REVIEW 4 major objections 6 minor 6 references
Spin Vector Control for Heisenberg-Inspired Probabilistic Computing
T0 review · 4 major / 6 minor · reviewed 2026-08-16 · deepseek-v4-flash
Pith's one-line read Two tilted ferromagnetic injectors on graphene add spin vectors in real space, with the summed vector's angle set by the injection current ratio.
desk verdict The device work is real and the tunable spin-projection result is solid, but the paper's centerpiece claim of verified real-space vector summation is underdetermined by a single-axis detector. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The central object is the tilted dual-injector non-local spin valve: two permalloy electrodes angled at 45° and 135° (or 15° and 165°) inject spin currents into a monolayer graphene channel, and a third electrode detects the non-local voltage. The argument runs through the vector-sum identity for the resultant spin direction, $\phi=\tan^{-1}\big((P_1 I_1 \sin\theta_1 + P_2 I_2 \sin\theta_2)/(P_1 I_1 \cos\theta_1 + P_2 I_2 \cos\theta_2)\big)$, together with the assumption that the detector signal is the projection of the summed spin accumulation onto its magnetization axis. This identity converts a current-ratio sweep into a continuous rotation of $\phi$, which is what lets the device tune magnitude and direction electrically. HSPICE spin-circuit simulations with 4×4 conductance matrices for charge and three spin components serve as the quantitative check on the analytical formula.
What would settle it
Measure the actual interface spin polarizations of the Py/Al2O3/graphene contacts on the same devices and insert them into the vector-sum formula; if the predicted $\Delta R_{NL}(\phi)$ curve shifts so far that the measured six-state step heights and signs no longer match, or if the MFM-reversal fields of the two tilted injectors do not line up with the resistance transitions, the linear vector-addition story is not the explanation.
Extended reading notes
Core claim
The central claim is that a non-local spin valve with two tilted ferromagnetic electrodes performs real-space vector addition of spin accumulations in monolayer graphene. With injector magnetizations at $\theta_1=45^\circ$ and $\theta_2=135^\circ$ and injection currents $I_1$ and $I_2$, the measured non-local resistance is argued to be $\Delta R_{NL}\approx R_\square P_{\rm det} (P_1 I_1 \cos\theta_1 + P_2 I_2 \cos\theta_2)/(I_1+I_2)$, where $P$ denotes interface spin polarization. The experimental signature is a six-state switching pattern whose step heights, signs, and order change with the current ratio; the extracted $\Delta R_{NL}$ as a function of the resultant angle $\phi$ follows the analytical vector-sum curve and HSPICE spin-circuit simulations, with deviations only in the region where the summed vector is nearly orthogonal to the detector. A second geometry with injectors at 15° and 165° shows the same sign reversal and linear dependence, presented as evidence that the vector-sum mechanism generalizes.
Load-bearing premise
The vector-addition interpretation assumes the two spin accumulations superpose linearly in the graphene channel and that the detector signal is exactly the projection of the summed spin vector onto its magnetization, with every interface assigned the same spin polarization of 0.5 rather than a value measured on these devices.
Editorial extensions
If this is right
- A spin accumulation vector with continuously adjustable angle can be produced in one graphene channel without changing magnet geometry, only by changing the ratio $I_1/I_2$.
- The six resistance states observed under balanced dual injection provide a direct fingerprint of the sequential magnetization reversal of the two tilted injectors, with the same sequence seen in magnetic force microscopy.
- The 15°/165° injector geometry reproduces the sign reversal and the linear relation between non-local resistance and injection current, indicating the vector-sum rule is not an artifact of the 90° symmetric tilt.
- Because the vector state is set electrically rather than by lithography, the same device is a reconfigurable analog spin-vector node, the ingredient needed for hardware mapping of Heisenberg-type energy functions.
Reading between the lines
- Inference: If the linear-superposition picture holds, reversing the sign of one injection current should produce vector subtraction, so the same two-injector pair could act as a compact analog vector arithmetic unit; the paper does not report such an experiment.
- Inference: The signal-suppression plateau near $\phi=90^\circ$ is a built-in orthogonality detector; a future experiment could exploit that plateau to measure the relative spin polarizations of the two injectors without additional magnetometry.
