Pith. sign in

REVIEW 4 major objections 5 minor 14 references

A Spatial-Physics Informed Model for 3D Spiral Sample Scanned by SQUID Microscopy

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A three-stage pipeline (phase rotation, affine alignment, FFT-Biot-Savart inversion) converts SQUID magnetic scans of a 3D spiral into current-density maps, sharpening the in-phase channel by 0.3% and cutting quadrature blur by 25%.

desk verdict A clear, well-cited pipeline paper for SQUID current imaging, but the reported gains are tiny and the evaluation is circular; the single-plane inversion assumption is a real limitation that the paper itself acknowledges. read the letter →

arxiv 2507.11853 v1 pith:SXB7IM5B submitted 2025-07-16 physics.ins-det cs.CV

J. Senthilnath , Jayasanker Jayabalan , Zhuoyi Lin , Aye Phyu Phyu Aung , Chen Hao , Kaixin Xu , Yeow Kheng Lim , F. C. Wellstood This is my paper · ORCID
classification physics.ins-detcs.CV PACS 85.25.Dq07.55.Ge
keywords SQUIDmicroscopymagneticfieldimagingspatial-physicsinformedmodeleddycurrentaffinetransformfastFourierBiot-Savartinversionnon-destructivetesting
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper is trying to establish that raw SQUID microscope images of a 3D spiral sample can be converted into usable current-density maps by a three-stage model: phase-aligning the lock-in channels to suppress eddy-current blur, applying an affine transform to correct a 0.3° scanning misalignment, and inverting the field with a combined Biot-Savart and FFT method. On a real 254-by-1166 pixel scan, the authors report that the phase step sharpens the in-phase (I) channel by 0.3% and reduces the quadrature (Q) channel's sharpness by 25%, and that a 0.3° skew correction aligns the image better than a rotation. The reason this matters is that magnetic field imaging is a non-contact route to seeing buried currents in semiconductor packaging, and current-density maps are what a failure analyst actually uses. If the pipeline holds up, it gives an end-to-end workflow for SQUID-based inspection without manual phase or alignment tuning.

What carries the argument

The engine of the model is the Fourier-domain Biot-Savart inversion. For a 2D current sheet at height $z$, the 2D Fourier transform of the measured vertical field $\tilde{B}_z(k_x,k_y,z)$ is related to the in-plane current components by $\tilde{J}_x = -(2/\mu_0)(k_y/k)\,e^{k z}\tilde{B}_z$ and $\tilde{J}_y = +(2/\mu_0)(k_x/k)\,e^{k z}\tilde{B}_z$, with $k=\sqrt{k_x^2+k_y^2}$; dividing by $k$ and multiplying by $e^{k z}$ recovers the lateral currents. The two preceding stages prepare the field image for this step: the phase rotation $I' = I\cos\phi + Q\sin\phi$ concentrates sharp wire signals in one channel, and the affine mapping $X_w=X$, $Y_w=Y+X\tan\theta$ removes skew before the inversion is applied with a hard cutoff filter.

What would settle it

Scan a known two-layer current test structure at two well-separated heights, run the SPIM inversion at the reported $z=120\,\mu$m, and compare the recovered currents with the known top-layer pattern: if increasing the bottom-layer current introduces growing artifacts, or if the residual between measured and reconstructed $B_z$ exceeds the 0.1 nT SQUID noise, then the single-plane assumption is falsified.

Watch

Extended reading notes

Core claim

The central claim is that the spatial-physics informed model (SPIM) converts raw SQUID magnetic-field scans of a 3D spiral into current-density images through three stages. Stage one rotates the lock-in phase between the in-phase (I) and quadrature (Q) channels so the sharp wire signal concentrates in the I' channel while the blurry eddy-current signal is pushed into the Q' channel; for the experimental scan a phase of 5.8° yields 0.3% more I-channel sharpness and 25% less Q-channel sharpness. Stage two applies an affine transformation that corrects a 0.3° skew in the raster scan, aligning the image better than a pure rotation. Stage three inverts the aligned vertical-field image using an FFT-based Biot-Savart inversion with a hard cutoff at $k_w = 3/z$ and $z = 120\,\mu$m, producing current density along x, along y, and overall. The authors demonstrate this conversion experimentally, not only in simulation.

