An integrated CMOS micro search‑coil magnetometer with on‑chip calibration and a low‑noise amplifier Authors: Hadi Tavakkoli, Moaaz Ahmed, Ikramullah Shah, Sile Fang, Wenhao Chen, Amine Bermak, Yi‑Kuen Lee Published: 2026 ABSTRACT AbstractThis paper presents a fully CMOS‑integrated micro search‑coil magnetometer (µSCM) that achieves a NEMI of 2.1 nT/√Hz at 1 kHz while consuming only 364 µW, approximately a tenfold improvement over previous CMOS search‑coil magnetometers (SCMs). The system integrates a low‑noise instrumentation amplifier (INA) and on‑chip calibration, optimized for operation from 1 Hz to 1 kHz. Implemented in a standard TSMC 180 nm CMOS process, it uses an Integrated‑within‑Aperture layout that places the INA entirely inside the planar spiral inductor’s inner diameter, maximizing area efficiency. A dedicated theoretical framework is used to determine the critical line width (wc) and inner diameter (Dic) to minimize the noise equivalent magnetic induction (NEMI). The integrated INA employs a nested‑chopped current‑feedback architecture to mitigate flicker‑noise, achieving a significant reduction in input‑referred noise from 73.45 nV/√Hz to 36.92 nV/√Hz at 100 Hz. Additional features, including a Hall‑effect sensor for DC magnetic field measurements and a temperature sensor, further enhance the system’s versatility. This compact, energy‑efficient solution is ideal for precision applications such as biomedical diagnostics, environmental monitoring, and energy‑efficient systems. Subject terms: Electrical and electronic engineering; Environmental, health and safety issues Introduction Magnetometers are widely used for measuring magnetic fields in various scientific, industrial, and biomedical applications. There are several types of magnetometers, including Hall‑effect sensors1, anisotropic magnetoresistance (AMR) sensors2, giant magnetoresistive (GMR) sensors3, magnetostrictive sensors4, tunneling magnetoresistance (TMR) sensors5, optical magnetometers6, MEMS Lorentz‑force magnetometers7,8, fluxgate magnetometers9, superconducting quantum interference devices (SQUIDs)10, and atomic magnetometers11,12. Each type has its own measurement range13, specific applications, advantages, and disadvantages. Hall‑effect sensors are compact and cost‑effective but exhibit limited sensitivity, while AMR and GMR sensors offer higher sensitivity and are widely used in industrial applications, but require more complex manufacturing processes. TMR sensors provide even greater sensitivity but at the cost of increased complexity and power requirements. Optical magnetometers and atomic magnetometers are extremely sensitive, with atomic magnetometers capable of detecting magnetic fields down to the femto‑Tesla range; however, their high cost and operational complexity limit their use to specialized scientific applications. MEMS Lorentz‑force magnetometers utilize microelectromechanical systems (MEMS) technology for compact and low‑power designs but are less effective at detecting weak magnetic fields. Fluxgate magnetometers are suitable for low‑frequency applications, but their bulky design and high‑power consumption make them unsuitable for compact systems. SQUIDs, on the other hand, achieve unparalleled sensitivity but require cryogenic cooling, which increases both cost and complexity14–17. SCMs have inherent advantages that set them apart from other magnetometer types. They are passive sensors that require no external power for their operation18, which inherently leads to low‑noise levels and ultra‑low‑power consumption. They can operate for several years on small batteries19. Additionally, SCMs exhibit a low temperature dependence20, making them stable in varying environmental conditions. Their simple implementation21 based on Faraday’s law of electromagnetic induction makes them a reliable and straightforward choice for many applications. Importantly, SCMs have the widest dynamic range among all types of magnetometers, allowing them to measure a broad range of magnetic field strengths from very weak to extremely strong fields without saturation22. However, traditional SCMs are bulky and not compatible with CMOS technology. This bulkiness severely hinders their application in cases that require miniaturized sensors, such as portable devices, biomedical implants, and compact IoT systems. Limited research has focused on the miniaturization of traditional SCMs, despite the growing demand for compact and power‑efficient magnetic field sensors23–27. Miniaturization of SCMs can be achieved using planar spiral inductors, either at the printed circuit board (PCB)28,29 level or at the CMOS level23,24. While PCB‑based planar spiral inductors offer a scalable and cost‑effective solution, they are generally limited in resolution and integration capabilities. CMOS‑based planar spiral inductors, on the other hand, provide an opportunity for seamless integration with on‑chip circuitry, enabling reduced size, improved performance, and lower power consumption. Planar spiral inductors have been extensively used in high‑frequency applications, such as RF circuits, due to their efficient inductance properties and ease of fabrication30,31. However, their application in low‑frequency magnetic field sensing has been largely unexplored. Sharma et al.23 presented the development of a 3­D AC magnetic sensor fully implemented in a standard 65­nm CMOS process. The sensor utilizes three orthogonal, highly dense on‑chip coils (X, Y, and Z axes) for electromagnetic induction‑based sensing, achieving a magnetic field resolution of <10 µTrms at a frequency of 500 Hz. It operates within a measurement range of ±10 mT and consumes an ultra-­low power of only 14.8 µW, making it highly suitable for power‑constrained biomedical applications. The sensor’s compact design integrates all sensing and signal processing circuitry, including a low‑noise INA, bandpass filter, programmable gain amplifier, peak detection circuit, and a 12‑bit SAR ADC, on a single chip. The coil resistance is 8 MΩ for the X and Y sensor coils and 5.4 MΩ for the Z sensor coil, corresponding to thermal noise floors of 364 nV/√Hz and 300 nV/√Hz, respectively. No coil‑design optimization was employed to improve the resolution. Eyre et al.24 presented a multi‑axis micro‑coil magnetic sensor using a standard CMOS fabrication process combined with post‑fabrication micromachining techniques. The sensor consists of three orthogonal sensing coils, with two coils rotated perpendicular to the substrate using aluminum hinges and one coil remaining on the substrate. Fabricated using a two‑layer aluminum metallization process, the coils are optimized for sensitivity and resolution, which are frequency‑dependent and exhibit a linear frequency response. The sensor demonstrates temperature independence, making it suitable for reliable operation in varying environmental conditions. Integrated on‑chip electronics, including amplification, filtering, and analog‑to‑digital conversion, minimize signal degradation and enhance performance. With a large inherent bandwidth (limited only by the associated amplifiers) and resistance, inductance, and capacitance values enabling gigahertz operation, the sensor achieves resolution as low as 35 nT at 1 MHz with a sensitivity of 40 V/T and a die size of less than 5 mm². This work requires post‑CMOS processing. Moreover, the coil design was not optimized. A low‑noise ASIC preamplifier presented