Can single-mode fiber arrays drive down cost and complexity of copackaged optics?

As the worlds of advanced semiconductor packaging, optics, photonics, and networking converge to accommodate the performance needs of high-performance computing, AI data centers, and quantum computing, it’s becoming essential to combine optical and electrical functions within the same package, a.k.a. copackaged optics.

Why turn to copackaged optics with silicon photonics as a solution? It accommodates bandwidth density and energy efficiency challenges. Copackaged optics typically involve optoelectrical chiplets that integrate a complementary metal-oxide semiconductor (CMOS) chip with a silicon photonic chip, and an external laser source transmits a signal through a fiber connector (attached to chiplets).

Polarization-maintaining fiber (PMF) arrays are the standard approach used today between external laser sources and chiplets to ensure stable polarization states required for the optical performance of silicon-photonics-based devices. While PMFs provide advantages such as robust polarization control, they tend to have high-precision alignment requirements, increase manufacturing complexity, and significantly drive up packaging costs.

During ECTC, Mehta and Nvidia colleagues will share an alternative approach they’ve designed to replace PMFs with single-mode fiber (SMF) arrays—targeting reduced costs and complexity without compromising system performance.

Coolest part of their SMF design? “If I had to pick one, it’d be the two-stage polarization tracking circuit,” says Mehta. “But the real message of this paper is that you don’t need complex or exotic photonic circuits—just a simple design is enough to eliminate PMF and unlock significant savings in packaging costs.”

One challenge with the new approach is that although “replacing PMF with SMF reduces connector assembly costs, it introduces polarization uncertainty at the transmitter input,” explains Mehta. “We addressed this with a polarization-tracking photonic circuit, but the key challenge was minimizing its power overhead. SMF also requires a specific type of optical input/output (I/O), which typically have higher insertion loss than conventional options and further adds to the photonic link budget. Keeping the combined overhead minimal was critical—otherwise, the cost and performance gains from using SMF could be negated.”

During ECTC, they’re sharing experimental results from a test chip, which demonstrates the viability of their solution, as well as a detailed cost-performance analysis comparing PMF- and SMF-based architectures.

One experimental test chip “demonstrates a photonic circuit designed to address the polarization uncertainty introduced when replacing PMF with standard SMF at the optical transmitter input,” says Mehta. “By leveraging the controlled, shielded environment in which the fiber is deployed, the circuit aims to minimize the electrical power consumption overhead.”

Use of efficient thermo-optical phase shifters will further reduce the power overhead—and allow the full benefits of SMF to be achieved. These include lower connector costs and significantly reduced connector assembly time, which are key factors that will help drive down the overall cost and production time of copackaged optics.

The prototype “is ready now,” he adds. “Once we complete optimization of the thermo-optic phase shifters, it’ll be production ready.”

If you’re attending ECTC, you can catch Mehta’s presentation—Session 2.6—on Wednesday, May 28th, at 11:35 am.