betaFeedback

WorksPhysics and Astronomy

Quantum Frequency Mixing using an N-VV Diamond Microscope

Abstract

Wide-field magnetic microscopy using nitrogen-vacancy (NV) centers in diamond can yield high-quality magnetic images of DC and AC magnetic fields. The unique combination of micron-scale spatial resolution of scalar or vector fields at room temperature and parallel camera readout make this an appealing technique for applications in biology, geology, condensed-matter physics, and electronics. However, while NV magnetic microscopy has achieved great success in these areas, historically the accessible frequency range has been limited. In this paper, we overcome this limitation by implementing the recently developed technique of quantum frequency mixing. With this approach, we generate wide-field magnetic images of test structures driven by alternating currents up to 70 MHz, well outside the reach of DC and Rabi magnetometry methods. With further improvements, this approach could find utility in hyperspectral imaging for electronics power spectrum analysis, electronics diagnostics and troubleshooting, and quantum computing hardware validation.

Introduction

Wide-field magnetic microscopy using nitrogen-vacancy (NV) centers in diamond has gained attention for its value in diverse applications including geology, biology, condensed-matter physics, and electronics [1, 2]. Unlike single-pixel scanning magnetic microscopes, NV-diamond-based microscopes can simultaneously image all pixels in a wide (mm-scale) field of view (FOV) without moving parts, and can reveal spatiotemporal dynamics of magnetic fields in samples under study. The technique offers micron-scale spatial resolution and the potential for operation over a broad frequency range, and is being applied to a range of interdisciplinary applications.

Despite the ongoing accomplishments and progress, most wide-field NV magnetic microscopes have been limited to measuring static or low-frequency (<1< 1 kHz) magnetic fields, with few exceptions [3, 4, 5, 6, 7]. This is due to the added complexity required for measuring MHz-frequency and GHz-frequency AC magnetic fields (including pulsed NV interrogation and gated optical readout) and the greater need for homogeneous laser, microwave (MW), bias magnetic field, and NV strain profiles across the FOV. Moreover, these measurements are restricted to specific and often narrow frequency ranges. For example, AC magnetometry using pulsed dynamical decoupling is limited by MW π\pi-pulse durations to typically <10< 10 MHz [8], while Rabi-oscillation AC magnetometry senses in a narrow frequency range (∼1\sim1 MHz) around the NV spin resonance frequency (usually near ∼3\sim3 GHz) [9]. Though varying a strong bias magnetic field can tune (and extend) the Rabi-oscillation AC magnetometry frequency range, field uniformity requirements render this method impractical for macroscopic NV ensembles. As a result, NV AC magnetic imaging demonstrations often sacrifice frequency range, FOV size, or both.

Here we demonstrate a pulsed diamond magnetic microscope with a 1.5×1.51.5 \times1.5 mm2^2 FOV size, and image DC and AC magnetic fields up to 70 MHz. To our knowledge, this is the largest-area pulsed NV magnetic imaging demonstration, as previous experiments have shown ∼0.5\sim0.5 mm2^2 at best [5]. To enable wide-frequency sensitivity, we apply the quantum frequency mixing (QFM) approach recently demonstrated with small NV ensembles [10] to our wide-field imager. We validate the technique by imaging AC magnetic fields from currents in a straight-wire test structure patterned on a glass slide, where we observe the expected frequency dependence of detected signals. We then image magnetic fields from a second test structure (an Archimedean spiral), demonstrating agreement between DC and AC images. Finally, we explore the QFM magnetometry dynamic range under different signal frequencies and amplitudes, leading to additional analytical expressions of the expected behavior. Combining high-performance wide-field NV-diamond microscopy with the QFM AC magnetometry technique opens the door to new magnetic microscopy applications, including high-resolution power spectrum analysis (PSA) imaging of integrated circuits, electronics failure analysis, validation of ∼\sim1-100 MHz electrical components, and even diagnostics of quantum computing hardware (e.g. ion trap chips) [11–13].

Quantum Frequency Mixing

Frequency mixing is a process that uses a nonlinear response to generate new frequencies from two periodic input signals. One example classical mixer, the unbalanced single-diode mixer, uses the nonlinear relationship between input voltage and response current to multiply two input signals, generating new frequencies at the sum and difference of the input frequencies. In our work using NV centers in diamond as a quantum frequency mixer, the NV quantum system is nonlinear, and we exploit this to multiply two AC magnetic signals within the quantum system itself [10]. The type of mixing employed here is heterodyne, and the net effect of our experiment is implementation of a heterodyne receiver in a quantum system.

When a quantum system is driven with two fields at different frequencies, its dynamics can be described as a quantum frequency mixer, where the effective Hamiltonian derivation in the multi-mode Floquet picture allows us to identify the dynamics to describe a frequency-mixer-based quantum sensing protocol [10]. Here, we summarize the principle of quantum frequency mixing and its application to this work.

To bring a signal of frequency ωs\omega_s into resonance with

an NV transition frequency ω0\omega_0, we apply a bias oscillating field with a frequency ωb\omega_b. Here ωs\omega_s and ωb\omega_b are analogous to the input radio-frequency (RF) and local oscillator (LO) of a classical mixer. The effective signal generated by the nonlinear process of the quantum system is at frequency ωT\omega_T, which is analogous to the intermediate frequency (IF) of a classical mixer.