by Rhouni et al.32 was designed and fabricated for traditional bulky SCMs, which require large input transistor pairs to achieve extremely low‑noise levels. To meet the stringent noise requirements, the first amplification stage employs input transistors with an exceptionally large gate area (W/L = 20,000 µm/10 µm). This design enables the preamplifier to achieve an input‑referred voltage noise of 4 nV/√Hz at 10 Hz, demonstrating its ability to process weak magnetic signals with high sensitivity. Although the preamplifier achieves an impressive NEMI of 2 pT/√Hz at 10 Hz, it is tested with a traditional SCM that uses a ferromagnetic core with a diameter of 3 mm and a length of 10 cm. These significant core dimensions are essential for amplifying weak magnetic signals, underscoring the compatibility of the ASIC preamplifier with large‑scale, conventional SCMs. A simulation‑driven design and fabrication of a µSCM optimized for low‑magnetic field detection was presented in ref.25. Using COMSOL simulations, the optimal coil design was determined to have a 60 µm thickness, 40 µm spacing, and 30 turns, balancing sensitivity and miniaturization. The use of air as the core simplifies fabrication, while flip‑chip bonding33 is introduced as an alternative to traditional fabrication methods. However, the paper does not include details on resolution, die size, or the frequency of the measured magnetic field, nor does it present any optimization to achieve a minimum NEMI. A micro‑coil sensor designed for measuring magnetic field leakage with high precision and compact size was presented by Hirota et al.20. The sensor operates based on Faraday’s law of electromagnetic induction, where the micro‑coil is positioned close to the source of the magnetic field leakage. As the leakage field varies, it induces a voltage in the coil, which is then measured as the output signal. This approach enables the detection of localized magnetic fields with fine spatial resolution. However, details regarding the sensor’s die size, specific fabrication process, or the frequency range of the measured magnetic fields are not explicitly mentioned in the paper. Cha et al.34 presented a thin‑film inductive microsensor for non‑contact and non‑destructive detection of surface flaws on both magnetic and non‑magnetic metallic specimens. The sensor features a planar inductive coil with a Co‑Ni‑Fe magnetic core, fabricated using sputtering, electroplating, dry etching, and photolithography, with compact dimensions of 288 × 360 µm and a thickness of 23 µm. Flaw detection is based on eddy current principles, where a wire‑based excitation coil (a single straight copper coil, 120 µm in diameter) generates an AC magnetic field (0.6–1.8 MHz, 0.1–1.2 A) that induces eddy currents in the specimen. Surface flaws disrupt these currents, causing magnetic field variations, which the planar inductive microsensor (with 40 turns) detects as induced voltage signals, enabling precise identification of flaw positions, shapes, and dimensions. However, the sensor demonstrates a high spatial resolution of 0.1 mm during flaw detection and imaging; the sensitivity, noise level, and resolution of the micro‑coil are not calculated or measured. Later, Cha et al.35 showed how increasing coil turns, magnetic permeability, driving frequency (0.5–2.0 MHz), and current (0.1–1.4 A) enhance sensitivity. The sensor enables precise detection and imaging of cracks with a spatial resolution of 50 µm per scan, offering both crack profile and material‑specific information. Despite a better spatial resolution, the performance metrics of a µSCM have not been studied comprehensively. One of the main advantages of a planar spiral inductor is its ability to operate across a wide range of input magnetic field frequencies. For instance, in ref.36, a planar spiral inductor is employed to detect magnetic fields in the megahertz range. In this work, a miniaturized on‑chip spiral inductor is developed as a loop antenna sensor for detecting partial discharge (PD) activity in high‑voltage equipment. Fabricated using CMOS technology, the sensor features a 70‑turn square planar inductor with a side length of 1.8 mm. It detects magnetic fields generated by PDs, producing an induced voltage of approximately 100 mV at a separation distance of 5 cm, with a resonant frequency of ~5 MHz. Although the design parameters of the planar spiral inductor and an experimental equivalent circuit model are included, no optimization method is proposed to improve the signal‑to‑noise ratio. Because of their operating principle based on Faraday’s law, fixed SCMs cannot be used to measure DC magnetic fields37. Although vibrating‑coil sensors can be used to measure DC magnetic fields, they require a more complicated fabrication process to realize moving parts38. A µSCM can also be implemented using low‑cost PCB multilayer technology. Ulvr et al.29 presented the design, fabrication, and comprehensive characterization of single‑layer, double‑layer, and 10‑layer PCB SCMs for measuring AC magnetic flux density up to 1 T in a low‑frequency range (up to 10 kHz) within a 10 mm air gap. The study compares analytical design methods, finite integration technique (FIT) simulations, and experimental results, demonstrating strong agreement in key parameters such as resistance, inductance, and the search‑coil constant. However, while analytical calculations for resistance, inductance, and the total area of the PCB coil are provided, the work does not address NEMI optimization through theoretical calculations, FEM simulations, or experimental approaches. Ka et al.39 introduced a µSCM designed for high‑precision magnetic field distribution measurements with a spatial resolution of 0.1 × 0.2 mm and a magnetic field resolution of 1 mT. The magnetometer incorporates a single‑turn SCM fabricated using PCB technology (0.2 mm width, 100 mm length) and a Terfenol‑D actuator vibrating at 1.8 kHz, which enhances sensitivity without requiring a bias magnetic field. To reduce noise caused by the stray AC magnetic field from the actuator, the system utilizes an even‑function λ‑H magnetostriction curve and a lock‑in amplifier technique. While the magnetometer demonstrated accurate magnetic field distribution measurements closely aligned with FEM simulations, it lacks a detailed theoretical analysis of its spatial resolution and sensor resolution. Donzier et al.40 reported an integrated magnetic field sensor based on a resonant silicon beam actuated by the Laplace force, with piezoresistive readout, achieving nanotesla‑level resolution at a mechanical resonance around 10 kHz in air. More recently, Niekiel et al.41 demonstrated fully MEMS‑fabricated piezoelectric cantilever sensors with integrated NdFeB powder magnets, reaching sensitivities of 34.6–43.4 kV/T and limits of detection of 7.2–8.5 pT/√Hz at resonance frequencies of 2.2–3.6 kHz. In a different direction, Manyosa et al.42 proposed a hybrid MEMS–TMR magnetometer using a movable flux concentrator to modulate the field at a commercial TMR element, achieving ≈20 nT/√Hz at 0.1 mHz and ≈1.5 nT/√Hz from 100 mHz to 1 Hz for space‑mission‑grade ultra‑low‑frequency monitoring. While these MEMS‑based approaches achieve excellent sensitivity in their respective (often narrow) frequency bands, they are not fully CMOS‑integrated SCMs and do not address systematic NEMI optimization in the 1 Hz–1 kHz range targeted in this work. Herein, we