Load-bearing premise

The inversion assumes all currents lie in a single flat plane 120 µm below the SQUID, so the spiral's second layer and its vertical wire segments are treated as invisible; if they actually contribute to the measured field, the recovered current-density maps are biased even when the alignment and phase steps work perfectly.

Editorial extensions

If this is right

  • At the found phase optimum, the I' image carries the sharp wire signals and the Q' image's $B_z$ scale is about 10 times smaller, so the alignment and inversion stages operate on a cleaner field.
  • A 0.3° skew correction is a better geometric model than a 0.3° rotation for this scan geometry, aligning the image along the y-direction with fewer residual features in the difference image.
  • With the cutoff $k_w = 3/z$ at $z = 120\,\mu$m, the FFT-Biot-Savart inversion localizes the spiral's current paths in both x and y components, so the output can be used to trace where the current actually flows.
  • Because vertical wire segments produce no z-component of the magnetic field, the recovered current-density images represent the lateral wires only; this is a physical property of the measurement, not a failure of the preprocessing.
  • The model is structured so the alignment and phase steps can be combined with any inversion kernel, opening the way to extensions that include vertical resolution and defect samples, as the authors note.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A decisive next test would be to scan a two-layer test structure with known currents in each layer and check whether the single-plane inversion at $z=120\,\mu$m recovers the top layer without artifacts from the bottom layer; if bottom-layer currents leak into the reconstruction, a multi-layer inversion would be needed.
  • The reported 0.3% I-channel sharpness gain is small enough that the practical benefit of SPIM probably lies in current-path localization rather than in sharpness itself; scoring the pipeline by recovered wire positions instead of total variation would be a stronger test.
  • The phase-rotation and affine-alignment stages are sensor-agnostic, so they could in principle be reused for GMR or quantum-diamond magnetic images, with only the inversion kernel replaced by the appropriate vector-field response.
  • The 0.3° skew parameter was chosen by visually comparing difference images; automating the affine search with a sharpness or alignment objective would make the correction quantitative and repeatable.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper proposes SPIM, a three-stage processing pipeline for SQUID microscopy images of a 3D spiral test sample. Stage A performs a lock-in phase rotation (Eq. 2) to concentrate sharp wire-field signals into an in-phase image I' and leave eddy currents in the quadrature image Q'. Stage B applies an affine transform to correct a claimed 0.3° rotational or skew misalignment. Stage C converts the processed magnetic field image into a current-density map using a Biot–Savart inversion in Fourier space with a hard cutoff kw (Eqs. 8–11). The authors report that SPIM improves I-channel sharpness by 0.3%, reduces Q-channel sharpness by 25%, removes a 0.3° misalignment, and produces feasible current-density maps at a cutoff of kw = 3/z with z = 120 µm.

Significance. If the claimed improvements were independently validated, the pipeline would be a useful practical workflow for SQUID-based current-path imaging in advanced packaging inspection. The paper has real experimental data, and the Fourier-domain Biot–Savart inversion in Eqs. (8)–(11) is correctly taken from the established literature, which is a strength. The use of a total-variation score to tune the lock-in phase and the use of affine transforms for scan-geometry correction are also sensible ideas. However, the reported gains are very small, the free parameters are tuned and evaluated on the same image, and the inversion model ignores the sample's two-layer structure. The significance is therefore conditional: the paper demonstrates a plausible workflow, but it does not yet provide evidence that the reconstructed current density is quantitatively reliable or that the preprocessing improvements are real rather than fitting artifacts.

