Sensitive multichannel zero-to ultralow-field NMR with atomic magnetometer arrays

Abstract

Despite its versatility and high chemical specificity, conventional nuclear magnetic resonance (NMR) spectroscopy is limited in measurement throughput due to the need for high-homogeneity magnetic fields, necessitating sequential sample analysis, and expensive devices. Here, we propose a multichannel NMR device that addresses these limitations by leveraging the zero-to ultralow-field (ZULF) regime, where simultaneous detection of multiple samples is carried out via an array of compact optically pumped magnetometers (OPMs). A magnetic field is used only for prepolarization, permitting the use of large-bore, high-field, inhomogeneous magnets that can accommodate multiple samples concurrently. Through systematic improvements, we demonstrate sensitive, high-resolution ZULF NMR spectroscopy with sensitivity comparable to benchtop ¹³C NMR systems. The spectroscopy remains robust without the need for field shimming for periods on the order of weeks. We show the detection of ZULF NMR signals from organic molecules without isotopic enrichment, and demonstrate the parallelized detection of three distinct samples simultaneously as a proof-of-concept, with the ability to scale further to over 100 channels at a cost comparable to traditional liquid state NMR systems. This work sets the stage for using multichannel “NMR camera” devices for inline reaction monitoring, robotic chemistry, quality control, and high-throughput assays.

Significance Statement.

Nuclear magnetic resonance (NMR) suffers from low sample throughput due to homogeneity requirements which afford only a small, sample-sized detection volume for chemical measurement. Zero-to ultralow-field (ZULF) NMR subverts this limitation by detecting samples in large, inherently homogeneous magnetic shields following preparation at elevated, yet inhomogeneous magnetic fields. This remedies low throughput while reducing overall instrumental cost. We leverage this regime to perform high-throughput, arrayed ZULF NMR detection for the first time. Through systematic improvements, we obtain ZULF spectra of organic molecules without isotopic enrichment with sensitivity comparable to benchtop ¹³C NMR, yet with inhomogeneous prepolarization fields. This demonstrates the viability of ZULF NMR as a scalable platform for high-throughput chemical analysis in a cost-effective manner.

Introduction

Nuclear magnetic resonance (NMR) spectroscopy is a key tool in chemical analysis due to its ability to examine chemical kinetics and molecular bonding noninvasively, and with high specificity (1, 2). However, conventional NMR spectroscopy is expensive and limited in chemical throughput, primarily due to the need for highly homogeneous superconducting magnets. The stringent requirements for magnetic field homogeneity (typically at the <10 ppb level (3)) engender small “sweet spots” ($<$5 cm³) where samples must be placed sequentially for analysis. Additionally, the need for spatiotemporal homogeneity makes the magnets bulky, and necessitates involved strategies for shimming and deuterium field lock (4). These limitations are particularly problematic in emerging fields like robotic chemical synthesis and combinatorial screening (5–7), where there is a critical need for in-line monitoring tools to support continuous AI-driven reaction optimization, but for which NMR is currently less well-suited.

In this work, we attempt to address this throughput limit by proposing an alternative strategy for parallelized NMR spectroscopy that exploits the zero-to ultralow-field (ZULF) regime. Here, the NMR spectra are dominated by J-couplings and can be complex even for small molecules (8). NMR detection is carried out in a magnetically shielded environment, while a magnet is solely used for prepolarization. As a result, the magnetic field homogeneity requirements are considerably relaxed. Detection is carried out by compact, commercially available optically pumped magnetometer (OPM) devices (9), and the lifting of homogeneity requirements means that one can arrange for an array of detectors to discern NMR spectra from multiple samples simultaneously. Additionally, the magnets used for sample prepolarization can be inhomogeneous, and accommodate a large number of samples concurrently. We demonstrate the first proof of concept of this vision outlined in Fig. 1.

Fig. 1.