This process can be described by Floquet theory [14, 15]. In Floquet analysis, the time-dependent Hamiltonian in the Hilbert space is converted to a time-independent Hamiltonian in Floquet space. The combined effect of two Fourier components HseiωbtH_{se^{i\omega_b t}} and HseiωstH_{se^{i\omega_s t}} associated with the input signal ωs\omega_s and the bias AC field ωb\omega_b, respectively, gives rise to a component at the target frequency ωT\omega_T. Upon block-diagonalization of the Floquet Hamiltonian, transforming back to the Hilbert space yields an effective Hamiltonian with a frequency ωb−ωs\omega_b-\omega_s [10]. With an appropriate choice of ωb\omega_b, the effective target signal is now at the NV resonance frequency

::: near 3 GHz, and existing quantum sensing protocols such as Rabi magnetometry can be used to probe the system.

ParameterDescription
ω0\omega_0NV resonance frequency
Ω0\Omega_0On-resonance Rabi frequency
ωb\omega_bBias frequency
Ωb\Omega_bBias amplitude
ωs\omega_sSignal frequency
Ωsz\Omega_{sz}Signal amplitude along zz
ωT\omega_TQFM target frequency
ΩQFM\Omega_{\mathrm{QFM}}QFM Rabi frequency

Table 1. A list of relevant parameters describing the QFM AC magnetometry protocol.

The physics of a signal magnetic field projection along the NV axis, with frequency ωs\omega_s, detected using a Rabi oscillation measurement protocol (see § 6 C below) can be described with the Hamiltonian:

H=ω02σz+Ωbcos⁡(ωbt+ϕb)σx+Ωszcos⁡(ωst+ϕs)σz,(1)H = \frac{\omega_0}{2}\sigma_z + \Omega_b \cos\left(\omega_b t + \phi_b\right)\sigma_x + \Omega_{sz}\cos\left(\omega_s t + \phi_s\right)\sigma_z, \tag*{(1)}

where Ωb\Omega_b is the amplitude of the bias field and Ωsz\Omega_{sz} is the amplitude of the signal frequency projected along the NV axis. In a rotating frame defined by U=e−i(ω0t/2)σzU=e^{-i(\omega_0t/2)\sigma_z}, the effective Hamiltonian contains two oscillating frequencies. When one is resonant with the NV transition, and the other is far-detuned and can be neglected in the spirit of the rotating-wave approximation. We choose a bias frequency ωb\omega_b such that

ωb−ω0−ωs=δz,(2)\omega_b-\omega_0-\omega_s=\delta_z, \tag*{(2)}

where ω0\omega_0 is the NV resonance frequency and δz\delta_z is a small detuning which we try to minimize in the experiment. Assuming Ωsz,Ωb≪ωs\Omega_{sz},\Omega_b\ll\omega_s, the amplitude of the effective target Rabi frequency is [10]

ΩQFM=ΩszΩbωs.(3)\Omega_{\mathrm{QFM}}=\frac{\Omega_{sz}\Omega_b}{\omega_s}. \tag*{(3)}

Table I lists the relevant parameters for this experiment.

Experimental Methods

Microscope Apparatus

Figure 1a shows a schematic of the microscope apparatus. This drawing shows the relevant coordinate systems: {x,y,z}\{x,y,z\} is the NV coordinate system (with zz along the N-V axis, which is also the diamond [111] crystallographic axis) and {X,Y,Z}\{X,Y,Z\} is the microscope coordinate system (with ZZ out of plane for the microscope image). A 5×5×0.5 mm35\times5\times0.5\ \mathrm{mm}^3 diamond sample with a 10 μm10\ \mu\mathrm{m} NV layer is illuminated with 1 W of 532 nm laser light, which is passed through an acousto-optic modulator (AOM) switch and flat-top beam shaper. We place the test structures on top of the NV layer, which is coated with 10 nm of Ti metal and 50 nm of Au to prevent light leakage. The diamond is attached to a printed circuit board (PCB) mount that applies a uniform MW magnetic field along the xx direction (which is also the lab-frame XX direction), and the NVs detect DC magnetic fields (with Ramsey magnetometry [16]) and AC magnetic fields (with QFM) along the zz axis. The mount also acts as a heat sink to dissipate heat deposited by the laser. We applied a ∼1\sim1 mT bias magnetic field B0B_0 along the +z+z direction using permanent magnets, which modifies the NV ground-state transition frequencies through the Zeeman effect. This bias field sets the transition frequencies between NV ground-state sublevels to be D±γB0D\pm\gamma B_0, where D≈2π×2870D\approx2\pi\times2870 MHz is the zero-field splitting and γ≈2π×28\gamma\approx2\pi\times28 GHz/T is the NV gyromagnetic ratio, and we address the ∣0⟩↔∣+1⟩\lvert0\rangle\leftrightarrow\lvert+1\rangle transition at ω0≈2π×2890\omega_0\approx2\pi\times2890 MHz. NV fluorescence light is imaged with an optical microscope onto a digital focal-plane array (DFPA) camera, which has high sensitivity, large dynamic range, 256×256256\times256 pixel resolution, on-chip processing, and fast data rates required for this work [17]. Laser, MW, test structure current, and camera trigger pulses are programmed with a 1.2 GS/s arbitrary waveform generator (AWG). The camera is triggered to acquire an NV fluorescence read-out image across a range of MW pulse durations, and we fit the resulting time traces in each pixel to extract the Rabi frequency for QFM magnetic imaging (Figure 1c).

