🔬 optics

Sidewall-Poled Nanophotonic Lithium Niobate with Bidirectional Characterization

This paper introduces a non-destructive, bidirectional characterization framework that utilizes classical power measurements to independently extract facet coupling losses and intrinsic nonlinear performance in sidewall-poled lithium niobate waveguides, achieving a record normalized second-harmonic generation efficiency of 1850% W⁻¹ and enabling high-throughput benchmarking across photonic platforms.

Original authors: Aditya Tripathi, Michael S. Bullock, Parash Thapalia, Pooja Kulkarni, Jaber Balalhabashi, Robert Kwolek, Shiva Behzadfar, Kazuki Hirota, Shion Eto, Rintaro Tsuchida, Liam Beaudoin, Sasan Fathpour, Raj

Published 2026-08-24

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Aditya Tripathi, Michael S. Bullock, Parash Thapalia, Pooja Kulkarni, Jaber Balalhabashi, Robert Kwolek, Shiva Behzadfar, Kazuki Hirota, Shion Eto, Rintaro Tsuchida, Liam Beaudoin, Sasan Fathpour, Rajveer Nehra

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

In the world of modern technology, the ability to manipulate light is the engine behind everything from the internet to advanced medical imaging. For decades, scientists have worked to shrink these optical systems down, packing them onto tiny chips similar to the silicon processors in our phones. These integrated devices can guide light through microscopic pathways, mixing colors, amplifying signals, and creating new forms of light. However, a persistent problem has plagued the field: when light enters or leaves these tiny chips, it often gets lost at the edges. This loss is not just a minor inconvenience; it hides the true performance of the device. If a scientist measures how well a chip converts one color of light into another, they cannot easily tell if the result is poor because the chip itself is inefficient, or simply because the light struggled to get in and out. This uncertainty makes it difficult to build the reliable, high-performance systems needed for the next generation of quantum computers and ultra-fast communication networks.

To solve this puzzle, a team of researchers has developed a clever new way to test these devices that separates the chip's true ability from the messy reality of its connections. They focused on a specific type of material called thin-film lithium niobate, which is a crystal known for its exceptional ability to change the color of light. While this material is powerful, the process used to pattern it on a chip often introduces tiny imperfections that scatter light and reduce performance. The researchers created a new method to measure the chip's internal efficiency without destroying it or needing complex, time-consuming microscopes. By sending light through the chip in two different directions and analyzing how the light changes, they could mathematically untangle the losses caused by the chip's edges from the losses caused by the material itself. This approach allowed them to see the device's true potential for the first time, revealing that their new manufacturing technique produces chips that are far more efficient than previously thought possible.

The core of this discovery lies in a technique the researchers call a bidirectional characterization framework. Imagine trying to judge the quality of a long, narrow hallway by measuring how much light reaches the other end. If you only look from one side, you cannot tell if the dim light at the exit is because the hallway walls are absorbing the light, or because the door you entered through was partially closed. In the past, scientists had to guess or make assumptions to separate these factors, which often led to inaccurate results. The new method, however, uses a specific interaction where a beam of light is converted into a different color, and then that new color is used to amplify a second beam. By running this process forward and then backward through the same chip, the researchers could compare the results. Because the light travels through the same internal path but enters and exits through different doors, the differences in the output reveal exactly how much light was lost at each specific end. This allows them to calculate the chip's intrinsic efficiency with high precision, removing the guesswork that has long clouded the field.

Using this method, the team tested a new way of building these chips called sidewall poling. Traditionally, manufacturers would first create the pattern inside the crystal and then carve the chip out of the material. This "poling-before-etching" approach often left the edges of the chip rough and uneven, causing significant light loss. The researchers flipped this process, carving the chip first and then applying the pattern to the sides. This "etching-before-poling" strategy resulted in much smoother edges and significantly less light scattering. When they measured the new devices, they found a normalized efficiency of 1850 percent per watt, a figure that stands among the highest ever reported for this type of material. This number represents the chip's ability to convert light on its own, stripped of the losses caused by the fiber optic cables used to connect it to the outside world. The device also demonstrated the ability to amplify light across a very wide range of colors, spanning more than 10 terahertz, which is a crucial capability for handling large amounts of data.