present, for the first time, a fully CMOS‑integrated µSCM with a low‑noise INA, accompanied by a comprehensive theoretical analysis for low‑frequency magnetic field measurements in the range of 1 Hz to 1 kHz. The critical line width and inner diameter are theoretically calculated and experimentally verified, enabling the development of a unique CMOS layout. This layout incorporates the calculated critical inner diameter, significantly optimizing the area utilization on the CMOS die. Additionally, the CMOS die includes a polysilicon‑layer inductor for on‑chip calibration, a Hall‑effect sensor for DC magnetic field measurement, and a temperature sensor for enhanced functionality. Our work marks a major advancement in the field, delivering a compact, energy‑efficient, and highly sensitive solution for low‑frequency magnetic field measurements, surpassing the performance of most commercially available Hall‑effect sensors. The innovation of this work lies not merely in the integration of established components, but in the systematic, performance‑driven optimization and novel co­integration that led to an order‑of‑magnitude breakthrough in NEMI for a fully CMOS µSCM. Specifically, the main contributions are: Novel design methodology for NEMI minimization: We present, for the first time, a comprehensive theoretical framework to systematically analyze the planar spiral inductor’s design space (including line width, spacing, and inner/outer diameters) and derive analytical expressions for the critical line width (wc) and critical inner diameter (Dic) that yield the absolute minimum NEMI. This optimization technique is a major contribution to µSCM design. Novel design methodology for NEMI minimization: We present, for the first time, a comprehensive theoretical framework to systematically analyze the planar spiral inductor’s design space (including line width, spacing, and inner/outer diameters) and derive analytical expressions for the critical line width (wc) and critical inner diameter (Dic) that yield the absolute minimum NEMI. This optimization technique is a major contribution to µSCM design. Unique integrated‑within‑aperture layout: We introduce a novel, area‑efficient co‑integration strategy where the optimized low‑noise INA is placed entirely within the inner aperture of the planar spiral inductor coil (defined by the critical inner diameter, Dic). This layout significantly maximizes area utilization and minimizes the required silicon footprint. Unique integrated‑within‑aperture layout: We introduce a novel, area‑efficient co‑integration strategy where the optimized low‑noise INA is placed entirely within the inner aperture of the planar spiral inductor coil (defined by the critical inner diameter, Dic). This layout significantly maximizes area utilization and minimizes the required silicon footprint. Advanced low‑noise readout circuitry: We designed a high‑performance nested‑chopped current‑feedback INA specifically tailored for µSCM readout. The nested‑chopping technique effectively suppresses the dominant 1/f noise (flicker‑noise) inherent in CMOS processes, leading to a significant reduction in input‑referred voltage noise (from 73.45 nV/√Hz to 36.92 nV/√Hz at 100 Hz). Advanced low‑noise readout circuitry: We designed a high‑performance nested‑chopped current‑feedback INA specifically tailored for µSCM readout. The nested‑chopping technique effectively suppresses the dominant 1/f noise (flicker‑noise) inherent in CMOS processes, leading to a significant reduction in input‑referred voltage noise (from 73.45 nV/√Hz to 36.92 nV/√Hz at 100 Hz). Benchmark performance and versatility: The combination of optimal coil design and low‑noise circuitry achieves a record‑setting NEMI of 2.1 nT/√Hz at 1 kHz while consuming only 364 µW, resulting in a power-normalized field-resolution efficiency ratio that is more than an order of magnitude superior to previously reported fully integrated magnetic sensors. Benchmark performance and versatility: The combination of optimal coil design and low‑noise circuitry achieves a record‑setting NEMI of 2.1 nT/√Hz at 1 kHz while consuming only 364 µW, resulting in a power-normalized field-resolution efficiency ratio that is more than an order of magnitude superior to previously reported fully integrated magnetic sensors. Results and discussion Working principle and realization of the sensor system The working principle of the presented µSCM is based on Faraday’s law of induction. When a planar spiral inductor is exposed to an external alternating magnetic field, a voltage (Vout) is induced at the output terminals of the coil as shown in Fig. 1a. A planar spiral inductor can be defined by its layout parameters: w, s, Di, Do, and line thickness (h). Since the metal thickness and sheet resistance are predetermined in any CMOS process, we used w, s, Di, and Do to optimize the presented µSCM. A planar spiral inductor can be approximated with closed‑loop turns. This approximation of the planar spiral inductor with closed‑loop turns and its design parameters (w, s, Di, and Do) are shown in Fig. 1b. The µSCM is implemented using the six metal layers available in the 1P6M TSMC 180 nm CMOS process. A planar spiral inductor is formed in each metal layer, and all six inductors are connected in series and in phase. An additional inductor is implemented in the polysilicon layer for on‑chip calibration. The schematic of six metal inductors and the polysilicon inductor is depicted in Fig. 1d. Working principle and realization of the CMOS die. a Working principle of the µSCM. b Approximation of a planar spiral inductor with closed‑loop turns. c 3D schematic view of the CMOS die. d Realization of the µSCM using six metal layers in the CMOS process. e Block diagram of the CMOS die. f Layout of the CMOS die. g Photograph of the wire‑bonded die. h QFN‑packaged CMOS die In this configuration, the six metal‑layer coils behave effectively as a single µSCM. Under the low‑frequency operating conditions of this work (1 Hz–1 kHz), the induced voltage on the µSCM is in the µV–mV range and is applied directly to the high‑impedance input of the PMOS input pair of the INA. As a result, the coil current is very small, and the magnetic field generated by this current—as well as the associated mutual fields among the stacked metal layers—is negligible compared with the externally applied magnetic field. Therefore, mutual coupling between the stacked metal coils does not materially affect the magnetic‑field‑to‑voltage transfer function or the NEMI analysis. In contrast, mutual inductance is intentionally exploited only between the polysilicon excitation coil and the metal‑layer µSCM in the on‑chip calibration structure described later. The block diagram of the CMOS die is illustrated in Fig. 1e. The CMOS die consists of a µSCM connected to a low‑noise INA, a temperature sensor, a Hall‑effect sensor, and a polysilicon inductor for on‑chip calibration of the µSCM. The total size of the CMOS die is 3 × 3 mm. A critical aspect of this work is the full integration and fabrication of the entire µSCM system—including the coil, the low‑noise INA, the on‑chip calibration coil, temperature sensor, Hall‑effect sensor, and associated passive components—using a standard commercial 1P6M TSMC 180 nm CMOS process. This approach eliminates the need for expensive, complicated, and low‑yield post‑CMOS MEMS processing steps (such as sacrificial-layer etching), ensuring