Image not extracted

Figure 1. (a) Schematic diagram of a QFM magnetic microscope. A green laser beam illuminates the NV layer. Different regions of the NV layer (highlighted by blue and red spins) respond to spatially varying magnetic fields and produce magnetic-field-dependent fluorescence, which is imaged onto a camera sensor using imaging optics. An example test structure (an Archimedean spiral), generating a spatially varying magnetic field, is shown at the sensor plane, and a magnetic map of its current is imaged at the focal plane. (b) The NV spins act as both quantum mixers and quantum sensors. In the presence of a signal field (yellow) at ωs\omega_s and a bias field (teal) at ωb\omega_b, a target field (green) at the sum or difference frequency ωT\omega_T is generated. When ωb\omega_b is chosen so that the sum or difference frequency is resonant with the NV resonance at ω0\omega_0, population oscillations occur between the two spin states. (c) The oscillation frequency, called the target Rabi frequency ΩQFM\Omega_{\mathrm{QFM}}, provides a measure of the signal field amplitude Ωsz\Omega_{sz}, given by Eq. (3). Spatial variations in Ωsz\Omega_{sz} result in variations in ΩQFM\Omega_{\mathrm{QFM}} from pixel to pixel, allowing wide-field imaging of AC signal amplitudes at arbitrary ωs\omega_s.

Test Structures

We fabricated two conducting test structures on a glass microscope slide using photolithography and electron-beam evaporation (10 nm Ti, 2 μ\mum Al, 10 nm Ti, 150 nm Pt, 200 nm Au), a straight wire and an Archimedean spiral (both with 50 μ\mum trace width), as shown in Figure 3a and Figure 4a. We attached wire leads to the test structures using silver paint (yielding few-Ω\Omega total resistances), after which they were placed on the diamond sample. When applying DC and AC currents using a function generator, we included a 50 Ω\Omega load resistor in series and measured the voltage across it to validate the DC and AC current amplitudes through the test structures. To avoid the possibility of shorting the test structures through the conductive film on the NV layer, we placed a 7.5 μm\mu\mathrm{m} Kapton film between the slide and the diamond.

Image not extracted

Figure 3. (a) Photograph of the straight-wire test structure. (b-c) Two QFM imaging examples for an AC current at ωs=2π×4\omega_s = 2\pi\times4 MHz and 2π×352\pi\times35 MHz (top). The Rabi oscillations and the extracted Rabi frequency for the center pixel (white cross) are also included (bottom). As expected, ΩQFM\Omega_{\mathrm{QFM}} at a specified pixel decreases as ωs\omega_s increases. (d) ΩQFM\Omega_{\mathrm{QFM}} for the center and right (green star) pixels as a function of ωs\omega_s and fit to an a/ωsa/\omega_s curve, demonstrating the inverse relationship described by Eq. (3). The pixel marked with a blue × is discussed in § 7 C.

Image not extracted

Figure 4. (a) Photograph of the Archimedean spiral test structure. (b) DC magnetic image of the spiral with 1 mA signal current measured using a Ramsey sequence. (c-d) Two examples of QFM imaging (top), with the extracted ΩQFM\Omega_{\mathrm{QFM}} for an example pixel (bottom). ΩQFM\Omega_{\mathrm{QFM}} for this pixel decreases by a factor 1/ωs1/\omega_s as the input signal frequency increases from ωs=2π×5\omega_s = 2\pi\times5 MHz to 2π×502\pi\times50 MHz.

Experimental Procedure

Figure 2 shows the pulse sequences used for magnetic imaging experiments. First, we perform a conventional Rabi oscillation measurement [5] with no current in the test structure (Figure 2a). To do this, we initialize the NVs to the ∣0⟩\lvert0\rangle state with a laser pulse, apply an on-resonance MW field with a variable duration τ\tau, and read out the NV final-states with a second laser pulse and the camera. Fitting the NV fluorescence intensity as a function of τ\tau for each pixel, this yields a map of the on-resonance Rabi frequencies Ω0\Omega_0 (and implicitly the Ωb\Omega_b bias amplitudes), and also the π/2\pi/2 pulse duration used for Ramsey magnetometry.

Image not extracted

Figure 2. (a) Conventional Rabi pulse sequence used to determine the on-resonance NV Rabi frequencies, with no current applied to the test structure. We initialize the NVs to the ∣0⟩|0\rangle state with a laser pulse, apply an on-resonance MW field of frequency ω0\omega_0 for a variable duration, and read out the NV final state with a second laser pulse. (b) Ramsey pulse sequence used for DC magnetic imaging. We apply a DC current to the test structure, and we apply a π/2\pi/2 MW pulse to create a superposition of the ∣0⟩|0\rangle and ∣+1⟩|+1\rangle states and allow the system to evolve for a fixed time τ\tau. We then apply a second π/2\pi/2 pulse, followed by a laser readout pulse. (c) Quantum frequency mixing (QFM) Rabi pulse sequence used to measure an AC magnetic field amplitude at frequency ωs\omega_s. This is similar to the conventional Rabi oscillation experiment, except with an applied bias frequency ωb\omega_b.