High-sensitivity, multichannel, ZULF NMR system. A) Multichannel NMR concept. Envisioned instrument consists of a high-field, inhomogeneous, prepolarization magnet here at $B_0=9.4$ T, along $\hat{\mathbf{x}}$. The magnet is shielded, horizontally oriented, and hosts a large bore (11 cm diameter). ZF NMR detection is carried out with an array of OPMs in a mu-metal shielded region. Samples are mechanically shuttled between HF prepolarization and ZF detection centers, separated by a 1.55 m travel distance, in $<$1 s. B) $>$100 Sample Array Concept of sample placement in (i) HF center for prepolarization, where magnet inhomogeneity plays insignificant role; and (ii) ZF center, where samples are probed noninvasively by an array of OPMs. We estimate $>$100 samples can be accommodated at the HF and ZF regions (see SI Section II). C) OPM operation for ZF NMR detection is schematically represented (see Ref. (12)). OPMs are compact (dimensions marked), allowing for arrayed operation. D) Photograph of assembled instrument with three channels. See SI Section I for discussion on device construction and additional photographs (Fig. S1 in SI). Coordinate axes (marked) highlight connection between the different panels.

A significant barrier for ZULF NMR has been its relatively low sensitivity—often at least two orders of magnitude worse than benchtop NMR spectroscopy—diminishing its appeal despite its advantages. Here, through a series of technical improvements, we demonstrate the ability to address this challenge, for the first time, showing high signal-to-noise ratio (SNR) spectroscopy matching the ¹³C NMR sensitivity of a commercial 1.9 T benchtop system. Our approach exploits prepolarization with a large-bore (inhomogeneous) superconducting magnet, while implementing several methods to combat experimental noise. Notably, our approach supports high-resolution spectroscopy with the complete absence of shimming or OPM recalibration for periods on the order of weeks, all while employing magnets too inhomogeneous for conventional NMR, yet with a large enough bore to uniquely enable multichannel detection.

Leveraging these advances, we report the first ZULF NMR spectra of organic molecules at natural ¹³C isotopic abundance without hyperpolarization. We demonstrate parallelized detection of three distinct samples simultaneously as a proof-of-concept; however, we estimate the ability to scale to $>$100 channels with comparable cost to a traditional liquid state NMR system (see SI Section II). We note parallel developments in the field of magnetoencephalography (MEG) with multichannel OPM detection (10, 11).

Experimental design

Figure 1A shows a conceptual depiction of the envisioned multichannel NMR device; our experimental demonstration is on a proof of concept with three parallel channels (Fig. 1D). It employs a cost-effective, inhomogeneous, $B_0=9.4$ T horizontal-bore magnet for prepolarization, positioned adjacent to a mu-metal magnetic shield for NMR measurements in the ZULF regime, or alternately, the zero-field (ZF) regime. The superconducting magnet enables higher prepolarization fields compared to previous experiments in the literature, augmenting ZF NMR signal strength. Inner shielding in the magnet allows a small separation between the HF and ZF centers ($\sim$1.55 m center-to-center) despite heightened fringe fields in the horizontal configuration. In operation, the samples are “shuttled” between both centers by a high-speed motor driven belt actuator in <1 s; in practice, a solenoid “guiding field” (GF₁) provides a switchable weak field over the path adjoining the centers (see Fig. 2A and Fig. S1 in SI). Signal loss during shuttling due to $T_1$ relaxation can be minimized when $T_1$ significantly exceeds the shuttling time.

Fig. 2.

