogeneous integration may balance between integration density, performance, throughput, yield, and cost.”

Source: “When do we need Pbps photonic chips and what are the challenges?,” by Yikai Su, Yu He, Ting Wang, Xuhan Guo, Xi Xiao, and Guo-Qiang Lo, Applied Physics Letters (2024). The article can be accessed at https://doi.org/10.1063/5.0232053 .

This paper is part of the Hybrid and Heterogeneous Integration in Photonics: From Physics to Device Applications Collection, learn more here .