Inverse-Designed Photonic Devices: A Tutorial on CWDM Demultiplexers

Introduction

The field of photonics has seen remarkable advancements in recent years, driven in part by the emergence of inverse-designed photonic devices. Unlike traditional design approaches that rely on intuition and experience, inverse design starts with the desired device performance and then employs optimization algorithms to find the optimal device structure. This innovative approach has the potential to unlock new and innovative photonic devices that were previously challenging or even impossible to realize.

However, the adoption of inverse-designed photonics faces some key challenges, particularly around device performance predictability and variability. The complex and often non-intuitive patterns in inverse-designed components can introduce fabrication complexities, making it crucial to demonstrate that these devices can deliver predictable performance and consistent behavior across multiple fabricated samples.

In this tutorial, we will explore the design, fabrication, and characterization of an inverse-designed silicon-based coarse wavelength division multiplexing (CWDM) demultiplexer (DEMUX) operating in the O-band (1270 nm to 1330 nm). This device serves as a case study to showcase the capabilities and considerations of inverse-designed photonics in a practical, real-world application.

Device Design

To demonstrate the real-world relevance of inverse-designed photonics, we start with the design of a CWDM4 demultiplexer, which separates four wavelength channels at 1270 nm, 1290 nm, 1310 nm, and 1330 nm in the O-band.

The device is designed on a silicon-on-insulator (SOI) platform, with the silicon thickness determined by the foundry's process design kit (PDK). The design optimization is performed using the topology optimization method introduced in [4], which discretizes the 14 μm × 16 μm design region into 10 nm pixels. Strict design rule constraints are enforced using a non-differentiable conditional generator and a straight-through estimator [5] to enable backpropagation through the generator. The electromagnetic simulations of the device are carried out using a finite difference time domain (FDTD) solver [6].

The resulting optimized design, shown in Figure 1(a), consists of a complex pattern of silicon and silicon oxide regions that satisfy the foundry's design rules. The energy intensity at the center wavelengths of the four channels (1270 nm, 1290 nm, 1310 nm, and 1330 nm) is depicted in Figure 1(b), illustrating the spatial distribution of the optical field within the device.

Figure 1. (a) The 14x16μm design of the CWDM4 demux obtained using the design methodology in [4]. The white material represents silicon and the black material is silicon oxide. (b) The energy intensity at the center wavelength of each channel, corresponding to wavelengths of 1270 nm, 1290 nm, 1310 nm, and 1330 nm. The gray contour indicates the material interface between silicon and silicon oxide.

Experimental Results

The inverse-designed CWDM DEMUX devices were fabricated in a passive flow at GlobalFoundries using their 45SPCLO silicon photonics process. Measurements were carried out at the wafer level, where a laser source swept through the different wavelengths, and a photodiode detected the corresponding output signals. The spectra of the device under test (DUT) were de-embedded by subtracting the grating reference measurement.

Figure 2(a) presents a comparison between the simulated and measured transmission spectra of the four DEMUX channels. The thick solid lines represent the simulated spectra, while the dashed lines represent the measured spectra. The good agreement between simulation and measurement indicates a high degree of performance predictability in the inverse-designed device.

To further examine the device variability, the transmission spectra from 34 different chips are overlaid in Figure 2(b). Zoomed-in views of the red (channel 4) and green (channel 3) channels are provided in Figures 2(c) and 2(d), respectively. These plots demonstrate the consistent performance across the 34 chips, with only minor variations in the measured results.

Figure 2. (a) Measured (solid) and simulated (dashed) transmission spectra for a single chip. (b) Measured transmission spectra from 34 chips. (c-d) Enlarged view of measured transmission spectra across channel 3 and channel 4.

The key performance metrics extracted from the measured results are summarized in Figure 3. Figure 3(a) shows the worst insertion loss for each channel, with a mean ranging from 2 dB to 3.3 dB. The one-sigma distribution of the insertion loss is 0.41 dB, demonstrating reasonably low variability. Figure 3(b) provides the statistics on the worst crosstalk, with a mean ranging from 19 dB to 26 dB. The one-sigma distribution of the crosstalk is 0.62 dB, further highlighting the consistent performance across the fabricated devices.

Figure 3. Measured (a) worst insertion loss and (b) worst crosstalk across 34 chips for each channel.

It is worth noting that while the overall performance is promising, there is room for improvement in certain aspects, such as the insertion loss and crosstalk of specific channels. The simulation-measurement discrepancy observed in channel 3 (green channel) suggests the need for further optimization and refinement in the inverse design process.

Conclusion

In this tutorial, we have demonstrated the design, fabrication, and characterization of an inverse-designed silicon-based CWDM DEMUX operating in the O-band. The key highlights of this work include:

1. Inverse Design Methodology: We have employed a topology optimization-based inverse design approach to create a CWDM DEMUX with a complex, non-intuitive structure that satisfies the foundry's design rules.

2. Performance Predictability: The good agreement between simulated and measured device performance indicates a high degree of predictability in the inverse-designed photonic device.

3. Device Variability: The consistent performance observed across 34 fabricated chips, with reasonably low one-sigma variations in insertion loss and crosstalk, showcases the viability of inverse-designed photonics for mass production.

This case study on the CWDM DEMUX highlights the potential of inverse-designed photonics to deliver innovative and practical photonic devices. While challenges remain, the demonstrated capabilities in terms of predictability and variability represent a significant step towards making inverse-designed photonics a viable technology for real-world applications.

References

[1] T. Van Vaerenbergh, S. Hooten, M. Jain, P. Sun, Q. Wilmart, A. Seyedi, Z. Huang, M. Fiorentino, and R. Beausoleil, "Wafer-level testing of inverse-designed and adjoint-inspired dual layer Si-SiN vertical grating couplers," J. Phys. Photonics 4(4), 044001 (2022).

[2] Yang, K.Y., Shirpurkar, C., White, A.D. et al. "Multi-dimensional data transmission using inverse-designed silicon photonics and microcombs," Nat Commun 13, 7862 (2022).

[3] A. M. Hammond, J. B. Slaby, M. J. Probst, and S. E. Ralph, "Multi-layer inverse design of vertical grating couplers for high-density, commercial foundry interconnects," Opt. Express 30(17), 31058–31072 (2022).

[4] Martin F. Schubert, Alfred K. C. Cheung, Ian A. D. Williamson, Aleksandra Spyra, and David H. Alexander, " Inverse Design of Photonic Devices with Strict Foundry Fabrication Constraints," ACS Photonics 9 (7), 2327-2336 (2022).

[5] Y. Bengio, N. Léonard, A. Courville, "Estimating or Propagating Gradients Through Stochastic Neurons for Conditional Computation," arXiv:1308.3432 [cs.LG] (2013).

[6] A. Taflove, S. C. Hagness, "Computational Electrodynamics: The Finite-Difference Time-Domain Method," 3rd ed., Artech, Norwood, MA (2005).