Robust, high SNR, ZF NMR via inhomogenous field polarization. A) Experiment schematic. Sample is prepolarized in the inhomogenous $B_0=9.4$ T field for $t_{\mathrm{pol}}$ (19 s for formic acid) before moving through the first guiding field (GF₁) solenoid (80 $\mu$T) along the shuttling path for the duration of travel in ($\approx 0.9$ s). Before GF₁ is adiabatically turned off over 30 ms, a 1 $\mu$T second guiding field (GF₂) is preemptively turned on uniformly throughout the shield. When the sample arrives, GF₂ is suddenly turned off immediately before acquisition, which proceeds for $t_{\mathrm{acq}}\sim T_2^{*}$. B) High SNR spectroscopy. Panel shows ZF-NMR spectrum of enriched formic acid over 20 scans with $t_{\mathrm{acq}}=11$ s in (i) linear and (ii) logarithmic scales. Characteristic peak at $J_{\mathrm{CH}}=222.1$ Hz is visible. Single shot SNR $=\mathrm{1,250}$, calculated from the spectral wing (shaded). C) High-resolution ZF NMR of enriched benzaldehyde with $t_{\mathrm{acq}}=30$ s shown over 200 scans. Phase correction and baseline subtraction is applied. Inset (i): Zoom in shows sub-hertz, $T_2^{*}$ limited linewidths. D) Long-time robust ZF NMR. Spectral FWHM of enriched formic acid (inset) is used as a reporter of temporal stability over 256 h ($\approx 10.67$ days). No shimming, calibration, or field compensation is applied during the entire period. Linewidths ($T_2^{*}$ limited) remain stable at 0.144 Hz to within error, and are reported as an average of 128 scans (dashed line). Error bar: linewidth standard deviation over 128 scans. E) Ultra-low-field NMR here conducted on water at (i) 12 nT and (ii) 244 nT. Panel shows single-shot free induction decays with (iii) high one-shot SNR value $=\mathrm{4,200}$ at 244 nT. All samples were prepared and measured under ambient atmospheric conditions (no deoxygenation) in standard 5 mm NMR tubes.

Since detection occurs at the ZF center, magnet inhomogeneity manifests as negligible variations in overall ZF NMR signal intensity without affecting spectral resolution, effectively subverting the otherwise usual line-broadening effects stemming from inhomogeneity. Hence, the method can tolerate magnets with inhomogeneities as large as a few percent, which only moderately affects SNR. This permits using cost-effective high-field magnets that host large-bores, allowing for greater sample accommodation than conventional NMR magnets. In our device, for instance, the magnet could feasibly accommodate a rack of $>100$ sample NMR tubes simultaneously (Figs. S2 and S3 in SI); this concept is schematically depicted in Fig. 1B(i).

Since homogeneity considerations are comparably trivial to handle at zero-field, the ZF chamber can be constructed to be spacious enough to accommodate a similarly large number of samples at low cost (Fig. 1A). Tunable low fields can be created within it; allowing us to access both the ZF and ultra-low-field (ULF) regimes (13). Key to multichannel operation in our work is conducting NMR measurements through an array of commercially available OPMs (QuSpin (14)), schematically shown in Fig. 1B(ii) as a concept, placed in proximity to the initially prepolarized NMR tube rack. Such arrayed detection exploits the compact form factor of the OPMs (Fig. 1C(i)), which can detect NMR signals at a relatively high standoff ($>$5 mm).

For clarity, Fig. 1C shows a schematic operation of one of the OPMs from the array (15). An integrated laser optically polarizes Rb atoms in a vapor cell within the unit as seen in Fig. 1C(i). Transmission of the light through the vapor cell is measured in the presence of a weak, localized oscillating transverse ($\hat{\mathbf{y}}$, $\hat{\mathbf{z}}$) magnetic field (<100 nT, $\sim$1 kHz). The quadrature demodulation signal is proportional to the background field component along the modulation axis, allowing sensitive measurement of fields from the ensemble of analyte nuclei in the sample container. Alternatively, one can envision configurations using a single cell excited by an array of laser beams instead of discrete OPM units (16).

Figure 1D shows a photograph of the assembled device implementing a proof-of-concept of the vision in Fig. 1A, but with three channels. Pictured is the horizontal magnet and ZF center. More details on device construction and photographs (Fig. S1) are provided in SI Section I.

Figure 2A displays a schematic of the experimental sequence. Coordinate axes show orientations of the applied fields, OPM and sample for clarity (see also Fig. 1). Samples are initially polarized using the $B_0=9.4$ T magnet for $t_{\mathrm{pol}}$=3–5 times $T_1$ ($\sim$19 s for formic acid), followed by transfer through the magnet fringe field along the GF₁ solenoid energized at 80 $\mu$T, collinear with $B_0$ (here $\hat{\mathbf{x}}$). Adiabatic turn-off of this field then transitions the sample to a secondary guiding field (GF₂) within the ZF center at 1 $\mu$T, that is preemptively activated to reorient nuclear spins along the OPM sensitive axis (here $-\hat{\mathbf{y}}$). GF₂ is subsequently rapidly turned off, prompting a nonadiabatic transition to ZF and initiating spin dynamics that generate the ZF NMR signal, captured by the OPM. Inset in Fig. 2A details the orientations ${\mathrm{GF}}_{1,2}$ relative to the OPM and sample. The entire field-switching protocol is driven using a low-cost, compact, controller (NMRduino (17)), equipped with high-current driver chips (DRV8838 and TB6612FNG for analog and digital channels, respectively) and a LTC1859 chip for signal acquisition (see Figs. S4–S6 in SI).

