A new implantable magnetic probe moves microscale neural magnetophysiology forward on sensitivity and directionality. Published on 4 September 2026 in Microsystems & Nanoengineering, Yi Wang and colleagues report a differential tunnel-magnetoresistance, or TMR, magnetrode with a detection limit of 68 pT/√Hz at 1 kHz, compared with 107 pT/√Hz for the same device operated single-ended. In an unshielded laboratory, the 50 Hz power-line peak falls from 18.961 to 0.433 nT/√Hz, roughly 98% common-mode suppression. Implanted in rat hippocampal CA1, the probe records spike-like magnetic transients associated with action potentials and polarity reversals consistent with different projections of local neural-current directions onto the sensor axis.
What changed
| Quantity | Previous/reference state | New result |
|---|---|---|
| TMR LOD at 1 kHz, 2024 design | about 300 pT/√Hz | 68 pT/√Hz |
| Same 2026 device, single-ended | 107 pT/√Hz | 68 pT/√Hz differential |
| 50 Hz interference | 18.961 nT/√Hz | 0.433 nT/√Hz |
| Differential baseline | 28 μm and 2.5 mm alternatives | 5 mm |
The baseline matters. In-vivo magnetophysiology was not invented by this paper. Earlier GMR magnetrodes recorded neural magnetic activity, and a 2025 Journal of Neurophysiology paper reported magnetic signatures of single-neuron action potentials in rats. TMR magnetrodes had reached about 300 pT/√Hz at 1 kHz in 2024 and about 140 pT/√Hz in a 2026 in-vivo LFP implementation. The correct claim is therefore a substantial TMR-system advance, not the first-ever neural magnetic recording.
Why magnetic recording is different
Electrical probes measure voltage differences created by ionic currents. The same currents generate magnetic fields. A magnetic measurement carries a vector sign relative to the sensor's sensitive axis, so reversing a current component can reverse the measured field. This gives magnetophysiology a potential source of directional information that a local voltage trace does not encode in the same way.
Engineering architecture
The MEMS probe uses a silicon shank thinned to 200 μm. Each sensing unit contains 12 magnetic tunnel junctions, each about 15 μm × 45 μm, connected in series. The sensing and reference units show TMR ratios of 121% and 119%, with linear sensitivities of 12.30%/mT and 12.96%/mT. One unit sits near the tip and the reference unit farther up the shank. Environmental magnetic noise is approximately common to both channels; the localized neural field decays rapidly with distance. Subtracting the channels rejects common-mode interference while retaining more of the local field. Fabrication differences are compensated with separate Wheatstone bridges and adjustable instrumentation-amplifier gains.
A short reference spacing fails because the reference still sees too much of the target near field. At 28 μm, differential operation produced negative SNR gain, down to −8.68 dB. At 5 mm, spatial decoupling gave the best result, with up to 4.19 dB SNR improvement. The 50 Hz suppression is eta50 = (1 - A_differential/A_single-ended) × 100%. With the measured values, eta50 = (1 - 0.433/18.961) × 100% ≈ 97.7%.
What was recorded in CA1
The device was implanted in hippocampal CA1 in male Sprague-Dawley rats. Signals were calibrated into magnetic-field units, band-pass filtered between 300 and 3000 Hz, and candidate spikes detected at three times the RMS background noise. Waveform clustering revealed opposite-polarity classes. Because the magnetic sensor is directional, opposite projections of local neural currents onto the sensor axis can produce opposite signs. The authors state a crucial limitation: the 12-MTJ unit has a finite footprint and can integrate flux from locally synchronous multi-unit activity. The supported claim is therefore action-potential-related local magnetic recording, not guaranteed isolation of one neuron per magnetic spike.
Biocompatibility and limits
The TMR stack uses a SiO2/Si3N4 dual passivation layer, 200 nm of each material, to isolate metallic layers and protect the device. Cell assays and seven-day histology were encouraging but do not establish chronic stability. The silicon shank also remains much stiffer than neural tissue. Chronic foreign-body response, micromotion, packaging, channel scaling, long-term drift and independent replication remain open.
New state of the art
The real delta is the combination of 68 pT/√Hz at 1 kHz, strong common-mode rejection, in-vivo CA1 recording and polarity-based directional information. It does not yet show chronic implantation, human recording, a large multichannel array, guaranteed single-neuron isolation for each magnetic spike, or independent replication of the exact device. The next discriminating experiments are simultaneous high-density electrical and magnetic recordings with known source geometry, chronic stability measurements, independent replication, and arrays able to exploit vector information across space.
Evidence status
Peer-reviewed · in-vivo animal experiment · data and supplementary material available through the paper · independent replication of this exact differential TMR implementation not yet identified.