major comments (4)
  1. [§III.B, Figs. 9–10, Eq. (2)] The phase value φ = 5.8° is selected by maximizing the TV score of the I' image on the same data that are then used to report the sharpness improvement. This makes the reported 0.3% I-channel sharpness gain a fitted result, not an independent measurement. Moreover, because Eq. (2) is a rotation, transferring sharp features into I' and out of Q' is exactly what the optimization is designed to do; the 25% Q-channel reduction is therefore partly definitional. To support the claimed improvement, the authors need to validate the phase choice on a held-out image, on a separate region, or with a known phase offset, and they need to report the improvement with error bars or noise propagation.
  2. [§III.C, Fig. 13, Eqs. (8)–(11)] The cutoff wavenumber kw = 3/z is chosen by visual inspection of the same current-density images that it produces (Fig. 13), with no quantitative selection criterion and no sensitivity analysis. Since the inversion factor exp(kz) in Eqs. (8)–(11) amplifies high-k noise exponentially, the 'feasible current density' claim depends critically on this ad hoc choice. The authors should define an objective rule for kw, evaluate the reconstruction against known current paths or a synthetic forward model, and report how the result changes over a range of kw values.
  3. [§III.A–C, Table I, Eqs. (6)–(11)] The inversion assumes a single 2D current sheet at a known height z = 120 µm, but the sample is explicitly a two-layer 3D spiral (Table I) with vertical connecting segments. The text states that wires in the z direction produce no Bz component, and the conclusion defers vertical resolution to future work, but this does not resolve the problem: for any current in the lower layer at depth z1 ≠ 120 µm, the measured Bz contains a factor exp(−k(z1 − 120 µm)), and the inversion multiplies by exp(k·120 µm), exponentially amplifying high-k components from that layer. Even if the phase and alignment stages are perfect, the recovered Jx and Jy are thus not quantitatively tied to the true 3D current distribution. The authors should either restrict the claims to top-layer lateral current paths, add a forward-model study quantifying the bias, or model both layers explicitly.
  4. [§III.B, Fig. 11] The 0.3° rotation and skew corrections are asserted but not measured or validated. No procedure is given for estimating θ from the image, and the evaluation is purely visual comparison of difference images (Figs. 11(c) and 11(e)). A quantitative alignment metric, such as residual line-edge misalignment or comparison with the known sample geometry, is needed to support the claim that misalignment was removed.
minor comments (5)
  1. [Abstract] The abstract states 'misalignments of 0.30 in a real image'; the units should be degrees (0.3°), not '0.30'.
  2. [Eq. (6)] Equation (6) is written with an approximate equality and unclear notation; the variables in the integrand and the limits of integration should be defined explicitly, or the equation should be replaced by the standard Biot–Savart expression.
  3. [Table I] The entry 'All lateral wires have with ℓ/z >> 1' contains a typo and the quantity ℓ is not defined.
  4. [§II.C] The text repeatedly refers to 'magnetic currents'; the quantity being reconstructed is electric current density J, and the terminology should be corrected for clarity.
  5. [§III.C, Fig. 12] The edge artifacts mentioned in the text are visible in Figs. 12(b) and 12(c), but no explanation is given for why they appear only on the right side; a brief comment on the edge-handling method would help the reader.

Circularity Check

2 steps flagged · score 6.0 of 10

The reported sharpness gains and the feasibility of the recovered current maps are in-sample optimization outputs, not independent predictions: the phase angle and FFT cutoff are tuned on the very images whose sharpness and current density are then quoted as results.

  1. fitted input called prediction [The phase-tuning and sharpness-quantification passage in Section III.B (Magnetic Image Analysis).]
    "After conducting a more detailed analysis of the phase values, we identify 5.8 as having the highest TV score. This selection ensures that the sharp features in the Iˈ image are preserved while minimizing sharp features in the Qˈ image... The adjusted phase, considering maximum and minimum field values for Iˈ and Qˈ images, shows an improvement in sharpness of 0.3% and a reduction in sharpness by 25%, respectively."

    The rotation angle is a free parameter fitted to maximize the total-variation score of the I' image, and the claimed sharpness improvement and Q reduction are then computed from that same image after applying the fitted angle. The direction of the result is enforced by the selection criterion (sharper I', less sharp Q'), and the percentages are in-sample statistics rather than predictions. No held-out image, independent phase measurement, or ground-truth sharpness is supplied, so the reported improvement is a description of the fit, not a tested outcome.