the system is highly manufacturable, low‑cost, and scalable. The final layout of the CMOS die, which includes the active and passive components of the designed INA, and the CMOS MEMS sensors, is shown in Fig. 1f. The fabricated die was wire‑bonded into a 48‑pin 5 × 5 mm Quad Flat No‑Lead (QFN) package (Fig. 1g) and then molded with an epoxy compound to protect the wire bonds. The QFN package was then mounted on the designed PCB for characterization. The photograph of the QFN‑packaged die after wire bonding and epoxy molding is shown in Fig. 1h. Amplifier characterization The designed and fabricated INA is specifically developed to amplify the induced voltage at the output terminals of the µSCM, which operates in a low‑frequency range (<1 kHz). In this frequency range, flicker‑noise becomes a dominant noise source, significantly impacting the NEMI of the µSCM. Addressing this challenge is critical to achieving the desired sensitivity and performance of the sensor system. Before applying the chopping technique, the input‑referred noise of the INA shows significant flicker‑noise contributions, as illustrated in Fig. 2a. The corner frequency of the flicker‑noise exceeds 6 kHz, meaning that the noise remains significant even at lower operational frequencies. For instance, at 100 Hz, the input‑referred noise of the INA is measured at 73.45 nV/√Hz, demonstrating that flicker‑noise substantially increases the overall noise floor of the system and degrades the NEMI. Simulation and measurement results for the designed and fabricated INA. a Post‑layout simulation results for the input­-referred noise of the INA before chopping. b Post‑layout simulation results for the input­-referred noise of the INA after chopping. c Output voltage noise measurement of the INA, demonstrating that chopping pushes the flicker‑noise corner to sub‑hertz frequencies. d Post‑layout simulation and measurement results for the gain frequency response of the fabricated INA. e Post‑layout simulation and measurement results for the phase frequency response of the fabricated INA To mitigate this issue, the chopping technique is employed. This technique effectively modulates the input signal to a higher‑frequency band, where flicker‑noise is less significant, and then demodulates it back to the baseband after amplification. By doing so, the flicker‑noise is pushed to frequencies well below the operational range, drastically reducing its impact. As shown in Fig. 2b, the flicker‑noise corner frequency is shifted from greater than 6 kHz to less than 100 mHz, effectively minimizing its contribution to the overall noise. The results demonstrate a substantial improvement in noise performance. At 100 Hz, the input‑referred noise is reduced from 73.45 nV/√Hz (without chopping) to 36.92 nV/√Hz (with chopping), representing a nearly 50% reduction. This significant decrease in noise ensures that the flicker‑noise no longer limits the NEMI of the µSCM, enabling higher sensitivity and better performance in low‑frequency magnetic field measurements. The successful implementation of the chopping technique highlights its effectiveness in overcoming the challenges posed by flicker‑noise in low‑frequency INAs. The measured output voltage noise up to 3.125 Hz is shown in Fig. 2c. The 1/f corner is below 1 Hz, confirming that the employed chopping technique effectively pushes flicker‑noise to sub‑hertz frequencies. The measurements align well with post‑layout simulations (output noise ≈ input‑referred noise × closed‑loop gain). The average output noise in the flat region is −100.47 dB, compared with −101.29 dB predicted by the post‑layout simulation results. This small discrepancy is mainly due to the output‑stage noise (eno) of the INA. The closed‑loop frequency response of the designed and fabricated INA was measured using a Vector Network Analyzer (Bode 100, Omicron Lab, Klaus, Austria) and obtained through post‑layout simulations performed with Cadence software. The results, including the gain and phase frequency responses, are compared in Fig. 2d and Fig. 2e, respectively. Both the simulated and measured results show excellent agreement, validating the accuracy of the design and fabrication processes. The measured and simulated DC gains are 47.38 dB and 47.27 dB, respectively, demonstrating minimal deviation between simulation and experimental results. Similarly, the measured and simulated bandwidths are 2.89 kHz and 2.86 kHz, respectively, showcasing a close match. This limited bandwidth acts as a natural low‑pass filter, effectively suppressing high‑frequency noise and electromagnetic interference, which is particularly important for low‑frequency applications such as precision sensing. The measurements in Fig. 2d show a monotonic magnitude response with no gain peaking at the chopper frequency or its harmonics. The absence of excessive gain peaking in the gain response (Fig. 2d) and the gradual phase roll‑off indicate that the system is stable under closed‑loop conditions. A stable phase roll‑off without abrupt changes suggests sufficient phase margin, implying that the INA is robust against instability. Furthermore, the close alignment between the measured and simulated phase response confirms that the stability considerations during the design stage (use of the nested Miller compensation technique) successfully translated to the fabricated device. µSCM characterization results The characterization setup for the µSCM is shown in Fig. 3a. This setup includes an in‑house electromagnet, a custom‑built power amplifier, a DC power supply (Tektronix PS280), and a function generator (Agilent 33120 A). The power amplifier was specifically designed and fabricated to provide sufficient current to drive the electromagnet. A commercial Hall‑effect sensor (A1324LLHLT‑T, Allegro Microsystems Inc., Manchester, NH, USA) was used as a reference to measure the magnetic field generated between the poles of the electromagnet as a function of the input voltage applied to the power amplifier. The Hall probe’s sensitivity was verified against its datasheet calibration and cross‑checked by repeated sweeps. Characterization of the fabricated µSCM. a In‑house characterization setup. b Time‑domain input‑output response at 100 Hz magnetic field. c Measurement results for the sensitivity of the µSCM as a function of the input magnetic field frequency. d Output noise measurement results of the µSCM connected to the INA. e Measured and predicted results for the NEMI of the µSCM as a function of input magnetic field frequency To measure the output voltage of the INA, which amplifies the induced voltage on the µSCM, a low‑input‑capacitance passive probe (TPP0101) connected to an oscilloscope (Tektronix TBS 1052B‑EDU) was utilized. The predicted and measured time‑domain input‑output responses of the CMOS µSCM are shown in Fig. 3b. The predicted output voltage of the INA was calculated using Eq. (6) and the simulated DC gain of the INA (47.27 dB). As illustrated in this figure, the measured results closely match the predicted values, demonstrating a strong correlation between the theoretical and experimental analyses. The sensitivity of the µSCM as a function of the input magnetic field frequency was measured using the characterization setup. The comparison between the measured and predicted sensitivities is depicted in Fig. 3c. Within the frequency range of interest (<1 kHz), the sensitivity is shown to be linearly