Next, we apply a DC current to the test structure and perform Ramsey magnetometry (Figure 2b). This ensures that the test structure is centered in the camera field of view and allows us to confirm that the test structure is flat on the diamond surface. We use double-quantum Ramsey spectroscopy for improved magnetic sensitivity and to avoid frequency shifts due to temperature and strain inhomogeneity [16, 18]. After laser initialization, the first π/2\pi/2-pulse places the NVs in a superposition on the equator of the Bloch sphere, and they accumulate a phase ϕ=2γΔBτ\phi= 2\gamma\Delta B\tau, where ΔB\Delta B is the field from the test structure and τ\tau is the phase accumulation time, then apply a second π/2\pi/2-pulse and a readout laser pulse. After optimizing the Ramsey imaging measurement parameters, we image the DC magnetic field map from the test structure. This provides a DC magnetic image to compare with subsequent AC QFM magnetic images.

Finally, we apply an AC current at frequency ωs\omega_s to the test structure, a MW field at ωb=ω0+ωs\omega_b = \omega_0 + \omega_s, and measure QFM Rabi oscillations (Figure 2c). The QFM experiment protocol is nearly identical to the Rabi oscillation experiment protocol, with the main difference being that we address the NVs with frequency ωb\omega_b instead of ω0\omega_0. Fitting the fluorescence time-trace data to extract ΩQFM\Omega_{\mathrm{QFM}}, we can then use Eq. (3) to get the AC magnetic field amplitude for each pixel (Bapplied=Ωsz/γB_{\mathrm{applied}} = \Omega_{\mathrm{sz}}/\gamma).

To improve the long-term stability and pixel-to-pixel noise floor of our apparatus, each of the above experiments modulates the signal amplitude between alternate camera exposures (either on/off for Rabi and QFM or positive/negative DC current for Ramsey) [5] [18, 19].

Results

Validation with a Straight Wire Test Structure

Figure 3a shows a photograph of the straight-wire test structure. Following the measurement procedure described above, we first imaged Rabi oscillations in the field of view with no current in the wire, then measured a 1 mA DC current through the wire to check its position in the field of view. The DC magnetic image yields a conservative estimate that the standoff distance between the NV layer and the wire is ∼\sim50 μ\mum.

For the QFM magnetic imaging measurement, we applied an 80 mA peak-amplitude AC current through the test structure and measured the projection of the created magnetic field along the zz-axis. The acquisition time for each ωs\omega_s image was ∼\sim5 minutes, roughly half of which was dead time due to camera transfer lag [20]. The Rabi frequency ΩQFM\Omega_{\mathrm{QFM}} in each pixel was extracted using the fit function C(τ)=Acos⁡(ΩQFMτ+ϕ)e−τ/TRabi+C0C(\tau) = A\cos(\Omega_{\mathrm{QFM}}\tau+ \phi)e^{-\tau/T_{\mathrm{Rabi}}} + C_0 where CC is the normalized fluorescence intensity [20], AA is the Rabi contrast amplitude, ϕ\phi is a phase offset, TRabiT_{\mathrm{Rabi}} is the Rabi coherence time, and C0C_0 is an offset. Figure 3b-c maps the ΩQFM\Omega_{\mathrm{QFM}} values for ωs=2π×4\omega_s = 2\pi\times4 MHz and 35 MHz, including the QFM Rabi oscillations for the center (white cross) pixel. The black stripe corresponding to ΩQFM\Omega_{\mathrm{QFM}} near 0 is where the wire magnetic field is perpendicular to the zz direction.

Figure 3d plots ΩQFM\Omega_{\mathrm{QFM}} as a function of ωs\omega_s for two pixels. As expected, ΩQFM\Omega_{\mathrm{QFM}} decreases as ωs\omega_s increases. These data are fitted with the fit function ΩQFM=a/ωs\Omega_{\mathrm{QFM}} = a/\omega_s (where aa is a free parameter), and are consistent with Eq. (3). For both pixels, Ωb\Omega_b was experimentally determined to be ∼2π×1.0\sim2\pi\times1.0 MHz by measuring the on-resonance Rabi oscillation frequencies (Figure 2a) and assuming the MW amplitude is the same at ωb\omega_b. Ωsz\Omega_{\mathrm{sz}} is estimated to be ∼2π×1.8\sim2\pi\times1.8 MHz and ∼3.8\sim3.8 MHz for the center and right (green star) pixels, respectively, based on the simulated magnetic field of a 50 μ\mum ribbon at ∼50\sim50 μ\mum standoff distance. The fitted aa coefficients are (2π)2×{1.3,2.3}(2\pi)^2 \times\{1.3, 2.3\} MHz2^2 for the center and right pixels, and given Ωb≈2π×1.0\Omega_b \approx2\pi\times1.0 MHz for both pixels, this yields Ωsz≈2π×{1.3,2.3}\Omega_{sz} \approx2\pi\times\{1.3, 2.3\} MHz. These extracted values are reasonably consistent with the estimated Ωsz≈2π×{1.8,3.8}\Omega_{sz} \approx2\pi\times\{1.8, 3.8\} MHz.

Previously, QFM Rabi magnetometry was demonstrated for a bulk (single-pixel) NV ensemble [10]. The above straight-wire test structure demonstration shows that this technique can also be applied to an NV magnetic imaging apparatus, and validates the expected frequency-scaling law predicted in Eq. (3).