  2. fitted input called prediction [The FFT cutoff selection passage in Section III.B (Magnetic Image Analysis), Fig. 12/13.]
    "In the FFT method, two important tunable parameters, z and kw, significantly influence the resulting current distribution. Figure 13 shows the effects of varying the cut-off (kw) values at a fixed z value of 120 µm... Among these, the cut-off value of 3/z (as shown in Figures 12 and 13(b)) proves to be the most effective in accurately locating the current path compared to the other cut-off values."

    The cutoff kw=3/z is chosen by visually comparing the reconstructed current images and selecting the one that appears to locate the current path best, using the same current-density images that the paper presents as outputs. The later statement that this cutoff provides a feasible current density and direction is therefore a description of the tuned output rather than an independent validation. Since no ground-truth current map or separate test image is used, the claimed effectiveness of the inversion is partly a consequence of the parameter being selected from the displayed result.

full rationale

The Fourier-space Biot-Savart inversion in Eqs. (8)-(11) is a standard transform and is not derived from the data, and the affine alignment in Eqs. (3)-(5) is a conventional geometric correction; those parts are self-contained. The single-plane assumption at z=120 µm for a two-layer spiral, together with the statement that 'wires in the z direction produce no z-component of the magnetic field,' is a real modeling limitation for the recovered current density, but it is a correctness/scope concern rather than circularity; the conclusion's admission that vertical resolution is left to future work confirms the point. The circularity that is present lies in the headline quantitative claims: the 0.3% sharpness gain and 25% Q-channel reduction are produced by choosing the lock-in phase to maximize a sharpness-like objective on the same image, and the current-density result is obtained after selecting the FFT cutoff from the same output. These are in-sample fits presented as demonstrated improvements, which is the fitted-input-called-prediction pattern. Self-citations such as [3], [8], [10], [11], and [13] involve the authors' prior work, but the inversion mathematics is standard electrodynamics and is not the load-bearing source of circularity.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claim rests on several fitted parameters (phase, cutoff, alignment angle) and on modeling assumptions that simplify the 3D sample to a 2D current sheet. None of these are independently validated.

free parameters (3)
  • Lock-in phase phi = 5.8 degrees
    Selected to maximize the total variation score on the same I-channel image (Figure 9); the reported 0.3% sharpness gain is measured on the image after this fit.
  • FFT cutoff wavenumber kw = 3/z = 25 mm^-1 at z=120 um
    Chosen by visual inspection of Figures 13(a)-(e) as the setting that best locates current paths; no independent criterion.
  • Misalignment angle theta = 0.3 degrees
    The skew/rotation angle applied in the affine transform is stated as 0.3 degrees without showing how it was measured; it is selected by visual fit to the image (Figure 11).
assumptions (4)
  • standard math Biot-Savart law and its 2D Fourier transform for Bz from a 2D current distribution (Eqs. 6-11)
    Taken from prior work [3]; the paper relies on this inversion without re-deriving it.
  • domain assumption The sample currents can be represented as a single 2D current sheet at height z=120 um
    Introduced in Section III.A/Table I and Section II.C; ignores the two-layer 3D structure and vertical wire segments.
  • domain assumption A single constant phase rotation separates sharp wire signals from blurry eddy currents
    This is the basis of Eq. (2), even though the paper notes in Section II.A that the local phase of eddy currents can vary systematically with position.
  • domain assumption The SQUID measures only the z-component of the magnetic field at a point, with no sensor size correction
    Eq. (6) treats the field as Bz(x,y,z) from an ideal point sensor; the finite pickup loop area is only used to calibrate voltage-to-field conversion.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Spatial-Physics Informed Model for 3D Spiral Sample Scanned by SQUID Microscopy." pith.science (2026). https://pith.science/paper/SXB7IM5B

@misc{pith2026250711853,
  author       = {Pith},
  title        = {Pith review of: A Spatial-Physics Informed Model for 3D Spiral Sample Scanned by SQUID Microscopy},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SXB7IM5B}},
  note         = {Machine review of arXiv:2507.11853}
}
read the original abstract