While the sequence in Fig. 2A itself follows prior work (18, 19), we include several innovations in the device to enable high-SNR multichannel operation (see SI Sections I and III). The shuttling stage and motor are placed at the opposite end of the $B_0=9.4$ T magnet, distancing them from the ZF center to mitigate electronic noise and vibration. ${\mathrm{GF}}_{1,2}$ coils, OPMs, and all control electronics operate on battery power to further reduce interference. Vibration is minimized by mounting the ZF center on Sorbothane vibration-dampening feet and using a flexible carbon fiber rod to support the samples during shuttling (see SI Section I). Finally, we adopt a “pulse-free” technique that does not require a DC pulse to initiate the free induction decay (FID). This removes the need for pulsing coils within the ZF center, simplifying multichannel operations (see Fig. 5) by applying GFs uniformly throughout the shuttling path and shield, effectively manipulating all chemical samples in an identical manner. This additionally decreases the risk of magnetizing the shield through strong pulses, ultimately leading to highly robust operation on the order of weeks (Fig. 2D). We found the removal of these pulsing coils also reduced background noise close to the sensor, contributing to improved sensitivity.

Fig. 5.

Multichannel ZF NMR via inhomogenous field prepolarization. A) Simultaneous ZF NMR spectra taken from three separate OPM channels, showing measurement of three distinct ¹³C-enriched chemical samples: (i) methanol (dark blue, left), (ii) formic acid (purple, middle), and (iii) benzaldehyde (orange, right). B) Inset (i): Photograph of multichannel setup showing sideview of OPM arrangement. Main panel: Schematic of top-view of OPM arrangement, arranged in a linear fashion, separated by 5.4 mm. Three samples are color-coded. Coordinate axes, corresponding to Fig. 1 shown for clarity. C) Simultaneously acquired FIDs of the three samples. Digital bandpass filters employed (top to bottom): 120–300 Hz, 221–225 Hz, 150–200 Hz. D) Measurement of OPM crosstalk. (i and ii) Pictorial representation of OPM separation d defining the edge-to-edge distance between sensors. Sample of ¹³C-enriched formic acid is placed in front of one OPM; crosstalk quantifies extent of signal sensed by the adjacent OPM. E) OPM crosstalk calculated as ratio of baseline corrected peak heights expressed as a percentage plotted against separation for two adjacent OPMs. Inset (i): example spectra measured by the OPM directly front of sample (left) and the one adjacent to it (right). Here, $d=20.4$ mm.

Figure 2B illustrates the outcomes of these instrumentation advances, initially focusing on a single channel. Using ¹³C enriched formic acid, a common benchmark used for J-spectroscopy at ZF, we achieve a single-shot SNR of $\sim$1,250, presented on both linear (Fig. 2B(i)) and logarithmic scales (Fig. 2B(ii)). The visible peak is at the ¹³${\mathrm{C}-}^1\mathrm{H}$ J-coupling value of 222.1 Hz, accompanied by the expected peak at 0 Hz (at the mean larmor frequency of ¹H and ¹³C at ZF). To our knowledge, this SNR is the highest reported in the literature and is achieved here with an off-the-shelf sample without sample deoxygenation (8, 9, 20). Figure 2B(i) also shows the spectrum is also notably free from spectral contamination, including line and motor noise.

Figure 2C displays the spectrum from a more complex sample, carbonyl labeled ¹³C benzaldehyde. The main panel displays a broad frequency range, once again emphasizing SNR and spectral purity, which matches previous reports but with minimal scans and sample preparation. Inset Fig. 2C(i) zooms into the relevant window near 175 Hz, showing distinct J-spectral features (8), with a $T_2^{*}$ limited linewidth <0.2 Hz.