Archimedean Spiral Test Structure

Having demonstrated that the QFM imaging method works as expected for a simple test case, we now apply it to a more complicated test structure, an Archimedean spiral (Figure 4a). Unlike the straight-wire, the spiral test structure has currents (magnetic fields) pointing in a wide range of directions, leading to a nontrivial magnetic field map. The test structure layout is similar to a planar spiral inductor, and generalizing the measurement technique to spiral inductors and other similar passive RF circuit elements is straightforward. Furthermore, since the spiral is a passive circuit element, we control the frequency and amplitude of current passing through it. By contrast, when measuring the magnetic fields from a more sophisticated IC, these parameters are likely uncontrollable (and are often unknown).

Figures 4b-c show QFM magnetic images for ωs=2π×5\omega_s = 2\pi\times5 MHz and 50 MHz, with 80 mA peak-amplitude current. Rabi oscillation data and fits for an example pixel are included, which also show the ΩQFM∝1/ωs\Omega_{\mathrm{QFM}} \propto1/\omega_s dependence expected from Eq. (3). Figure 4d includes a DC magnetic image for comparison, acquired using Ramsey magnetometry and showing a similar magnetic field amplitude structure. Note that the DC experiment measures both positive and negative magnetic field values while the AC experiments measure amplitudes, though the absolute value of the DC image is consistent with the AC images.

Large-sz Small-s Regimes

In Figure 3 and Figure 4, we used Eq. (3) to confirm that the measured QFM Rabi frequencies ΩQFM\Omega_{\mathrm{QFM}} were behaving as expected. However, this expression is valid when

Ωb,Ωsz≪ωs,(4)\Omega_b,\Omega_{sz} \ll\omega_s, \tag*{(4)}

which is not necessarily true for all experimental conditions. For large Ωsz\Omega_{sz} and small ωs\omega_s, ΩQFM\Omega_{\mathrm{QFM}} is more accurately described by

ΩQFM=ΩbJ1(2Ωszωs),(5)\Omega_{\mathrm{QFM}} = \Omega_b J_1\left(\frac{2\Omega_{sz}}{\omega_s}\right), \tag*{(5)}

where J1J_1 is the Bessel function of the first kind [20]. Figure 5a plots ΩQFM\Omega_{\mathrm{QFM}} as a function of ωs\omega_s for the left (blue ×) pixel in Figure 3, showing the necessity of Eq. (5).

This pixel has a larger Ωsz=2π×2.5\Omega_{sz} = 2\pi\times2.5 MHz amplitude than the right pixel because of its proximity to the wire, which we estimated using the ωs≥2π×30\omega_s \ge2\pi\times30 MHz tail and Eq. (3). We also measured its ΩQFM\Omega_{\mathrm{QFM}} frequencies down to a lower-bound ωs=2π×2\omega_s = 2\pi\times2 MHz. These ΩQFM\Omega_{\mathrm{QFM}} frequencies were measured with the same 80 mA peak-amplitude AC current as in Figure 3. Figure 5a verifies that Eq. (5) (solid black line) is a more appropriate description of the ΩQFM\Omega_{\mathrm{QFM}} behavior for small ωs\omega_s than Eq. (3) (blue dashed line).

Figure 5b shows ΩQFM\Omega_{\mathrm{QFM}} as a function of Ωsz\Omega_{sz} (i.e. applied current) for the center pixel previously examined in Figure 3, using ωs=2π×5\omega_s = 2\pi\times5 MHz. For sufficiently large Ωsz\Omega_{sz}, the measured ΩQFM\Omega_{\mathrm{QFM}} deviates from the linear dependence predicted by Eq. (3) (blue dashed line), and is more accurately described by Eq. (5) (solid black line).

Discussion

Our analysis shows the dynamic range of the QFM AC magnetometry and imaging method. In the large-Ωsz\Omega_{sz} and small-ωs\omega_s regime, conversion from ΩQFM\Omega_{\mathrm{QFM}} to Ωsz\Omega_{sz} (which is straightforward using Eq. (3)) requires inverting Eq. (5). ΩQFM\Omega_{\mathrm{QFM}} does not scale monotonically with ωs\omega_s and Ωsz\Omega_{sz}, and near a maximum in ΩQFM\Omega_{\mathrm{QFM}}, ΩQFM\Omega_{\mathrm{QFM}} is insensitive to Ωsz\Omega_{sz}. These details may complicate the QFM AC magnetometry method when applying it to more general experimental situations.

Equation 3 implies that one can increase Ωb\Omega_b to compensate for a small Ωsz\Omega_{sz}, since a slow ΩQFM\Omega_{\mathrm{QFM}} is difficult to measure if its period is slow compared to the coherence time TRabiT_{\mathrm{Rabi}}. However, this may also require additional analysis to interpret the measured ΩQFM\Omega_{\mathrm{QFM}} values [20]. Alternatively, in a regime where ΩQFM\Omega_{\mathrm{QFM}} is slow compared to the TRabiT_{\mathrm{Rabi}} coherence time, one can apply an on-resonance π/2\pi/2-pulse at frequency ω0\omega_0 before the QFM Rabi pulse at frequency ωb\omega_b, making the NV fluorescence intensity proportional to sin⁡(ΩQFMT+ϕ)e−T/TRabi\sin(\Omega_{\mathrm{QFM}}T+\phi)e^{-T/T_{\mathrm{Rabi}}} instead of cos⁡(ΩQFMT)e−T/TRabi\cos(\Omega_{\mathrm{QFM}}T)e^{-T/T_{\mathrm{Rabi}}} [9]. For small ΩQFM\Omega_{\mathrm{QFM}}, the fluorescence change will be proportional to ΩQFMT\Omega_{\mathrm{QFM}}T (sometimes called “sine magnetometry”) instead of (ΩQFMT)2(\Omega_{\mathrm{QFM}}T)^2 (“cosine magnetometry”) for improved sensitivity and minimum-detectable ΩQFM\Omega_{\mathrm{QFM}}, though implementing this also requires careful calibration [8].