proportional to the frequency. The excellent agreement between the measured and predicted results confirms the validity of the theoretical analysis and the accuracy of the system design. To evaluate noise performance, the output noise of the INA at zero input magnetic field was measured using a spectrum analyzer (HP DSA Dynamic Signal Analyzer 35665A) as shown in Fig. 3d. The measured NEMI of the µSCM as a function of input magnetic field frequency is depicted in Fig. 3e. Using the measured output noise, the input current noise of the INA was calculated based on Eq. (1). This equation is derived from Eq. (7), which is explained in detail in the “Theoretical analysis” section. The eno/G term was neglected in this calculation, as the output noise voltage (eno) is typically small and further divided by the gain (G), rendering this term negligible43,44. The calculated input current noise of the INA was found to be 291.8 pA/√Hz. Notably, the measured DC resistance R1 (coil formed by M6, M5, and M4 layers) is 47.75 kΩ, while R2 (coil formed by M3, M2, and M1 layers) is 67.66 kΩ. Moreover, comparison of the INA noise measurements and simulations in Fig. 2 shows that the output‑stage contribution is <1%. Using (en)² = (eni)² + (eno/G)² and Fig. 2c with G = 233.62, indicates that the contribution of eno is <1% across the fitted band. Therefore, the eno/G term can be safely ignored in the current noise calculation of Eq. (1). Finally, after obtaining the sensitivity and noise measurement results, the NEMI of the µSCM was computed using Eq. (8). At a 1 kHz magnetic field, the NEMI was determined to be 2.1 nT/√Hz, which is significantly better than most commercially available fully CMOS magnetic field sensors, such as commercial Hall‑effect sensors. This result highlights the superior noise performance and sensitivity of the proposed µSCM design, validating its potential for high‑precision low‑frequency magnetic field sensing applications. Characterization of the on‑chip calibration structure, temperature sensor, and Hall‑effect sensor To implement the on‑chip calibration structure on the CMOS die, the polysilicon layer beneath the µSCM was utilized to fabricate an excitation coil, as illustrated in Fig. 1c, d. The design parameters of the polysilicon coil include a line width of 20 µm, a line spacing of 0.5 µm, outer dimensions of 2960 × 2850 µm, and a total of 51 turns. The working principle of the on‑chip calibration is based on the mutual inductance between the polysilicon coil and the µSCM, as described in previous studies45–47. When a current is passed through the polysilicon coil, it induces a voltage on the µSCM due to the mutual inductance. This feature is particularly useful for long‑term applications where periodic calibration of the µSCM is essential to maintain accuracy and reliability. The calibration process was tested by applying a 200 mVpp sinusoidal voltage with a frequency of 250 Hz to the polysilicon coil. The output voltage measurement results are shown in Fig. 4a. These results clearly demonstrate the advantage of the narrow bandwidth of the INA, which acts as a low‑pass filter, effectively suppressing high‑frequency noise. Furthermore, the relationship between the peak‑to‑peak output voltage of the µSCM and the frequency of the excitation voltage applied to the polysilicon coil is shown in Fig. 4b. This characterization highlights the functionality of the on‑chip calibration structure across different frequencies. Measurement results of other features of the CMOS die. a Time‑domain input‑output response of the on‑chip calibration at 250 Hz, and 100 mV amplitude (200 mVpp) input voltage applied to the polysilicon coil. b On‑chip calibration measurement results as a function of excitation frequency for an input amplitude of 200 mVpp. c Measurement results for the Hall‑effect sensor. d Measurement results for the temperature sensor Because the INA consumes only 364 µW, this approach offers significant potential for future applications, including wireless powering of the sensor system. Future work will focus on integrating wireless power transfer techniques to enhance the usability of the µSCM in portable and remote sensing applications. The CMOS die also integrates additional features that complement the μSCM. In particular, a Hall‑effect sensor was integrated to enable DC magnetic field measurement, providing a low‑frequency/DC modality that complements the μSCM, which targets AC fields (1 Hz–1 kHz). The Hall element is implemented as an 80 × 80 µm crossbar. Its input–output response is shown in Fig. 4c. The measured offset voltage of 4.36 mV was subtracted when calculating the reported sensitivity. Without any amplification, the Hall‑effect sensor exhibits a sensitivity of 50 µV/mT, demonstrating suitability for low‑power DC magnetic field sensing. Additionally, a temperature sensor fabricated using n‑doped polysilicon is integrated into the CMOS die and shown in Fig. 1f. The input‑output response of the temperature sensor is displayed in Fig. 4d. The temperature sensor serves a dual purpose: monitoring the environmental temperature and enabling temperature compensation for the µSCM. However, studying the temperature’s effect on the µSCM’s performance and implementing precise temperature compensation requires a more sophisticated measurement setup than the one shown in Fig. 3a. This investigation will be reported in future work. In this prototype, the integrated Hall‑effect and temperature sensors are not on‑chip calibrated (trimmed); they are included as auxiliary monitors to enable future temperature and DC bias compensation, rather than to derive the main µSCM NEMI results. Both sensors were characterized off‑chip, using external calibrated references to extract their sensitivity. Finally, a comparative analysis between the presented µSCM and commercially available magnetic field sensors, as well as previously reported µSCMs, is summarized in Table 1. This table is restricted to fully integrated or microfabricated CMOS µSCMs for the low‑frequency range, and products with clearly reported sensitivity and noise/resolution from which NEMI‑like metrics can be taken or reliably derived. Bulky wire‑wound/core‑based µSCMs are discussed qualitatively for context but are not included in Table 1 due to their fundamentally different form factor and implementation. The comparison highlights the superior performance of the presented µSCM, particularly in terms of NEMI. The NEMI of the presented µSCM outperforms other designs by more than one order of magnitude. While input‑referred field noise density (nT/√Hz), sensitivity (V/T), and power consumption remain the primary comparison metrics, we additionally report an efficiency figure defined as NEMI⁻¹/P × 10⁹ (nT⁻¹ × W⁻¹ × √Hz). This composite metric is not a community standard, but it compactly captures the trade‑off between resolution (lower NEMI is better) and power (lower P is better) in a single number: higher values indicate more field‑noise‑limited sensitivity per watt. The rationale is that, for readout‑limited inductive sensors, reducing amplifier noise typically requires higher bias and thus higher power; NEMI⁻¹/P × 10⁹ provides a normalized view of how effectively power is converted into magnetic resolution, complementing the conventional side‑by‑side reporting of nT/√Hz and power. We emphasize that all conclusions are based on standard metrics (nT/√Hz at specified