The development of advanced packaging is essential in the semiconductor manufacturing industry. However, non-destructive testing (NDT) of advanced packaging becomes increasingly challenging due to the depth and complexity of the layers involved. In such a scenario, Magnetic field imaging (MFI) enables the imaging of magnetic fields generated by currents. For MFI to be effective in NDT, the magnetic fields must be converted into current density. This conversion has typically relied solely on a Fast Fourier Transform (FFT) for magnetic field inversion; however, the existing approach does not consider eddy current effects or image misalignment in the test setup. In this paper, we present a spatial-physics informed model (SPIM) designed for a 3D spiral sample scanned using Superconducting QUantum Interference Device (SQUID) microscopy. The SPIM encompasses three key components: i) magnetic image enhancement by aligning all the "sharp" wire field signals to mitigate the eddy current effect using both in-phase (I-channel) and quadrature-phase (Q-channel) images; (ii) magnetic image alignment that addresses skew effects caused by any misalignment of the scanning SQUID microscope relative to the wire segments; and (iii) an inversion method for converting magnetic fields to magnetic currents by integrating the Biot-Savart Law with FFT. The results show that the SPIM improves I-channel sharpness by 0.3% and reduces Q-channel sharpness by 25%. Also, we were able to remove rotational and skew misalignments of 0.30 in a real image. Overall, SPIM highlights the potential of combining spatial analysis with physics-driven models in practical applications.

Figures

Figures reproduced from arXiv: 2507.11853 by the authors.

Figure 1
Figure 1. An overview of the SPIM The signal recorded in a SQUID data file consists of the in-phase (VI, I-channel) and quadrature-phase (VQ, Q￾channel) output voltages (Vout) from a lock-in detector. This detector is connected to the output (Vf) of a SQUID Flux￾Locked Loop (FLL) measuring system, as shown in [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 3
Figure 3. SQUID scanned raw magnetic image with (a) r [PITH_FULL_IMAGE:figures/full_fig_p002_3.png] view at source ↗
Figure 4
Figure 4. 3D visualization of the connected wire path [PITH_FULL_IMAGE:figures/full_fig_p003_4.png] view at source ↗
Figures from the paper (4 more)
Figure 7
Figure 7. Figure 7: SQUID scanned raw magnetic image in Volts w [PITH_FULL_IMAGE:figures/full_fig_p004_7.png]
Figure 9
Figure 9. Figure 9: Total variation of Iˈ for different phase va [PITH_FULL_IMAGE:figures/full_fig_p004_9.png]
Figure 12
Figure 12. Figure 12: Magnetic current image, (a) current flow d [PITH_FULL_IMAGE:figures/full_fig_p005_12.png]
Figure 13
Figure 13. Figure 13: Analysis of cut-off (kw) values, (a) cut-off at kw=1/z, (b) cut-off at kw=2/z, (c) cut-off at kw=3/z, (d) cut-off at kw=4/z, (e) cut-off at kw=5/z. IV. CONCLUSIONS In this paper, we experimentally demonstrate the acquisition of magnetic field imaging using a SQUID mic…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

14 extracted references · 14 canonical work pages

  1. [1]

    Transfer learning-based artificial intelligence-integrated physical modelin g to enable failure analysis for 3 nanometer and smaller silicon-based CMOS transistors,

    J. Pan, J., K.L. Low, J. Ghosh, S. Jayavelu, M.M. F erdaus, S.Y. Lim, E. Zamburg, Y. Li, B. Tang, X. Wang, J.F. Leong, S. Ramasamy, T. Buonassisi, C.K. Tham, and A.V.Y. Thean , “Transfer learning-based artificial intelligence-integrated physical modelin g to enable failure analysis for 3 nanometer and smaller silicon-based CMOS transistors,” ACS Applied N...

    work page 2021

  2. [2]

    Defect Detection and Localization using 2D Slicing Method on 3D X- ray Microscopy

    A.P.P. Aung, Z. Lin, R.S. Pahwa, R.I. Made, and J. Senthilnath, “Defect Detection and Localization using 2D Slicing Method on 3D X- ray Microscopy", AI4X 2025

    work page 2025

  3. [3]