As illustrated in Fig. 2C, ZF-NMR spectra, dominated by scalar J-couplings, exhibit a high degree of spectral complexity. This spectral richness offers potential advantages for molecular discrimination, as even small structural differences, such as variations in functional groups, yield distinct zero-field NMR spectra (8, 20). Such spectra could serve as unique molecular fingerprints or chemical “barcodes” for analyzing complex mixtures. Moreover, the compatibility of ZF-NMR with machine learning (ML) techniques, such as those trained on DFT simulations (21), presents a promising pathway for solving the “inverse” problem—i.e. identifying unknown molecules from their ZF spectra. Additionally, weak residual fields can be reintroduced via guiding solenoids to operate in the ultra-low-field (ULF) or low-field (LF) regime without affecting the sensor environment. Here, J-couplings become comparable to or weaker than the Larmor field, allowing for the resolution of single-spin transitions akin to high-field NMR (13, 22). This hybrid regime offers a range of J- and B-dominated fingerprints, potentially enhancing chemical analysis capabilities further.

A useful feature of our instrument is its robust temporal stability, highlighted in Fig. 2D. Here continuous ZF NMR measurements, conducted in full operation as in Fig. 2A, on ¹³C enriched formic acid reveal a stable, $T_2^{*}$ limited, spectral linewidth of $\sim$145 mHz, calculated as the FWHM (inset shows representative spectrum). Importantly, this linewidth remains stable for $>$10 days without any shimming or degaussing of the mu-metal shield (23), nor OPM recalibration during the entire period. No spectral degradation was observed over this period to within error, indicating that the peak may remain stable for much longer. This stability contrasts with typical benchtop NMR systems where field instability is around 0.0014 ppm/h (0.083 Hz at 60 MHz) assuming <1°C change in temperature, often necessitating frequent shimming or active field compensation. From our experience, the stability observed in Fig. 2D is also better than ZF NMR apparatuses constructed with permanent (Halbach) magnets and pulsing coils. We attribute this to the low spatial gradient of the fringe field of the superconducting $B_0=9.4$ T magnet, which is effectively shielded by the multilayered mu-metal in the ZF center, as well as reliable shield integrity in the absence of strong pulses. However, similar shields with a ferrite core layer can possibly mitigate these concerns.

Stable shield conditions also enable sensitive ultralow field (ULF) NMR measurements. Figure 2E(i and ii) demonstrates this with water at bias fields of 12 nT and 244 nT respectively, applied along the GF₁ coil. Here, the FID is sampled every ∼165 $\mu$s, and the spins exhibit a slow Larmor precession of 0.51 and 10.39 Hz that can be clearly discerned. The ULF measurement here is of similarly high SNR (Fig. 2E(iii)); we estimate a one-shot SNR $\approx$ 4,200 for the 244 nT bias field. Figure 2E suggests applications towards ULF NMR with reintroducing controlled Zeeman fields to gain in chemical resolution (22, 24) or in exploiting enhanced relaxation–dispersion at ultralow fields (13).

ZULF NMR at natural abundance

The sensitivity of our apparatus enables ZF NMR signal measurements at natural isotopic abundance (NA) with no deoxygenation, previously unfeasible due to low sensitivity. Historically, ZF NMR required ¹³C enriched molecules (25, 26) or those containing specific spin-1/2 heteronuclei like ¹⁹F or ³¹P (27, 28). Here we present, to our knowledge, the first direct ZULF detection of organic molecules at natural abundance of ¹³C nuclei. This is demonstrated in Fig. 3A with formic acid, comparing ZF NMR signals from fully enriched and NA samples on a logarithmic scale, showing the expected 100-fold SNR reduction. Dark solid lines here are Lorentzian fits. NA methanol and benzene are similarly analyzed in Fig. 3B and C(i), respectively. The benzene spectrum corresponds well with simulations (Fig. 3C(ii)) despite being taken from a previously opened stock solution where degradation has likely occurred, faithfully reproducing chemistry laboratory conditions.

Fig. 3.

