Improving Optical Metrology by Engineering the Target Environment

Technical Summary: Improving Optical Metrology by Engineering the Target Environment

Problem Statement
In optical metrology, the ultimate limit on measurement precision is not the diffraction limit, but rather measurement noise. This limit is formalized by the Cramér-Rao lower bound, where the minimum achievable variance is the inverse of the Fisher information (FI). While strategies to improve precision typically focus on reducing detector noise or refining estimation algorithms, this work addresses the propagation of Fisher information itself. The authors posit that in coherent light systems, FI is a physical quantity that propagates, resonates, and interferes like an electromagnetic wave. A specific challenge arises in scattering-based position metrology: when a subwavelength object (such as a nanowire) undergoes lateral displacement, the resulting "information sources" at the object's edges often interfere destructively along the optical axis. Consequently, a detector centered on the axis—where optical energy is typically maximized—captures only a small fraction of the available Fisher information, limiting measurement precision.

Methodology
The authors propose an "information-based inverse design" approach to engineer the electromagnetic environment surrounding a measurement target. The core hypothesis is that by structuring the space around the target, one can optimize the generation and transmission of Fisher information to a finite-aperture detector.

• System Model: The study uses optical scattering-based measurements of the lateral position of a subwavelength nanowire suspended in a gap within an opaque screen.
• Theoretical Framework: The authors define a "Fisher information flow" vector ($\mathbf{S}_{FI}$) analogous to the Poynting vector. They demonstrate that for lateral displacement, FI sources are of equal magnitude but opposite phase, leading to destructive interference on the optical axis. The goal is to engineer a "reference field" (via surrounding structures) such that the phase difference between the sensitivity field and the scattered field maximizes the Fisher information density accessible via intensity-only measurements ($FI \propto (\partial_\theta I)^2 / I$).
• Optimization Strategy: Using numerical simulations (angular spectrum method and 3D Finite-Difference Time-Domain (FDTD)), the authors employ Bayesian optimization to determine the optimal widths and positions of additional parallel slits flanking the nanowire. The objective function is the maximization of detected Fisher information integrated over a central detector window.
• Experimental Validation: The optimized designs were fabricated on gold-coated silicon nitride membranes using focused ion beam milling. Experiments were conducted at $\lambda = 640$ nm. A neural network estimator, trained on a large dataset of scattering patterns corresponding to known nanowire positions, was used to retrieve unknown positions from single-shot scattering patterns.

Key Contributions and Results

1. Demonstration of FI Control: The work establishes that the flow of Fisher information can be controlled and redirected by structuring the target environment, much like energy flow.
2. Simulation Results: Numerical simulations showed that adding pairs of optimized slits significantly enhances detected FI. A single pair of slits increased detected FI by 39×, while four pairs yielded an enhancement of nearly 50× compared to an unstructured screen. This corresponds to a 7.8× increase in Fisher information per unit of detected power (per photon). The optimized structures create sub-diffraction limit hotspots and superoscillatory field features that are highly sensitive to source perturbations.
3. Experimental Performance: In experimental tests using a nanowire flanked by three pairs of optimized slits, the mean standard deviation of position measurements ($\sigma$) was reduced from 397 pm (unstructured case) to 61 pm. This represents a 6.5× improvement in localization precision, achieving a precision beyond $\lambda/10,000$.
4. Discrepancy Analysis: The experimental improvement (6.5×) slightly exceeded the simulation prediction based on FI enhancement ($\sqrt{28} \approx 5.3\times$). The authors attribute this to factors such as a slightly larger detector acceptance angle in the experiment than assumed in simulations, and the potential for fabrication imperfections or non-ideal wavefronts to provide beneficial information to the neural network training process.

Significance and Claims
The paper claims to establish "target environment engineering" as a powerful and broadly applicable strategy for advancing measurement and sensing performance. The significance lies in shifting the paradigm from solely improving detectors or algorithms to actively shaping the electromagnetic environment to optimize the physical flow of information. The authors assert that this approach can be applied across various platforms, ranging from the optical characterization of micro- and nano-objects to microwave radars and optical LiDAR navigation systems. The work demonstrates that substantial gains in precision are achievable not by overcoming fundamental physical limits, but by ensuring that the information generated by a physical process is efficiently coupled to the detection system.