    Scanning SQUID microscopy of integrated cir cuits,

    S. Chatraphorn, E.F. Fleet, F.C. Wellstood, L.A. Kn auss, and T.M. Eiles, "Scanning SQUID microscopy of integrated cir cuits," Applied physics letters vol. 76 (16), pp. 2304-2306, 2000

    work page 2000

  4. [4]

    Magnetic field imaging for electrical f ault isolation,

    A. Orozco, “Magnetic field imaging for electrical f ault isolation,” Microelectron. Fail. Anal, 111-131, 2019

    work page 2019

  5. [5]

    Current imaging using sq uid and gmr sensors for failure analysis,

    Y. Kaneko, and A. Orozco, “Current imaging using sq uid and gmr sensors for failure analysis,” In Proc. 37th Annu. NANO Test. Symp., 2017

    work page 2017

  6. [6]

    Vector magnetic current imaging of an 8 nm process node chip and 3D current distributions using the quantum diamond microscope,

    S.M. Oliver, D.J. Martynowych, M.J. Turner, D.A. Ho pper, R.L. Walsworth, and E.V. Levine, “Vector magnetic current imaging of an 8 nm process node chip and 3D current distributions using the quantum diamond microscope,” In International Symposium for Testing and Failure Analysis (Vol. 84215, pp. 96-107). ASM International, 2021

    work page 2021

  7. [7]

    Integr ation of SQUID microscopy into FA flow,

    R. Dias, S. Lars, W. Zhiyong, and S. David, “Integr ation of SQUID microscopy into FA flow,” International Symposium f or Testing and Failure Analysis, vol. 30859, pp. 77-81, 2001

    work page 2001

  8. [8]

    Room-temperature magnetic microsco py using a high-T (c) SQUID,

    S. Chatraphorn, “Room-temperature magnetic microsco py using a high-T (c) SQUID,” University of Maryland, College Park, 2000

    work page 2000

Show all 14 references
  1. [9]

    The SQUID handbook: Applications of SQUIDs and SQUID systems,

    J. Clarke, and A.I. Braginski, “The SQUID handbook: Applications of SQUIDs and SQUID systems,” John Wiley & Sons, eds., 2006

    2006

  2. [10]

    Multiobjective discrete particle swarm optimizatio n for multisensor image alignment,

    J. Senthilnath, S.N. Omkar, V. Mani, and T. Karthik eyan, “Multiobjective discrete particle swarm optimizatio n for multisensor image alignment,” IEEE Geoscience and Remote Sensin g Letters, 10(5), pp.1095-1099, 2013

    2013

  3. [11]

    Accurate point matching based on multi-objective genetic algorithm for multi-sensor satellite imagery,

    J. Senthilnath, N.P. Kalro, and J.A. Benediktsson, “Accurate point matching based on multi-objective genetic algorithm for multi-sensor satellite imagery,” Applied Mathematics and Computa tion, 236, pp.546-564, 2014

    2014

  4. [12]

    A mesh-free algorithm for ROF model,

    M.A. Khan, W. Chen, A. Ullah, and Z. Fu, “A mesh-free algorithm for ROF model,” EURASIP Journal on Advances in Signal Processing, 1- 16, 2017

    2017

  5. [13]

    Metacognitive Decision- Making Framework for Multi-UAV Target Search Withou t Communication,

    J. Senthilnath, K. Harikumar, and S. Suresh, “Metacognitive Decision- Making Framework for Multi-UAV Target Search Withou t Communication,” IEEE Transactions on Systems, Man, and Cybernetics: Systems , 54(5), pp. 3195-3206, 2024

    2024

  6. [14]

    Advances in scanning SQUID microscopy for die-level and package- level fault isolation,

    L.A. Knauss, A. Orozco, A., S.I. Woods, and A.B. Ca wthorne, “Advances in scanning SQUID microscopy for die-level and package- level fault isolation,” Microelectronics Reliability, 43(9-11), pp.1657- 1662, 2003

    2003

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.