frequencies, sensitivity, and power). The NEMI⁻¹/P × 10⁹ figure is included only as an auxiliary indicator of power–resolution efficiency to facilitate quick visual ranking across designs operating in comparable bandwidths. Under identical normalization, our μSCM achieves an efficiency of 1308 (nT⁻¹ × W⁻¹ × √Hz) on this metric, exceeding the other entries in Table 1 by more than an order of magnitude. We provide the underlying nT/√Hz, sensitivity, and power alongside this figure to ensure transparency and reproducibility. Table 1Comparison between the presented µSCM, selected commercial magnetic field sensors, and previous worksRefSensor typeS1*NEMI (resolution)P2* (mW)NEMI‑1/P × 109 3*A13244*Hall‑effect505*(R6*)130 (nT/√Hz) (R6*)34.5 (D7*)0.223 (D7*)HMC5883L8*AMR10909* (R6*)500 (nT/√Hz) (R6*)0.25 (D7*)8 (D7*)LSM303AGR10*AMR1.59* (R6*)300 (nT/√Hz) (R6*)0.5 (D7*)6.67 (D7*)TI‑DRV5056A111*Hall‑effect2005* (R6*)130 (nT/√Hz) (R6*)30 (D7*)0.26 (D7*)MLX91205LB12*Hall‑effect2805* (R6*)125 (nT/√Hz) (R6*)55 (D7*)0.146 (D7*)Sharma et al.23CMOS µSCM413*(D7*)1.2 µTrms14* (R6*)0.0148 (R6*)NA15*Eyre et al.24CMOS µSCM4016* (D7*)3517* (R6*)NA15*NA15*This workCMOS µSCM670818*2.119*0.364130820*1* Sensitivity, 2* Power consumption, 3* (nT⁻¹ × W⁻¹ × √Hz), 4* Allegro Microsystems, 5* mV/mT, 6* Reported (R), 7* derived (D), 8* Honeywell, 9* LSB/Gauss, 10* STMicroelectronics, 11* Texas Instruments, 12* Melexis, 13* (mV/mT) @ 500 Hz, 14* @ 500 Hz (in 100 kHz BW, averaging 40 samples), 15* not available (NA), 16* (mV/mT) @ 1 MHz, 17* in 100 kHz BW @ 1 MHz, 18* (mV/mT) @ 1 kHz, 19* (nT/√Hz) @ 1 kHz, 20* @ 1 kHz Comparison between the presented µSCM, selected commercial magnetic field sensors, and previous works 1* Sensitivity, 2* Power consumption, 3* (nT⁻¹ × W⁻¹ × √Hz), 4* Allegro Microsystems, 5* mV/mT, 6* Reported (R), 7* derived (D), 8* Honeywell, 9* LSB/Gauss, 10* STMicroelectronics, 11* Texas Instruments, 12* Melexis, 13* (mV/mT) @ 500 Hz, 14* @ 500 Hz (in 100 kHz BW, averaging 40 samples), 15* not available (NA), 16* (mV/mT) @ 1 MHz, 17* in 100 kHz BW @ 1 MHz, 18* (mV/mT) @ 1 kHz, 19* (nT/√Hz) @ 1 kHz, 20* @ 1 kHz This significant improvement underscores the exceptional efficiency and performance of the presented µSCM, making it a strong candidate for high‑precision, low‑power magnetic field sensing applications. Discussion The experimental results confirm that the proposed NEMI‑driven design methodology and integrated‑within‑aperture layout translate directly into system‑level performance gains. The measured NEMI of 2.1 nT/√Hz at 1 kHz with a power consumption of only 364 µW agrees well with the predictions from the analytical framework in Eqs. (5–8) and the extracted INA noise parameters. In particular, we observed the linear frequency dependence of the sensitivity up to 1 kHz, the measured coil resistances consistent with design, and the input‑referred noise reduction from 73.45 nV/√Hz to 36.92 nV/√Hz at 100 Hz after nested chopping. This validates both the analytical identification of the critical line width and the inner diameter and the practical design rule of optimizing design parameters under process constraints as an effective means to minimize NEMI in CMOS µSCMs. When compared to prior CMOS micro‑coil sensors and MEMS magnetic sensors, the proposed µSCM occupies a distinct design space. Earlier CMOS µSCM works by Eyre et al. and Sharma et al. focused on demonstrating multi‑axis integration and wide‑band operation but did not systematically optimize the coil geometry for NEMI at low frequencies, nor did they co­design the readout amplifier to match the thermal noise of the coil. Resonant MEMS magnetometers based on permanent magnets or magnetostrictive layers20,22,40–42 achieve pT/√Hz‑level resolution at kHz mechanical resonance or, in the case of hybrid MEMS–TMR systems, at sub‑millihertz frequencies for space applications. However, they typically operate in narrow frequency bands, rely on hybrid stacks or specialized materials, and are not realized as fully monolithic CMOS inductive sensors. In contrast, our µSCM combines a broadband inductive sensing principle (1 Hz–1 kHz), a standard 180 nm CMOS process, and a NEMI‑optimized stacked‑coil plus INA co‑design, yielding more than an order of magnitude improvement in NEMI and NEMI⁻¹/P over both prior CMOS µSCMs and commercial Hall‑based CMOS sensors. These results suggest several directions for future work. On the sensor side, the same analytical framework can be extended to other technology nodes, metal stacks, and form factors (e.g., multi‑axis implementations or arrays), and can incorporate parasitic capacitance and eddy‑current effects at higher frequencies. On the circuit side, further reductions in NEMI could be achieved by modestly increasing INA bias current, exploring auto‑zeroing or correlated‑double‑sampling schemes that are co‑optimized with the coil resistance, or adding on‑chip digital calibration and temperature compensation using the integrated temperature sensor. Finally, the demonstrated on‑chip calibration via a polysilicon excitation coil opens a path toward long‑term, self‑verifying µSCM nodes that could be wirelessly powered and networked for distributed, low‑frequency magnetic monitoring in IoT, biomedical, and energy‑monitoring applications. Materials and methods Theoretical analysis The working principle of the µSCM is based on Maxwell‑Faraday’s induction law, which can be expressed by (2): where ∇⃗× is the curl operator, E⃗ is the induced electric field, and Bin⃗ is the input magnetic field. If the µSCM is parallel to the xy‑plane, and the input magnetic field is uniform and perpendicular to the surface of the coil (Bin⃗ = Bm sin(2πft)z⃗), the electric field can be obtained in the cylindrical coordinates by (3): where Bm is the maximum input magnetic field, f is frequency, t is time, r is the distance from the center of the coil, and φ⃗ is the unit vector in the cylindrical coordinates. A planar spiral inductor can be approximated by closed‑loop filaments in magnetostatics, as shown in Fig. 1b, and the induced voltage on each filament can be calculated by the line integral of the electric field around each turn. Therefore, the maximum induced voltage on the µSCM (Vout_m) can be calculated by (4): where Ai and n are the area of the i-th turn and the number of turns of the planar spiral inductor, respectively. The sensitivity (S) of a µSCM is defined by the ratio of the Vout_m to Bm and is formulated by (5): The sensitivity of the µSCM can be calculated using (6), provided that all available metal layers (M) in the CMOS process, where M is fixed for a given CMOS process, are utilized for designing the stacked inductors. The NEMI (which is the minimum detectable magnetic field and called resolution) of a µSCM is defined by the ratio of the input‑referred noise to the sensitivity19,32,48. If an INA with input current noise (ini), input voltage noise (eni), output noise (eno), and a known DC gain (G) is connected to the output terminals of a µSCM, the total input‑referred noise of the µSCM is calculated by (7): where R1 and R2 represent the resistances of the middle‑pin‑grounded coil (as shown in Fig. 1d), measured between the ground and the input terminals of the INA, and enR1 and enR2 are the thermal noise of R1 and R2, respectively. Therefore, the NEMI of the µSCM, which is connected to the INA in the CMOS process with M metal layers, can be calculated by (8): In the implemented