Conclusion and outlook

In this work we demonstrated an NV wide-field magnetic imaging apparatus capable of measuring AC magnetic fields using QFM up to 70 MHz. We implemented a Rabi-based magnetometry scheme with linearly-polarized signal and bias fields to measure a signal-field component oscillating along NV zz-axis. The imager itself overcomes several of the common challenges for NV pulsed magnetic imaging, achieving sufficient homogeneity to realize a ∼1.5×1.5\sim1.5 \times1.5 mm2^2 FOV. After validating that the technique works with a simple straight-

caption: (a) ΩQFM\Omega_{\mathrm{QFM}} as a function of ωs\omega_s with a fixed Ωsz\Omega_{sz} for the left pixel in Figure 3. The dashed pink line shows the expected ΩQFM\Omega_{\mathrm{QFM}} predicted by Eq. (3). The black line shows the expected ΩQFM\Omega_{\mathrm{QFM}} from Eq. (5), and the blue dots are the experimental data. (b) ΩQFM\Omega_{\mathrm{QFM}} as a function of Ωsz\Omega_{sz} at a fixed ωs\omega_s for the center pixel in Figure 3.

:::

wire test structure, we then imaged an Archimedean spiral test structure to show that this technique can be generalized to spiral inductors and other passive RF circuit elements. Additional extensions could use spin-locking and pulsed dynamical decoupling methods (such as CPMG), or use circularly-polarized signal and bias fields to measure a signal field oscillating transverse to the NV zz-axis[eq] [10].

Building on previous demonstrations using NV magnetic imaging of DC currents for electronics interrogation and failure analysis (FA) [13, 21], our QFM magnetic imaging approach could also be applied to FA, fabrication process monitoring, and non-contact wafer testing. This would generalize the technique to include AC magnetic field sources at arbitrary frequencies. One could measure the frequencies and amplitudes of currents in high-speed test structures (e.g. ring oscillators [22]), which are included on dice or wafers for fabrication process troubleshooting. To achieve this, one could use QFM magnetic imaging to implement a spectrum-analyzer imager (a form of hyperspectral imaging), generating image for a range of ωs\omega_s values [11, 23]. The resulting hyperspectral data cube could then be analyzed for PSA applications, such as counterfeit detection, component aging, and fabrication process variation over time or across a wafer [24, 25]. This could also be useful for validating how internal current paths within a device change as a function of frequency and impedance (for example, high-frequency capacitive shorting).

As a further extension, our QFM magnetic imaging approach could be adapted for space-domain reflectometry for open-circuit fault localization. This was previously demonstrated with a scanning RF SQUID sensor [12], but could potentially achieve improved spatial resolution and sensitivity with an NV magnetic microscopy apparatus. Finally, this technique could be used to evaluate the RF and MW images of hardware used for quantum computing, such as ion trap chips and superconducting qubits. We note the emerging synergy between quantum technologies, and here one quantum technology (quantum sensing and imaging) would troubleshoot hardware of another technology (quantum computing).

Acknowledgements

We thank M. Ricci for help with test structure fabrication and packaging, C. Reed for help with preparing Figure 1a, and M. Jordan, C. Nordquist, and D. Krawczyk for useful discussions.

References

Supplemental Material for “Quantum Frequency Mixing using an N-V Diamond Microscope”

Samuel J. Karlson; Pauli Kehayias; Jennifer M. Schloss; Guoqing Wang; Andrew C. Maccabe; Adam Libson; David F. Phillips; Paola Cappellaro; Danielle A. Braje

Experiment Dead-Time Assessment

For an experiment measuring QFM Rabi oscillations (Figure 2c in the main text), each pulse sequence takes 125 μs\mu\mathrm{s}. The pulse sequence includes a 25 μs\mu\mathrm{s} laser initialization pulse, a MW pulse ranging from τ=50\tau= 50 ns to 31 μs\mu\mathrm{s}, and a 8 μs\mu\mathrm{s} fluorescence readout laser pulse, with the total pulse sequence lasting 65 μs\mu\mathrm{s}. For each MW pulse duration, the sequence repeats for 32 camera exposures (125 μs×32=4\mu\mathrm{s} \times32 = 4 ms), and the fluorescence counts are summed and off-loaded from the camera to the PC frame grabber card as one “frame”, taking 0.5 ms to off-load. This corresponds to 4.5 ms spent acquiring per MW pulse duration.

This procedure is repeated for each of the 200 MW pulse durations used, where every other time point disables the bias field (sets Ωb=0\Omega_b = 0) for background subtraction. This yields two stacks of 100 images, which take 0.9 s to acquire. Additional software lag extends this to ∼1.5\sim1.5 s per “capture” of a full stack of QFM Rabi oscillation images. Of this 1.5 s, roughly 0.1 s (camera transfer) + 0.6 s (software lag) = 0.7 s is dead time, though definitions of “dead time” may also vary.