Beyond simply measuring efficiency, the researchers used their bidirectional method to map the internal structure of the chip. They discovered that the pattern they created was remarkably uniform along the entire length of the device, with only minor variations in the thickness of the crystal film. This level of uniformity is essential for the device to work correctly over long distances. By comparing their optical measurements with physical images of the chip, they confirmed that their new manufacturing process successfully minimized the defects that usually plague these devices. The study also highlighted that the traditional method of making these chips introduced substantial extra loss, confirming that the new sidewall approach is a superior path forward. The researchers verified their findings by adding known amounts of loss to the system and showing that their method could accurately detect and locate these changes, proving the reliability of their technique.

The implications of this work extend far beyond a single experiment. By providing a way to accurately measure and optimize these devices without damaging them, the researchers have offered a tool that can be used to test entire wafers of chips quickly and efficiently. This is a critical step toward mass-producing the complex optical circuits needed for future technologies. The ability to distinguish between a poorly made connection and a truly inefficient chip means that engineers can now focus on improving the actual performance of the materials rather than fighting against measurement errors. As the field moves toward more complex quantum systems and faster communication networks, having a clear, unbiased view of how these devices perform is no longer just a luxury; it is a necessity. The researchers have shown that with the right measurement strategy, the full potential of these advanced materials can finally be realized, paving the way for a new era of integrated photonics.

Technical Summary: Sidewall-Poled Nanophotonic Lithium Niobate with Bidirectional Characterization

Problem Statement

Accurate benchmarking of integrated nonlinear photonic devices is frequently compromised by unknown facet-coupling losses and fabrication-induced non-uniformities. Conventional single-pass characterization methods conflate intrinsic nonlinear conversion efficiency with extrinsic losses (coupling and propagation), leading to systematic over- or underestimation of on-chip performance. This is particularly critical for quantum applications, where sub-decibel errors can significantly distort inferred metrics such as squeezing and entanglement generation. Furthermore, characterizing longitudinal quasi-phase-matching (QPM) non-uniformities typically requires time-intensive microscopy or destructive imaging, which does not scale to wafer-level testing.

In the specific context of thin-film lithium niobate (TFLN), a leading platform for second-order ( $\chi^{(2)}$ ) nonlinearities, a trade-off exists between fabrication strategies. Traditional "poling-before-etching" approaches often result in sidewall corrugations and increased roughness due to differential etching of inverted domains, introducing excess propagation loss. While "etching-before-poling" (sidewall poling) strategies aim to mitigate this, rigorous, non-destructive methods to quantify their intrinsic performance and longitudinal uniformity have been lacking.

Methodology

The authors propose and demonstrate a unified bidirectional characterization framework that utilizes classical, phase-insensitive power measurements to decouple facet-specific coupling efficiencies from intrinsic nonlinear responses. The method relies on cascaded second-harmonic generation (SHG) and non-degenerate optical parametric amplification (NOPA) interactions.

1. Bidirectional Coupling Estimation

By pumping a cascaded SHG-NOPA process from opposite propagation directions (forward and backward), the system exploits the nonlinear dependence of the generated idler power on the coupled pump power.

Asymmetry Detection: In the low-gain regime, the ratio of forward to backward idler powers provides a direct estimate of the facet coupling asymmetry ( $\eta_1/\eta_2$ ).
MAP Estimation: To address bias in highly efficient or asymmetric cases, the authors employ a Maximum A Posteriori (MAP) estimator. This incorporates a prior on total transmission (derived from linear insertion-loss measurements) to independently extract input and output facet coupling efficiencies without requiring internal field access or quantum measurements.

2. Unbiased SHG Efficiency Extraction

The framework corrects for facet asymmetry in SHG efficiency measurements. By combining forward and backward SHG measurements, the authors derive an unbiased normalized conversion efficiency ( $\eta_{norm}$ ). The geometric mean of the apparent efficiencies, normalized by the total throughput at fundamental and second-harmonic wavelengths, cancels out facet asymmetry. For high-efficiency devices, the method utilizes pump-depletion regimes and numerically integrated coupled-mode equations to determine absolute on-chip efficiency from a single direction, provided input coupling is known.

3. Longitudinal Phase-Matching Reconstruction

The framework treats the recovery of the longitudinal QPM profile as a phase-retrieval problem. By measuring SHG efficiency spectra in both propagation directions, the authors obtain the squared magnitudes of the $z$ -transform of the nonlinear envelope function $h(z)$ on two concentric circles in the complex plane. This dual-circle data allows for a unique solution (up to a global phase), enabling the reconstruction of the longitudinal phase-mismatch function $h(z)$ and the identification of thin-film thickness variations without destructive imaging.

Key Results

The methodology was experimentally validated using sidewall-poled TFLN waveguides (2.1 cm length, ~1.3 $\mu$ m wide).

Coupling Loss Characterization: The bidirectional MAP estimator successfully resolved asymmetric facet losses. In a device with a visible defect, the method attributed 8.9 dB loss to the input facet and 5.8 dB to the output facet, isolating 3.1 dB of excess loss specifically at the defect site.
Nonlinear Efficiency: After correcting for coupling losses, the device exhibited a normalized SHG efficiency of $(1850 \pm 20) \% \text{W}^{-1}$ . At an on-chip fundamental power of 95 mW, the device achieved an absolute on-chip conversion efficiency of approximately 56%, among the highest reported for sidewall-poled TFLN.
Longitudinal Uniformity: The reconstructed phase-mismatch function $h(z)$ showed a slow spatial drift consistent with independent ellipsometry measurements of thin-film thickness variations. The amplitude of the envelope remained relatively uniform, indicating that duty-cycle errors were modest compared to phase drifts.
Broadband Parametric Gain: The devices demonstrated broadband non-degenerate optical parametric amplification (NOPA) and optical parametric generation (OPG).
- Gain: A phase-insensitive signal gain of 4.8 dB (phase-sensitive gain ~10 dB) was measured.
- Bandwidth: The OPG spectrum confirmed a 3-dB bandwidth exceeding 10 THz, enabled by dispersion engineering with low group-velocity dispersion (0.16–0.22 ps $^2$ /m).
Fabrication Comparison: Resonator-based loss extraction compared "poling-before-etching" and "sidewall-poling" strategies. Sidewall poling substantially suppressed poling-induced excess loss compared to the traditional approach. The sidewall-poled waveguides achieved a duty cycle of $42.5\% \pm 1.5\%$ , whereas poling-before-etching resulted in reduced uniformity ( $55.3\% \pm 5.8\%$ ).

Significance and Claims

The paper claims that this work provides a general, non-destructive route to rigorous benchmarking of integrated nonlinear photonic circuits. The significance lies in three main areas:

Decoupling Extrinsic and Intrinsic Performance: The framework eliminates the systematic biases inherent in single-pass measurements by independently extracting facet coupling losses and intrinsic nonlinear efficiency using only classical power measurements. This is essential for accurate performance prediction in quantum circuits.
Wafer-Scale Compatibility: Unlike microscopy or ellipsometric mapping, this method requires no time-intensive imaging or calibrated internal references, making it compatible with high-throughput, wafer-scale testing.
Optimization of TFLN Platforms: The study validates sidewall-poling as a superior fabrication strategy for minimizing propagation loss while maintaining strong nonlinear interactions. The bidirectional metrology closes the loop on fabrication optimization by providing a direct diagnostic of residual non-uniformities (e.g., film thickness drift) left after dicing and adaptive poling.

The authors conclude that these advances address central bottlenecks in scaling TFLN photonics: the trade-off between optical confinement and poling-induced loss, and the conflation of coupling losses with intrinsic device performance. This facilitates the scalable development of low-loss, high-modal-purity nonlinear devices for quantum information processing and communications.