design, the abstract parameters in Eq. (8) are instantiated by the concrete layout values. These choices uniquely define the number of turns per layer, the total effective area ∑Ai, and the coil resistances R1 and R2 that enter the thermal‑noise terms23 in Eq. (8), thereby linking the geometric parameters directly to the measured NEMI. To study the effects of all design parameters (M, Di, Do, w, and s) of the µSCM on NEMI, Eqs. (9)–(13) must be solved simultaneously: The effects of the five aforementioned design parameters on NEMI are illustrated in Fig. 5a–e. According to the theoretical analysis and the results shown in Fig. 5a–e, the following conclusions can be drawn: The NEMI decreases monotonically when M increases. The NEMI decreases monotonically when M increases. The NEMI decreases monotonically when Do increases. The NEMI decreases monotonically when Do increases. The NEMI decreases monotonically when s decreases (at low frequencies, where the coupling capacitance of the µSCM can be ignored37). The NEMI decreases monotonically when s decreases (at low frequencies, where the coupling capacitance of the µSCM can be ignored37). There is a critical Di that results in minimum NEMI. This critical Di is very important since the available area from this critical Di is used for the design of the integrated INA. This unique layout configuration (as shown in Fig. 1f) has saved a substantial area on the CMOS die compared with ref.23. There is a critical Di that results in minimum NEMI. This critical Di is very important since the available area from this critical Di is used for the design of the integrated INA. This unique layout configuration (as shown in Fig. 1f) has saved a substantial area on the CMOS die compared with ref.23. There is a critical w that results in minimum NEMI. The line width is the most important design parameter that has a significant effect on NEMI. Almost 50 percent lower NEMI will be achieved by choosing the critical w, as shown in Fig. 5e.In all NEMI plots where a single design variable is swept (Fig. 5a–e), the remaining parameters are not chosen arbitrarily but are fixed either by process/area constraints or by the NEMI‑minimizing solution of our analytical model. Specifically, the number of metal layers is set to the maximum available in the 1P6M 180 nm process (M|6), the outer dimensions are set by the largest feasible footprint within the 3 × 3 mm die (Do ≈ 2.9 mm, 2850 × 2960 µm), and the line spacing is chosen as the minimum allowed by the PDK for each layer (s|smin). Under these fixed constraints, NEMI depends jointly on the inner diameter and line width, (Di|w). To make this coupling explicit, Fig. 5f presents a 2‑D contour map of the normalized NEMI (NEMIn) as a function of (Di|w), revealing a well‑defined minimum region. In our 1P6M 180 nm implementation, this optimum occurs near (w ≈ 5 µm for M1‑M5 layers, and w ≈ 2.6 µm for M6 layer), and Di ≈ 850 µm for all layers (red marker in Fig. 5f). The 1‑D curves in Fig. 5d, e are therefore slices taken through this optimum valley: when one variable (e.g., Di) is swept, the other (w) is held at its NEMI‑optimal value, and vice versa.In the fabricated CMOS die in this paper: There is a critical w that results in minimum NEMI. The line width is the most important design parameter that has a significant effect on NEMI. Almost 50 percent lower NEMI will be achieved by choosing the critical w, as shown in Fig. 5e. In all NEMI plots where a single design variable is swept (Fig. 5a–e), the remaining parameters are not chosen arbitrarily but are fixed either by process/area constraints or by the NEMI‑minimizing solution of our analytical model. Specifically, the number of metal layers is set to the maximum available in the 1P6M 180 nm process (M|6), the outer dimensions are set by the largest feasible footprint within the 3 × 3 mm die (Do ≈ 2.9 mm, 2850 × 2960 µm), and the line spacing is chosen as the minimum allowed by the PDK for each layer (s|smin). Under these fixed constraints, NEMI depends jointly on the inner diameter and line width, (Di|w). To make this coupling explicit, Fig. 5f presents a 2‑D contour map of the normalized NEMI (NEMIn) as a function of (Di|w), revealing a well‑defined minimum region. In our 1P6M 180 nm implementation, this optimum occurs near (w ≈ 5 µm for M1‑M5 layers, and w ≈ 2.6 µm for M6 layer), and Di ≈ 850 µm for all layers (red marker in Fig. 5f). The 1‑D curves in Fig. 5d, e are therefore slices taken through this optimum valley: when one variable (e.g., Di) is swept, the other (w) is held at its NEMI‑optimal value, and vice versa. In the fabricated CMOS die in this paper: M was set to six, since we used the 1P6M 180 nm RF general‑purpose TSMC CMOS process. M was set to six, since we used the 1P6M 180 nm RF general‑purpose TSMC CMOS process. The outer dimensions of the fabricated µSCM were chosen to be 2850 × 2960 µm due to the available die size in this tape‑out. The outer dimensions of the fabricated µSCM were chosen to be 2850 × 2960 µm due to the available die size in this tape‑out. s was chosen to be 280 nm for M1 to M5 layers and 2.5 µm for the M6 layer due to the PDK of the CMOS process. s was chosen to be 280 nm for M1 to M5 layers and 2.5 µm for the M6 layer due to the PDK of the CMOS process. The critical Di for the utilized outer dimensions (2850 × 2960 µm) is 850 µm for all metal layers. The critical Di for the utilized outer dimensions (2850 × 2960 µm) is 850 µm for all metal layers. The number of turns (n) of the planar spiral inductor is 200 for layers M1 to M5, and 207 for the M6 layer. The number of turns (n) of the planar spiral inductor is 200 for layers M1 to M5, and 207 for the M6 layer. The critical w for M1 to M5 layers is 5 µm, and for the M6 layer is 2.6 µm. Notably, the thickness of the M6 layer in the utilized CMOS process is 4 µm, which leads to much smaller sheet resistance compared to M1‑M5 layers. The critical w for M1 to M5 layers is 5 µm, and for the M6 layer is 2.6 µm. Notably, the thickness of the M6 layer in the utilized CMOS process is 4 µm, which leads to much smaller sheet resistance compared to M1‑M5 layers. Effects of design parameters on NEMI. a Effect of M on NEMI. b Effect of Do on NEMI. c Effect of s on NEMI. d Effect of Di on NEMI. e Effect of w on NEMI. f Contour plot for the normalized NEMI (NEMIn) of the μSCM as a function of inner diameter (Di) and line width (w) We distilled our optimization into a general design rule: maximize total effective area per resistance (ΣAi/R) under process constraints. Practically, this means stacking all available metals in series and in‑phase, maximizing the outer diameter, using minimum spacing at low frequencies, and choosing the line width w and inner diameter Di at their NEMI optima (w = w*, Di = Di*). The optimal Di* defines a center aperture that we utilize to place the INA, preserving minimum NEMI while reclaiming silicon area. Using this rule in 180 nm CMOS yielded a six‑layer design with w* = 5 µm (M1–M5 layers), 2.6 µm (M6 layer), Di* ≈ 850 µm, and Do ≈ 2.9 mm, achieving 2.1 nT/√Hz at 1 kHz with 364 µW power consumption. Instrumentation amplifier design The presented µSCM is designed for measuring low‑frequency magnetic fields (<1 kHz), where flicker‑noise is the dominant noise source in conventional amplifiers. To address this issue, a nested‑chopped current‑feedback INA configuration was selected. The proposed INA is realized using a 180 nm CMOS process integrated with the µSCM. To minimize mismatch, the input and feedback transconductors are designed with identical transconductance (gm) and arranged using a common‑centroid layout. These transconductors are followed by a fully differential cascode circuit, with the output stage implemented as a folded‑cascode design. Unlike traditional designs, which rely on voltage feedback, CFIAs use current feedback, allowing for a higher slew rate and wider bandwidth that remain nearly constant across varying gain settings. Additionally, CFIAs exhibit lower power consumption and reduced distortion, as their architecture avoids the high gain‑bandwidth trade­offs typically seen in traditional INAs. The input impedance in CFIAs is also high, which minimizes loading effects on the source signal. Furthermore, their design inherently reduces the influence of parasitic capacitances, leading to improved stability and better performance in low‑frequency noise‑sensitive environments, such as biomedical or precision magnetic field measurements. A detailed design analysis of this configuration has been presented in our previous works49–52. Although chopper stabilization is a well‑established technique for suppressing low‑frequency noise, our contribution lies in how it is co‑designed and integrated with the µSCM and the INA at the system level. First, the NEMI‑driven optimization of the six‑metal‑layer µSCM defines a critical inner diameter, which we exploit to place the nested‑chopped CFIA physically inside the coil aperture (Fig. 1f), enabling a highly area‑efficient, fully CMOS‑integrated µSCM. Second, the CFIA itself is tailored to this optimized coil: its input‑referred noise (in particular, the sizing of the input transistor pair and the biasing) is chosen such that the amplifier noise is commensurate with, but does not dominate, the thermal noise of the high‑resistance stacked µSCM, matching the targeted NEMI of 2.1 nT/√Hz at 1 kHz. Third, the three‑stage, nested‑Miller‑compensated architecture is designed to provide a customized and limited but well‑controlled bandwidth of approximately 2.9 kHz, which both covers the 1 Hz–1 kHz sensing band and acts as an effective low‑pass filter for chopper artifacts and out‑of‑band interference. Together, these choices link the nested‑chopped CFIA design directly to the measured magnetic field resolution of the complete integrated sensor, rather than applying chopping as a generic stand‑alone noise‑reduction technique. The block diagram of the designed and fabricated INA is shown in Fig. 6a, while the transistor‑level schematic of the first‑stage amplifier is depicted in Fig. 6b. Additionally, the transistor‑level schematic of the second and third stages (a two‑stage class‑AB output) is illustrated in Fig. 6c. The three‑stage amplifier is designed to remain stable across all process corners by employing a combination of nested Miller compensation (NMC) and improved cascode compensation. To achieve unity‑gain stability, the NMC method requires the condition in Eq. (14) to be satisfied for a phase margin of 60°53. where CM1, CM2, and CL are shown in Fig. 6a. In our design, we aim for conditional stability at closed‑loop gains of 50 (34 dB) or higher to optimize power efficiency. This approach is practical, as the application demands precise closed‑loop gains of approximately 230. Consequently, the stability condition defined in Eq. (14) can be relaxed, leading to a new stability constraint expressed in Eq. (15). To ensure the conditional stability of our amplifier, the criteria outlined in Eq. (15) were met. The final design parameters are as follows: gm1 = 104.4 µA/V, gm2 = 36 µA/V, gm3 = 843.5 µA/V, CM1 = 2 pF, CM2 = 25.22 pF, and CL = 50 pF. Nested‑chopped current‑feedback INA. a Block diagram of the INA. b Transistor level schematic of the first‑stage amplifier. c Transistor level schematic of the output stage (second and third stages) amplifier Table 2 provides the detailed transistor sizes for the first‑, second‑, and third‑stage amplifiers shown in Fig. 6. Table 2Size of all transistors in the first‑, second‑, and third‑stage amplifiers of the proposed INATransistorW/L (µm/µm)TransistorW/L (µm/µm)TransistorW/L (µm/µm)M112/2M111.2/2M213/2M215/2M121.6/2M223/2M380/2M131.6/2M232/2M480/2M141.6/2M242/2M580/2M151.6/2M252/2M680/2M166/2M262/2M712/2M1710/2M271/4M880/5M1810/2M2845/2M980/5M193.5/2M2954/2M101.2/2M203.5/2 Size of all transistors in the first‑, second‑, and third‑stage amplifiers of the proposed INA This work presents a novel CMOS‑integrated µSCM with optimized sensitivity, noise performance, and power efficiency. By leveraging the TSMC 180 nm CMOS process, the proposed µSCM utilizes a planar spiral inductor design with critical dimensions optimized to minimize NEMI. The integration of a nested­-chopped current‑feedback INA effectively suppresses flicker‑noise, improving the system’s input‑referred noise performance and enabling precise low‑frequency magnetic field measurements. The on‑chip calibration structure, realized using a polysilicon excitation coil, provides a reliable and long‑term calibration mechanism, enhancing the sensor’s usability in various applications. Experimental results demonstrate the µSCM’s exceptional performance, achieving a NEMI of 2.1 nT/√Hz at 1 kHz with a power consumption of only 364 µW. The system also includes additional features, such as a Hall‑effect sensor for DC magnetic field measurements and a temperature sensor for compensation, making the design versatile for multi‑functional sensing applications. Comparative analysis with commercial and previously reported µSCMs highlights the significant advancement achieved in terms of power-normalized field-resolution efficiency ratio (NEMI-1/P × 10⁹), which is more than one order of magnitude better than existing solutions. The proposed µSCM represents a major step forward in the development of compact, energy‑efficient, and high‑performance magnetic field sensors. Its application potential spans a wide range of fields, including biomedical diagnostics, environmental monitoring, and Internet of Things (IoT) systems. Future work will focus on integrating wireless power transfer and advanced temperature compensation techniques to further enhance the system’s performance and applicability. Supplementary information supplementary document Acknowledgements This work was partially funded by the Hong Kong Center for Construction Robotics (Hong Kong ITC‑sponsored InnoHK Center), the Hong Kong RGC projects (No. 16208522 and 16206423), and Shenzhen Sci. Tech. Innovation Commission (Ref. No. SZSTI21EG13). The authors also acknowledge the technical support of all staff at HKUST NFF and MCPF for supporting experiments in this research. Author contributions Hadi Tavakkoli designed the µSCM, conducted the experiments, and drafted the manuscript. Moaaz Ahmed, Ikramullah Shah, and Prof. Amine Bermak contributed to the layout design, design rule checks, and submission of the final design to TSMC’s 180 nm CMOS process. Sile Fang and Wenhao Chen contributed to the experimental setup design and realization, and Prof. Yi‑Kuen Lee supervised this work and checked it for final submission. Conflict of interest The authors declare no competing interests. Supplementary information The online version contains supplementary material available at 10.1038/s41378-026-01334-w. References Associated Data Supplementary Materials supplementary document