The total acquisition time for the time-averaged images in Figure 3 and Figure 4 in the main text is ∼\sim5 minutes (200 captures).

Fluorescence Normalization

NV fluorescence intensities are often normalized to the ∣0⟩\lvert0\rangle state fluorescence intensity. We present our results differently because the DFPA camera uses on-chip subtraction, reporting intensity differences, not the original intensities [S1]. Fortunately, since we extract Rabi frequencies from the time traces of each pixel, the units of the vertical scale (e.g. photoelectrons, volts, or normalized dimensionless units) themselves are not critical. However, to provide a means for comparing the fluorescence intensity contrast between pixels and measurements, the fluorescence intensities in Figure 3 and Figure 4 in the main text are scaled to the minimum and maximum fluorescence intensities of the selected example pixel in Figure 4 (white cross). The fluorescence intensities at τ=0\tau= 0 are not modified, and the minimum fluorescence intensity in the lower sub-figure of Figure 4b is used as the normalization value.

Magnetic Noise Floor Assessment

Here we estimate the pixel-to-pixel noise floor [S2, S3] for AC magnetic imaging using the QFM method. We can solve Eq. (3) in the main text to convert a measured ΩQFM\Omega_{\mathrm{QFM}} to an AC magnetic field amplitude Ωsz\Omega_{\mathrm{sz}},

Ωsz=ωsΩbΩQFM.(6)\Omega_{\mathrm{sz}} = \frac{\omega_s}{\Omega_b}\Omega_{\mathrm{QFM}}. \tag*{(6)}

If Ωb\Omega_b is homogeneous and is known to a small uncertainty before the QFM measurement, and ωs\omega_s is fixed, then the pixel-to-pixel noise floor in Ωsz\Omega_{\mathrm{sz}} is

δΩsz=ωsΩbδΩQFM.(7)\delta\Omega_{\mathrm{sz}} = \frac{\omega_s}{\Omega_b}\delta\Omega_{\mathrm{QFM}}. \tag*{(7)}

Ideally δΩQFM\delta\Omega_{\mathrm{QFM}} is determined by the photon shot noise of the NV fluorescence light imaged onto each camera pixel, which can be evaluated experimentally (keeping track of measurement repetition rate and dead time) and compared to the expected theoretical value.

QFM RABI FREQUENCY IN THE LARGE-Ωsz\Omega_{\mathrm{sz}} SMALL-ωs\omega_{\mathrm{s}} REGIMES

The system Hamiltonian is

H=ω02σz+Ωszσzcos⁡(ωst)+Ωbσxcos⁡(ωbt).(8)H = \frac{\omega_0}{2}\sigma_z + \Omega_{\mathrm{sz}}\sigma_z\cos(\omega_{\mathrm{s}}t) + \Omega_{\mathrm{b}}\sigma_x\cos(\omega_{\mathrm{b}}t). \tag*{(8)}

We can go into a first rotating frame at the ωb\omega_{\mathrm{b}} frequency:

H2=σz2[2Ωszcos⁡(ωst)−(ωb−ω0)]+Ωb2σx.(9)H_2 = \frac{\sigma_z}{2}\left[2\Omega_{\mathrm{sz}}\cos(\omega_{\mathrm{s}}t) - (\omega_{\mathrm{b}}-\omega_0)\right] + \frac{\Omega_{\mathrm{b}}}{2}\sigma_x. \tag*{(9)}

We neglect the counter-rotating term at 2ωb2\omega_{\mathrm{b}}. This is valid if ωb≫∣ω0−ωb∣\omega_{\mathrm{b}} \gg\lvert\omega_0-\omega_{\mathrm{b}}\rvert, which is the case here (but the approximation still fails for large Ωb\Omega_{\mathrm{b}}). Note that (ωb−ω0)=ωs+δz(\omega_{\mathrm{b}}-\omega_0)=\omega_{\mathrm{s}}+\delta_z, with δz≈0\delta_z\approx0.

We can define yet another rotating frame set by the σz\sigma_z term in H2H_2, H2(z)H_2^{(z)}. The rotating frame transformation is given by the unitary

U3=exp⁡(−i∫0tH2(z)(t′) dt′)≡exp⁡(−iϕσz/2)=exp⁡(−iσz2[2Ωszωssin⁡(ωst)−(ωs+δz)t]).(10)\begin{aligned} U_3 = \exp\left(-i\int_0^t H_2^{(z)}(t')\,dt'\right) \equiv\exp(-i\phi\sigma_z/2) \\ &= \exp\left(-i\frac{\sigma_z}{2}\left[\frac{2\Omega_{\mathrm{sz}}}{\omega_{\mathrm{s}}}\sin(\omega_{\mathrm{s}}t)-(\omega_{\mathrm{s}}+\delta_z)t\right]\right). \tag*{(10)} \end{aligned}