- Inference: Replacing the static permalloy injectors with low-barrier stochastic nanomagnets would turn this demonstration into a fluctuating vector source, and the time-averaged non-local signal would then encode the equilibrium orientation distribution of a Heisenberg-type spin; that step is the natural path to probabilistic computing but is not tested here.
- Inference: Because the plateau near 90° makes the detector nearly blind in that angular window, vector-based probabilistic nodes will likely need a second detection axis or a symmetry-breaking bias to read states near orthogonal alignment; the paper leaves that design constraint open.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The manuscript reports non-local spin valve measurements in monolayer graphene with two ferromagnetic injectors patterned at 45° and 135° relative to the detector electrode (Py4). By varying the two injection currents independently, the authors observe non-local resistance levels whose sign and magnitude change monotonically with the injection ratio, and they interpret these observations as real-space vector summation of two spin accumulations with the detector reading the projection of the summed vector onto its easy axis. Supporting evidence includes single-injection control measurements, MFM imaging of sequential magnetization switching, an analytical spin-circuit expression, and HSPICE simulations with tabulated parameters. The paper positions the device as a building block for Heisenberg-inspired probabilistic computing based on vector spin states.
Significance. If the vector-summation claim were conclusively established, the work would be a useful step toward continuous angular control of spin accumulation in a two-dimensional material, extending earlier scalar spin-majority gates and the tilted dual-injector idea from metallic channels to graphene. The paper is clearly written, the device architecture is transparent, and the combination of an analytical expression, spin-circuit simulations, and tabulated parameters is a strength. However, the central claim that the measurement demonstrates true vector addition rather than scalar superposition is not supported by the single-axis detection scheme used here. The significance is therefore contingent either on additional evidence of the transverse spin component or on a more modest interpretation of the results.
major comments (4)
- [Theoretical Vector Model and Angular Dependence (p. 11–13), Eqs. (1)–(2); Fig. 4b] The central claim of "true vector addition" is underdetermined because all measurements use a single detector (Py4) whose easy axis lies along x. Equation (1) reduces to ΔRNL ∝ P_det (P1 I1 cosθ1 + P2 I2 cosθ2)/(I1+I2), which is algebraically identical to independent scalar superposition of two spin signals projected onto the detector axis. Since no second projection (e.g., a y-axis detector or a Hanle measurement of the transverse component) is reported, every data point in Figs. 4b and 5b is consistent with two uncoupled scalar spin injections. The sentence in the Introduction (p. 3) distinguishing "real-space addition of two spin vectors, not merely scalar projections" is therefore not supported by the presented evidence.
- [Theoretical Vector Model and Angular Dependence (p. 12–13), Fig. 5b] The horizontal axis angle φ is not an observable; it is computed from Eq. (2) using the assumed geometry θ1=45°, θ2=135° and the assumed polarizations P1=P2=Pdet=0.5. The analytical line and the HSPICE simulation implement the same vector-sum model. Consequently, the agreement between the data squares and the line in Figs. 4b and 5b is a consistency check rather than an independent validation of the vector model. The authors should either provide an independent measurement of φ (for example, from Hanle spin precession) or reframe the horizontal axis in terms of directly measured quantities such as the current ratio.
- [Fig. 5b caption and Supplementary Fig. S3] The most discriminating prediction of the vector model is the near-null signal for φ≈90°, yet this is precisely the region where the grey points in Fig. 5b are excluded post hoc, and the main text notes that both ΔRNL and its modulation are intrinsically small there. In addition, Supplementary Fig. S3 reports only qualitative agreement and states that absolute values do not match quantitatively. These exclusions and mismatches should be presented transparently in the main text, and the sensitivity of the claimed "excellent agreement" to the excluded points and to the assumed polarizations should be quantified.
- [Table S1 and Supplementary section 4] The quantitative agreement relies on assumed values of the interface spin polarizations (P1=P2=Pdet=0.5) and transport parameters (G_c, λ_s, a, b) taken from Table S1 rather than measured on these devices. The paper states that the same polarizations are assumed. A sensitivity analysis over a plausible range of P and λ_s is needed to show that the sign reversal and the shape of ΔRNL(φ) are robust, since the claimed "excellent agreement" depends on these free parameters.
minor comments (6)
- [p. 13] The phrase "following a cos-1 dependence" should read "a cos φ dependence"; the x-projection of the spin vector is cos φ, not arccos φ.
