Photonic-integrated quantum sensor array for microscale magnetic localisation

TL;DR

This work demonstrates a scalable, photonic-integrated platform for multi-NV quantum sensing, enabling eight nanoscale magnetic sensors to operate in parallel without bulk optics. By integrating NV ensembles on a silicon-nitride PIC and employing a CNN-based magnetic localisation pipeline, the authors achieve microscale localisation of a 30 $\mu$m needle with errors below the tip size and dynamic tracking capabilities. Simulations further show potential applications to magnetic microrobots, including position and orientation tracking with single-NV sensors, highlighting the platform’s relevance for biomedical and in vivo contexts. The approach promises robust, low-crosstalk sensing under real-world conditions and provides a pathway toward scalable, chip-based quantum sensing for medical and bioengineering applications.

Abstract

Nitrogen-vacancy centres (NVs) are promising solid-state nanoscale quantum sensors for applications ranging from material science to biotechnology. Using multiple sensors simultaneously offers advantages for probing spatiotemporal correlations of fluctuating fields or the dynamics of point defects. In this work, by integrating NVs with foundry silicon-nitride photonic integrated circuits, we realise the scalable operation of eight localised NV sensors in an array, with simultaneous, distinct readout of the individual sensors. Using the eight NV sensors and machine-learning methods for multi-point magnetic field reconstruction, we demonstrate microscale magnetic localisation of a 30 $μ$m-sized needle tip. Experimentally, the needle tip can be localised with an error below its dimension and tracked dynamically with high fidelity. We further simulate the feasibility of our platform for monitoring the position and orientation of magnetic microrobots designed for biological and clinical purposes. Without the complexity of bulk optics, our photonic-integrated multi-sensor platform presents a step towards real-life biomedical applications under out-of-the-lab conditions.

Figures (5)