Since eiϕσz/2σ+e−iϕσz/2=e−iϕσ+e^{i\phi\sigma_z/2}\sigma_+e^{-i\phi\sigma_z/2}=e^{-i\phi}\sigma_+, we want to calculate

exp⁡(−i[Ωszωssin⁡(ωst)−(ωs+δz)t])(11)\exp\left(-i\left[\frac{\Omega_{\mathrm{sz}}}{\omega_{\mathrm{s}}}\sin(\omega_{\mathrm{s}}t)-(\omega_{\mathrm{s}}+\delta_z)t\right]\right) \tag*{(11)}
=∑n=0∞Jn(2Ωsz/ωs)e−i[nωs−(ωs+δz)]t,(12)= \sum_{n=0}^{\infty}J_n(2\Omega_{\mathrm{sz}}/\omega_{\mathrm{s}})e^{-i[n\omega_{\mathrm{s}}-(\omega_{\mathrm{s}}+\delta_z)]t}, \tag*{(12)}

where JnJ_n are Bessel functions of the first kind. The Hamiltonian in the rotating frame, H3=U3†H2U3−H2(z)H_3=U_3^\dagger H_2U_3-H_2^{(z)}, is then

H3=Ωb2∑nJn(2Ωsz/ωs)(σxcos⁡[nωs−(ωs+δz)]t+σysin⁡[nωs−(ωs+δz)]t).(13)H_3=\frac{\Omega_{\mathrm{b}}}{2}\sum_n J_n(2\Omega_{\mathrm{sz}}/\omega_{\mathrm{s}})\left(\sigma_x\cos[n\omega_{\mathrm{s}}-(\omega_{\mathrm{s}}+\delta_z)]t +\sigma_y\sin[n\omega_{\mathrm{s}}-(\omega_{\mathrm{s}}+\delta_z)]t\right). \tag*{(13)}

The dominant time-independent term for δz≈0\delta_z\approx0 is n=1n=1. Note that instead for ωs\omega_{\mathrm{s}} relatively large we do not have just a single dominating term, and thus this approach does not give a good solution. Here instead, this approximation gives an amplitude

ΩQFM=ΩbJ1(2Ωszωs),(14)\Omega_{\mathrm{QFM}}=\Omega_{\mathrm{b}}J_1\left(\frac{2\Omega_{\mathrm{sz}}}{\omega_{\mathrm{s}}}\right), \tag*{(14)}

which reproduces the data shown in Fig. 5 of the main text.

Corrections

We can consider the lowest-order correction for H3H_3 at δz=0\delta_z=0:

H3≈Ωb2[J1(2Ωsz/ωs)σx+J0(2Ωsz/ωs)(σxcos⁡(ωst)−σysin⁡(ωst))].(15)H_3\approx\frac{\Omega_{\mathrm{b}}}{2}\left[J_1(2\Omega_{\mathrm{sz}}/\omega_{\mathrm{s}})\sigma_x +J_0(2\Omega_{\mathrm{sz}}/\omega_{\mathrm{s}})\left(\sigma_x\cos(\omega_{\mathrm{s}}t)-\sigma_y\sin(\omega_{\mathrm{s}}t)\right)\right]. \tag*{(15)}

Note that for 2Ωsz<ωs2\Omega_{sz}<\omega_s we have J1<J0J_1<J_0, thus the oscillating term might dominate. While this looks similar to Eq. (9), here ωs<ΩbJ1(2Ωsz/ωs)\omega_s<\Omega_b J_1(2\Omega_{sz}/\omega_s), so we cannot take the same route. We can instead use the Floquet expansion used in [S4]. We then obtain the following detuning as the leading correction,

H3≈Ωb2J1(2Ωsz/ωs)σx+Ωb2J02(2Ωsz/ωs)4ωsσz.(16)H_3 \approx\frac{\Omega_b}{2}J_1(2\Omega_{sz}/\omega_s)\sigma_x+\frac{\Omega_b^2J_0^2(2\Omega_{sz}/\omega_s)}{4\omega_s}\sigma_z. \tag*{(16)}

References

  1. [1]E. V. Levine, M. J. Turner, P. Kehayias, C. A. Hart, N. Langellier, R. Trubko, D. R. Glenn, R. R. Fu, and R. L. Walsworth, Nanophotonics 8, 1945 (2019).
  2. [2]S. C. Scholten, A. J. Healey, I. O. Robertson, G. J. Abrahams, D. A. Broadway, and J.-P. Tetienne, Journal of Applied Physics 130, 050902 (2021).
  3. [3]L. M. Pham, D. L. Sage, P. L. Stanwix, T. K. Yeung, D. Glenn, A. Trifonov, P. Cappellaro, P. R. Hemmer, M. D. Lukin, H. Park, A. Yacoby, and R. L. Walsworth, New Journal of Physics 13, 045021 (2011).
  4. [4]S. J. DeVience, L. M. Pham, I. Lovchinsky, A. O. Sushkov, N. Bar-Gill, C. Belthangady, F. Casola, M. Corbett, H. Zhang, M. Lukin, H. Park, A. Yacoby, and R. L. Walsworth, Nature Nanotechnology 10, 129 (2016).
  5. [5]A. Horsley, P. Appel, J. Wolters, J. Achard, A. Tallaire, P. Maletinsky, and P. Treutlein, Phys. Rev. Applied 10, 044039 (2018).
  6. [6]F. Ziem, M. Garsi, H. Fedder, and J. Wachtrup, Scientific Reports 9, 12166 (2019).
  7. [7]K. D. Briegel, N. R. von Graffenstein, J. C. Draeger, P. Blümler, R. D. Allert, and D. B. Bucher, arXiv:2402.18239 (2024).DOI
  8. [8]C. L. Degen, F. Reinhard, and P. Cappellaro, Rev. Mod. Phys. 89, 035002 (2017).

Paper details

vlnc:W_sgsqzyce
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents