Imec Breakthrough: 4x Increase in Wavelength Channels with WDM Advancement

Imec, a leading research and innovation hub in nanoelectronics and digital technologies, recently showcased a significant breakthrough in silicon-based wavelength-division multiplexing (WDM). The demonstration featured a compact 32-channel silicon-based wavelength filter with remarkable attributes such as low loss and high tuning efficiency. This advancement paves the way for a fourfold increase in the number of transmitted and received wavelength channels compared to current commercial transceivers.

This technological feat holds immense promise for the future scalability of bandwidth density and power efficiency in next-generation silicon-photonics-based transceivers. It specifically addresses the growing demands for short-reach optical interconnects in high-performance AI and machine learning compute clusters.

According to Joris Van Campenhout, a key figure at Imec, "Imec’s optical I/O R&D programme is focused on achieving disruptive scaling in both bandwidth density and power consumption of silicon-integrated optical interconnects, which often present conflicting requirements." The newly developed 32-channel WDM filter allows for Tbps-scale aggregate bandwidth per optical fiber using individual lane rates of just 32Gbps. This breakthrough enables error-free link operation without the need for power-intensive digital signal processing.

The compact footprint, low optical losses, and high wavelength tuning efficiency of the demonstrated WDM filter offer a clear pathway to achieving Tbps/mm-scale optical interconnects with exceptional power efficiencies, as low as 1pJ/bit.

Continued advancements in optical interconnect technologies are crucial for unlocking further performance and efficiency gains in AI and machine learning compute clusters. Silicon photonics stands out as a pivotal technology platform for the cost-effective integration of Tbps-scale transceivers. However, meeting the stringent requirements for bandwidth density, power efficiency, and latency in AI and machine learning system interconnects necessitates significant enhancements at the process, device, and circuit levels.

One of the key challenges in scaling up link bandwidth by adding more wavelength channels is the potential for substantial optical losses and excessive tuning power when utilizing silicon-integrated components. Imec has successfully tackled these challenges by leveraging the advanced Si patterning capability in its 300mm silicon photonics platform (iSiPP300) and implementing innovative designs for Si ring resonator and interleaver components.

The WDM filter developed by Imec supports 32 wavelengths at a 100GHz channel spacing in the O-band, aligning with the specifications of the CW-WDM multi-source agreement (MSA). Despite the high channel count, the design optimization has ensured low optical insertion losses of approximately 2dB and crosstalk levels better than 16dB. Through the utilization of Imec’s local Si substrate undercut process and optimized heater designs, the thermo-optic tuning efficiency of the filter ranks among the best-reported values at 290GHz/mW.

Imec's groundbreaking advancements in silicon-based wavelength-division multiplexing not only push the boundaries of current optical interconnect technologies but also hold the potential to revolutionize the landscape of high-performance computing and data-intensive applications in the near future.