• Figure 1: Multi-NV quantum sensing platform using photonic integrated circuits. a A schematic representation of multi-NV quantum sensing using photonic integrated circuits. NVs on the photonic waveguides are pumped through the top excitation fibre array and distinctly read out with PL coupled off-chip through the collection fibre array. The eight sensors are used collectively to magnetically localise the position of a microscale magnetic source. Enabled by the high-index-contrast photonic channel, magnetic localisation remains feasible when direct top-down visibility is limited (for example, obscured by a liquid). b Schematic diagram of the experimental setup. Eight NV-spin sensors are pumped by a 515 nm laser split into eight channels. The distinct PL signal is collected by parallel single-mode waveguides and sent to SPAD for detection. The eight SPAD signals are sent to a multi-channel time tagger for spin-state measurements. The NV spins are driven by a global microwave antenna (yellow ring in the diagram). c The silicon-nitride photonic chip. Experimentally, we use only eight out of the sixteen waveguides in parallel. The middle read area is unclad for nanodiamond positioning. The edge couplers (enclosed by the yellow dashed-line box) are fabricated with a deep etch to ensure surface smoothness. d Image of a nanodiamond array. Here, the nanodiamonds are visualised by scattered laser light when a red laser is coupled into the waveguides. Scattered spots look extended due to scattering of light around the alignment markers. e A scanning-electron-microscopic image of a lithographically-positioned nanodiamond site, showing a cluster of nanodiamonds in the rectangular window. f The schematic diagram of waveguide tapering to the edge couplers. The 150 nm end size is limited by the foundry minimum feature size. g Comparison of the coupling efficiency over the NV spectrum (blue) for edge couplers (green) and grating couplers (yellow). Wavelength abbreviated Wl. in the plot. h Experimental setup, showing the excitation and collection fibre array, the chip, and microwave delivery through a PCB antenna.
• Figure 2: Parallel sensor operation and characterisation. a The PL signal of the eight NV sensors collected from each waveguide when the excitation (exc.) fibre array is scanned. Here, the X direction is parallel to the waveguides. Fitted centres of the Gaussian spots are labelled in µ m. The colour bar unit is counts per second. b The CW ODMR measurements of the eight NV sensors (NV 1-8) in parallel under zero field. Microwave frequency is abbreviated as MW. freq. c Experimental Zeeman-splitting showing the NV ensemble response. This corresponds to simulations in d. By fitting the resonance with Gaussian dip(s), the FWHM width and contrast can be extracted. e Characterisation of the Zeeman splitting for the eight sensors. Linear splitting is found with respect to the external magnetic field strength. f Characterisation of the CW-ODMR sensitivity as a function of magnetic field strength. The insensitivity (increased $\eta_\textrm{CW-ODMR}$) under higher field is due to resonance broadening and reduced contrast.
• Figure 3: Experimental magnetic localisation with multi-point field reconstruction.a Image showing the magnetic needle tip above the chip, the experimental (exp.) operation area, and the sensor relative positions. Note that the same X-Y coordinates (the needle tip motor stage reference frame) in µ m are used throughout this figure. b The Zeeman splittings of each NV sensor when the tip moves in the X direction, showing increased gradient. c The Zeeman splittings recorded when the tip moves in the Y direction showing the translation in peak field position. d Experimental magnetic localisation. The sensor array is used to estimate the needle position in X and Y. The experimental operation area is enclosed by the grey box with the sensor positions labelled. e Residual analysis for the magnetic localisation result in d, comparing the needle tip X and Y positions. f The linear decrease of averaged estimation error with increased number of sensors used simultaneously. g The $1/\sqrt{t}$ scaling of the average estimation error with the integration time $t$. All eight sensors are used here. h Relationship between average estimation error and training dataset size $N$, showing $1/\sqrt{N}$ scaling. All eight sensors are in use here with a 360 s integration time.
• Figure 4: Experimental dynamical tracking of moving magnetic object. The needle tip, when moving at different speeds, is tracked with different frame rates for comparison. Note that the same machine-learning model is used for magnetic localisation and the trajectory falls within the operation area in Fig. \ref{['fig3']}a. The deviation of the reconstructed (rec.) trajectory from the true trajectory is quantified by the Mean Perpendicular Distance (MPD), which increases with higher frame rate due to shot noise and higher moving speed due to undersampling.
• Figure 5: Application to position and orientation tracking of magnetically-driven microrobot. a Targeted application scenario. The cartoon diagram shows that a 100-µ m sized magnetically-driven microrobot, on a mission of pin-point drug delivery, requires the ability to navigate through optically-inaccessible environments with precise control. With our magnetic localisation technique, the microrobot can be efficiently localised and tracked when performing biomedical tasks using an array of NV ensemble or single-NV sensors. Note that the magnetic field used to drive the microrobot is homogeneous across the sensors whose contribution can be easily eliminated in the magnetic localisation calculations. b Error map of simulated microrobot magnetic localisation, with eight ensemble NV sensors. The robot-position-dependent estimation error is plotted with the relative positions of the NV ensemble sensors labelled. The good operation area (with estimation error $\leq50$ µ m) is shown by the dashed-line box. c The y-averaged estimation error and the averaged magnetic field gradient at the sensor locations for different robot x positions. When the robot moves away from the sensors, the estimation error increases as a result of decreased field gradient. See also Supplementary Section D for the microrobot magnetic field profile. The dashed vertical line at x=-750 µ m shows where the good operation range is defined (when estimation error stays $\leq 50$ µ m). d Position tracking of the microrobot along a trajectory (abbreviated as traj.) with eight single NV sensors. The single-NV sensor positions are labelled. We consider the single NV sensors only 150 µ m below the microrobot since their superior sensitivity at higher field allows the robot to be tracked with a smaller z standoff. e Monitoring of microrobot rotations. The change of orientations (abbreviated as ori.) alongside the trajectory (as visualised by the robot diagram) can be precisely tracked with the single NV sensors. Arrows in the plot show the true and estimated orientations for every two points along the trajectory.