- [Methods (p. 16)] The Methods text describes "three Permalloy (Py) ferromagnetic electrodes" but the device layouts in Figs. 1a and 2a contain more electrodes; the count and labeling (Py1–Py4, Py3 detector vs. Py4 detector) should be reconciled.
- [Figs. 4b and 5b] No error bars or replicate counts are shown. At minimum, the number of devices and repeated sweeps should be reported, particularly for the small-signal region near φ≈90° where measurement uncertainty is acknowledged.
- [Author contributions (p. 25)] The abbreviation "S.D." in the author contributions does not correspond to a listed author; please identify the contributor or correct the initials.
- [Introduction (p. 3)] The phrase "unique geometrically engineered injector" overstates novelty given that ref. 17 used a similar tilted dual-injector geometry in a Cu channel; the novelty here is the graphene implementation and the electrical control, which could be stated more precisely.
- [Conclusion (p. 15)] The Heisenberg-machine framing is forward-looking rather than demonstrated; the conclusion should clarify that the experiments show static spin transport, not stochastic p-bit operation.
Circularity Check
The manuscript asserts that spin-vector angles/magnitudes are 'extracted from the labeled ∆R_NL values' and then used as the basis for the model comparison, making the angular agreement in Figs. 4b/5b partially circular.
-
self definitional
[End of 'Spin Vector Control Using Dual Injection' section, immediately before 'Theoretical Vector Model and Angular Dependence'; Fig. 4b caption]
"The angular dependence and magnitude of these spin vectors, extracted from the labeled ∆R_NL values, provide the foundation for further analysis in the next section."
This sentence asserts that the spin-vector angle and magnitude feeding the theoretical analysis are obtained from the same measured ∆R_NL data that the model is then used to 'predict.' If the φ-axis in Figs. 4b/5b is produced by inverting the measured non-local resistance, then plotting ∆R_NL against φ and showing agreement with the analytical vector-sum line is tautological: the x-axis is a function of the y-axis. The next section's equations define φ from current ratios and assumed angles, which would be non-circular, but the explicit 'extracted from the labeled ∆R_NL values' statement makes the data-extraction route load-bearing; with only one detector projection, the measurement cannot distinguish true vector addition from scalar superposition of the two injector signals.
full rationale
The central derivation is mostly self-contained: the analytical ΔR_NL expression is derived from the 4×4 spin-circuit formalism with stated assumptions, and P = 0.5 is an assumed value rather than a parameter fitted to the reported dual-injection curves. The spin-circuit self-citations (refs. 16, 28, 29, 33) are methodological and carry no uniqueness claim, so they do not by themselves raise the circularity score. The main circularity concern is the manuscript's own statement that the spin-vector angles and magnitudes are 'extracted from the labeled ∆R_NL values' and then used as the foundation for the model comparison. If that extraction produced the φ-axis, the agreement with the analytical line is a self-consistency check, not an independent test. The alternative equations define φ from current ratios and assumed fixed angles, which would make the comparison a genuine quantitative prediction, and the six-state switching, single-injection limits, and MFM imaging provide independent qualitative support. Because the paper contains both routes, the circularity is real but partial; the central vector-summation claim retains independent experimental content, but the angular tunability demonstration is partly self-definitional.
Assumptions & free parameters
free parameters (4)
- Interface spin polarizations P1, P2, Pdet =
0.5 each (Table S1)
- NM spin diffusion length lambda_s =
500 nm
- Interface conductance G_c =
0.001 S
- Spin-mixing conductance parameters a, b =
a = 0.001, b = 0
assumptions (5)
- domain assumption Spin accumulation from two injectors superposes linearly in graphene (vector addition).
- domain assumption The non-local detector voltage measures only the projection of the spin accumulation along the detector magnetization axis.
- domain assumption G_c << 1 and lambda_s / L = 1 for the analytical RNL expression.