ZF NMR at natural ¹³C abundance in organic molecules. A) SNR comparison. Spectra corresponding to ¹³C enriched and natural abundance (NA) (1.1%¹³C) (turquoise, top and blue, bottom respectively) of formic acid plotted on a logarithmic scale (10 scans each). NA signal is 100 times weaker, but discernible. B) ZF spectrum of NA methanol (497 scans) in a neat solution shown in two windows centered at 140 Hz and 280 Hz. Lorentzian fits for formic acid and methanol are in purple. C) ZF spectrum of NA benzene in a neat solution, taken over 1,436 scans. (i) Experimental in orange (top) with (ii) simulated data below in dark blue. D) (i) NA formic acid ZF spectra taken with varying number of scans N and (ii) SNR plotted against $\sqrt{N}$. SNR $=175$ after 169 scans.

Given the stability of our detection apparatus, SNR can be enhanced through averaging. Figure 3D(i) displays individual spectral traces with increasing averages N, demonstrating an expected SNR scaling $\propto \sqrt{N}$. For $N=169$, we achieve an SNR of 175 for NA formic acid, with an average single-shot SNR of 13.

We emphasize that significant further SNR improvements are achievable. Our experiments utilize standard 5 mm NMR tubes that are not optimal and under-utilize the OPM detection volume. Additionally, for the current shielding conditions, the commercial OPMs we employ have a sensitivity $\approx$40 fT/$\sqrt{\mathrm{Hz}}$ (14). Emerging prototype OPMs, however, have reported sensitivities closer to 1 fT/$\sqrt{\mathrm{Hz}}$ (29). By using these more sensitive OPMs and optimizing the sample tube to aid detection, the signal could be improved by over an order of magnitude. Furthermore, the incorporation of a ferrite core in the innermost layer of the magnetic shield is reported to yield up to a tenfold reduction in magnetic noise (30) relative to the magnetic shield used in this study (31). In parallel, we anticipate even further signal gains from recent advances in high temperature superconducting (HTS) magnet technology (32–34) which can enable higher, albeit inhomogeneous, polarizing fields.

Even at current sensitivity, however, our apparatus compares well with benchtop high-field NMR technology. Traditionally, ZULF NMR sensitivity has been significantly lower—often by more than two orders of magnitude—compared to benchtop NMR; however, we demonstrate here a promising step towards bridging this gap. In particular, we compare our system to ¹³C NMR conducted on a Magritek Spinsolve 80 Carbon (detection field $\sim$1.9 T), using identical samples of neat NA formic acid solutions in conventional 5 mm NMR tubes.

Figure 4A(i and ii) first compares the resulting NMR spectra at ZF and HF (1.9 T), displaying J-resolved and chemical-shift-resolved spectra respectively, and demonstrating that ZF NMR spectra offers comparable chemical information for small organic molecules. The bar chart in Fig. 4B analyzes the signals over eight averages. From the upper two bars in Fig. 4B, corresponding to the data in Fig. 4A, it is evident that the that SNR of ZF NMR in our apparatus is comparable to that of a standard, ¹³C benchtop NMR measurement, although further improvement is necessary to match the sensitivity of a ¹H NMR experiment. Notably, our ZF apparatus does not require shimming or recalibration between experiments (see Fig. 2D), nor deuterium locking, unlike the benchtop HF instrumentation where repeated adjustments are necessary. Importantly, while the HF NMR device can accommodate only single samples, which have to be measured serially, our ZF NMR device can allow multichannel operation (see Fig. 5).

Fig. 4.

SNR comparison with benchtop HF NMR. A) ZF and HF NMR spectrum of NA formic acid, showing complementary features of J-couplings and chemical shifts. B) SNR comparison between our apparatus and various ¹³C NMR experiments done on a Spinsolve 80 Carbon ($\sim 1.9$ T detection field). Top to bottom: ZF-NMR via our apparatus, ¹³C NMR, ¹³C NMR with NOE enhancement, ¹³C NMR with proton decoupling, and ¹³C with NOE enhancement and proton decoupling. SNR is reported over eight scans, while error bars are calculated over 30 repetitions of such experiments.

The other bars in Fig. 4B depict enhancements to HF signals via employing the nuclear Overhauser effect (NOE), proton decoupling (¹H DCP), and their combinations. The ZF NMR signal remains within an order of magnitude of these enhanced cases.

Simultaneously acquired multichannel ZULF NMR

Leveraging prepolarization with the inhomogeneous magnet, we now move to multichannel operation, presented in Fig. 5 as a proof-of-concept of the vision outlined in Fig. 1. As a representative building block of larger arrayed detection, we constructed a $3\times 1$ OPM array at the ZF center capable of simultaneously acquiring ZF-NMR spectra from three distinct samples—here, enriched methanol, formic acid, and benzaldehyde. Figure 5A shows a zoomed view into the J-resolved spectra of these samples. Figure 5B details the setup within the ZF shield. For simplicity, the samples are arranged linearly, separated by $d=5.4$ mm, corresponding to OPMs placed side-by-side. Inset Fig. 5B(i) shows a side-view photograph of the OPM arrangement, while the main panel presents a schematic top view with axes marked, corresponding to same coordinate system in Figs. 1 and 2A, although the OPMs are placed beside the sample instead of below for ease of access and adjustments. Here, samples are contained in short 5 mm tubes (length $\sim$15 mm) and subjected to the sequence in Fig. 2A simultaneously. Figure 5C displays the FID signals from the three samples. We note that SNR was not fully optimized in these experiments, as the implementation of the data acquisition unit was not specifically configured to minimize low-frequency noise, although modifying the NMRduino using 3 on board ADC chips (LTC1859) poses no in-principle challenge.

An important metric for feasibility of arrayed detection is “crosstalk,” where spectra may bleed into adjacent channels due to correlations at short OPM separations d, caused by sensor interference or spurious detection of a sample’s magnetization by neighboring sensors. In Fig. 5D and E, we use a pair of OPMs to quantify crosstalk between adjacent channels. The setup is depicted in Fig. 5D(i–ii), with coordinate axes marked for reference, and crosstalk is measured as a function of d (Fig. 5E). We use a ¹³C-enriched formic acid sample placed in front of one OPM, measure the ZF-NMR signal from both OPMs simultaneously, and quantify crosstalk as the ratio of spectral magnitudes. Inset Fig. 5E(i) illustrates this for OPMs separated by $d=20.4$ mm.

The bar chart in Fig. 5D, derived from 10 consecutive shots, show that even at the closest side-by-side distance ($d=5.4$ mm, the closest distance enabled in our custom 3D printed OPM holder in Fig. 5B(i)), the crosstalk is $<10\mathrm{\%}$. It declines sharply thereafter with increasing d, and stabilizes at $\sim 2\mathrm{\%}$ at $d=15.4$ mm. The saturation value likely reflects inherent crosstalk within the data acquisition unit itself, since this steady-state crosstalk value was observed in the channels even when the empty (adjacent) channel was completely turned off.

While these results serve as a proof-of-concept, prospects for scaling to arrayed OPM detection at a larger scale appear promising. Even with the current shield and magnet arrangement, and separating the OPMs by $d=5.4$ mm, we estimate the capacity to accommodate $>$100 samples simultaneously (See SI Section II). The low GF values ($\mu$T) employed further support this, as homogeneous low field solenoids can be easily fabricated to accommodate large sample arrays. Extensive academic (11, 35, 36) and commercial (10) work on OPM-magnetoencephalography (OPM-MEG) has been reported using multichannel operation in close proximity on the scalp to detect minute magnetic signals emanating from the brain. Recent reports demonstrate 180 channel operation using the same sensors employed in this work (11). This makes MEG data a valuable benchmark and lends credence to our vision of multichannel NMR operation at scale, exploiting OPM arrays for high-throughput analysis in conjunction with inhomogeneous magnets.