- standard math Standard magnetoelectronic circuit theory (Bauer-Brataas-Kelly) and the modular spin-circuit formalism.
- domain assumption Magnetization states of Py electrodes during field sweeps follow the sequence inferred from MFM and coercive fields.
Cite this review
Pith. "Pith review of Spin Vector Control for Heisenberg-Inspired Probabilistic Computing." pith.science (2026). https://pith.science/paper/V5GFDPOA
@misc{pith2026260812568,
author = {Pith},
title = {Pith review of: Spin Vector Control for Heisenberg-Inspired Probabilistic Computing},
year = {2026},
howpublished = {\url{https://pith.science/paper/V5GFDPOA}},
note = {Machine review of arXiv:2608.12568}
}
read the original abstract
Probabilistic bits (p-bits) have emerged as a cornerstone of probabilistic computing, enabling energy-efficient hardware implementation for probabilistic inference and combinatorial optimization. A critical challenge in advancing this field beyond binary p-bits lies in realizing and manipulating vector spin information, essential for mapping complex energy-based models such as the Heisenberg Hamiltonian.Here, we demonstrate a spintronic platform capable of real-space vector summation by using dual ferromagnetic spin injections into a monolayer graphene channel. By electrically tuning the spin polarization through independently controlled injection currents, we achieve continuous control over the magnitude and direction of the resulting spin accumulation vector. Experimental observations, supported by theoretical vector summation models and spin-circuit simulations, reveal coherent vector interactions and angular tunability of the spin state. This approach enables direct implementation of vector-based spin logic and lays the groundwork for mapping classical Heisenberg models using stochastic low-barrier magnets. Our results establish a scalable pathway for realizing probabilistic spin circuits based on two-dimensional materials, offering new opportunities for low-power, non-Boolean computing architectures.
Reference graph
Works this paper leans on
-
[1]
Reciprocal spin transport characteristics of non-local lateral graphene spin valves Fig. S1. Reciprocal readout with opposite spin injection polarity. a, Three repeated non-local spin valve measurements with spin injection through FM electrode 3 and detection at electrode 2, as schematically illustrated in Fig. 1a; b, Three repeated non-local spin valve m...
-
[2]
The spatial dependence of the electrochemical potential μ for spin-up and spin-down electrons in the graphene non-local spin valve configuration Fig. S2. Spatial evolution of electrochemical potentials in the graphene non-local spin valve. a, Schematic of a monolayer graphene lateral NLSV device. b-f, The spatial profiles of the electrochemical potentials...
-
[3]
Spin transport of dual spin injection between 45°/135° through non-local lateral graphene spin valve Fig. S3. Non-colinear spin vector readout by graphene non-local spin valve. a, Non-local spin valve signals obtained by varying IPy1 while keeping IPy2 = 1 µA. b, Corresponding ∆𝑅!" (squares) as a function of the angle ∅, with modulation achieved by changi...
-
[4]
HSPICE Simulation Fig. S4. Simulated non-local spin valve using spin-circuit. The device transport is captured by two modules: the Non-Magnet (NM) and the Ferromagnet and non-magnet interface (FM|NM). The NM module represents the graphene channels while, the FM|NM modules represent the interface between graphene and the Permalloy (Py) ferromagnet. All cir...
-
[5]
, as shown in Fig. S6d. Fig. S6f shows the linear relationship between 𝑅!
Spin transport of dual spin injection between 15°/165° through non-local lateral graphene spin Fig. S6. Spin vector readout by graphene non-local spin valve and theoretical modeling. a, Schematic of lateral non-local spin valve device on monolayer graphene spin valve, with Py ferromagnetic electrodes and Al2O3 tunneling barrier; b, Non-local spin signal w...
work page 2006
-
[11]
Aadit, N. A. et al. Massively parallel probabilistic computing with sparse Ising machines. Nat. Electron. 5, 460–468 (2022). 12. Daniel, J. et al. Experimental demonstration of an on-chip p-bit core based on stochastic magnetic tunnel junctions and 2D MoS2 transistors. Nat. Commun. 15, 4098 (2024). 13. Bunaiyan, S., Datta, S. & Camsari, K. Y. Heisenberg m...
Reviewed August 16, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.