Lastly, we compare our methodology with alternative approaches for high-throughput NMR analysis. One strategy involves parallel NMR detection employing numerous microcoil receivers and small sample volumes at HF (37, 38). However, the magnet’s homogeneous “sweet spot” remains constant, setting an upper bound on the number of samples that can be accommodated simultaneously. Smaller coils might increase capacity, but they complicate the design and reduce efficiency. Alternately, approaches for simultaneous multinuclear detection enhance the richness of spectral information (39, 40), but they do not expand the total sample throughput. An emerging strategy involves magnetic resonance spectroscopy imaging (MRSI) (41). However, this method faces challenges in the expense of imaging-grade magnets, a low per-sample filling factor, and slower measurements due to the indirect dimension of spectroscopy.

In contrast, arrayed OPM-based ZULF detection is designed for direct spectroscopy and can be optimized for each sample’s filling factor. It is also inherently extensible, since the mu-metal shield can be made large enough to accommodate an arbitrarily large number of samples while also offering relaxed requirements for shimming and calibration.

Outlook

In certain contexts, multichannel ZULF NMR may already offer advantages over HF NMR, particularly for high-throughput applications. SI Section II presents a cost analysis for an arrayed ZULF NMR system. Using conservative per-channel estimates based on commercial list prices and factoring in economies of scale, our findings reveal a benign cost scaling, suggesting the feasibility of constructing a ZULF NMR device with $>$100 channels for a cost comparable to that of a conventional liquid-state HF NMR system operating at 400–500 MHz (see Fig. S2 in SI). We note that such arrayed OPM devices have already been assembled in the in the context of MEG applications (10, 11, 35, 36, 42).

We emphasize, however, that HF NMR systems have broader capabilities, including the ability to analyze a wide range of samples and perform multidimensional NMR, and currently exhibit higher sensitivity. However, we argue that for high-throughput 1D spectroscopy applications like in robotic chemistry (43, 44), where there is an unmet need for an in-line, noninvasive, analysis tool for kinetic measurements, ZULF NMR could offer distinct advantages. The horizontal bore configuration of our system (Fig. 1D) is particularly beneficial, allowing ZF NMR analysis to be performed diametrically opposite the chemical synthesis region where samples can be loaded and shuttled.

Furthermore, we anticipate straightforward improvements to our apparatus that could enhance sensitivity by more than an order of magnitude without compromising multichannel capacity. This includes optimizing the filling-factor and integrating next-generation OPMs (29). Additionally, HTS magnets, free from cryoshim requirements, could be optimized to house numerous samples in a compact form-factor. The capacity of multichannel ZF NMR systems to operate for extended periods without shimming or temporal field-lock (Fig. 2D) presents strong advantages for certain applications. It also portends new strategies to exploit cross-correlations between neighboring channels to suppress measurement noise (45–47).

Ultimately, this work suggests the potential for an “NMR camera” capable of observing multiple samples simultaneously with a benign cost scaling. This development could lead to assays for quality control (48, 49), and disease detection (50–52), as well as applications that exploit the ZF regime’s ability to enhance relaxation dispersion (13), eliminate susceptibility broadening artifacts, and penetrate metal containers (53, 54), potentially enabling in situ, in operando battery diagnostics (55) at scale.

Supplementary Material

Supplementary material is available at PNAS Nexus online.

Funding

We gratefully acknowledge funding from NSF Partnerships for Innovation (PFI) (2141083), NSF EArly-concept Grants for Exploratory Research (EAGER) (2231634), NSF Major Resaerch Instrumentation Program (MRI) (2320520), Hamamatsu Photonics (20201452), CIFAR Azrieli Foundation (GS23-013), and the Spanish Ministry of Science project RYC2022-035450-I, funded by MCIN/AEI /10.13039/501100011033.

Author Contributions

B.A.: conceptualization; investigation; writing-original draft; project administration; writing-review conceptualization; formal analysis; visualization; methodology; writing-review and editing. M.L.: conceptualization; investigation; writing-original draft; formal analysis; visualization; methodology; writing-review and editing. Z.W.: data curation; formal analysis. N.K.: conceptualization; methology. M.T.: methodology; writing-review and editing. E.D.: conceptualization; methodology; project administration. A.A.: conceptualization; resources; draft; project administration; writing-review and editing.

Preprints

This manuscript was posted on a preprint: arXiv:2407.00929.

Data Availability

The data that support the findings of this study are available at 10.5281/zenodo.15588118.