Spintronic Poisson Bolometers: Digital IR Sensors

• Spintronic Poisson Bolometers are solid-state IR detectors that encode temperature changes as discrete stochastic switching events in nanoscale MTJs.
• By integrating plasmonic nanoantenna absorbers with engineered MTJ stacks, these devices achieve sub-100 mK NETD and fast response times for high-speed imaging.
• Their CMOS-compatible digital output leverages Poisson statistics for robust noise control, enabling applications in robotics, environmental monitoring, and edge AI.

Spintronic Poisson bolometers are solid-state infrared detectors that exploit stochastic magnetization switching in nanoscale magnetic tunnel junctions (MTJs) to achieve event-based, digital readout of thermal signals. Unlike conventional analog bolometers which transduce temperature into continuous resistance or voltage shifts, these devices encode scene temperature as discrete stochastic flips, producing a Poissonian count stream whose mean rate varies with incident infrared radiation. By integrating spintronic transduction layers with engineered plasmonic nanoantenna absorbers, spintronic Poisson bolometers offer ultra-broadband sensitivity, sub-100 mK noise-equivalent temperature differences (NETD) at room temperature, and fast response times—all within a CMOS-compatible architecture (Mousa et al., 16 Jan 2026, Mousa et al., 26 Jan 2026, Yang et al., 13 Dec 2025, Singh et al., 7 Oct 2025, Yang et al., 16 Dec 2025, Yang et al., 13 Dec 2025).

1. Physical Architecture and Materials

The canonical device structure comprises a nanoscale MTJ stack capped by a plasmonically engineered IR absorber. The MTJ stack typically consists of the following layers, from substrate upwards: Ta (seed), CoFeB (pinned layer), Ru, CoFeB (synthetic antiferromagnet, SAF), MgO (tunnel barrier), and CoFeB free layer with tailored perpendicular magnetic anisotropy. The free-layer’s energy barrier, $E_b \sim 20–60\,k_BT$ at 300 K, is optimized for stochastic switching. Capping layers (Ta, Pt) serve for oxidation protection and facilitate spin–Hall coupling (Mousa et al., 16 Jan 2026).

Thermal sensitivity is enhanced via a plasmonic nanoantenna array (e.g., 40 nm Au nanodisks, diameter 300 nm, pitch 320 nm) deposited atop a Ge/Ti bilayer. These antennas induce localized surface plasmon resonances, boosting the absorption and field concentration in the $3–14\,\mu \text{m}$ spectral band, with COMSOL simulations and experimental measurement confirming absorptance between 60–80% across the operational range (Mousa et al., 16 Jan 2026, Yang et al., 13 Dec 2025).

Fabrication uses magnetron sputtering for the MTJ stack, e-beam lithography for nanopillar definition ($<0.1\,\mu\text{m}^2$ active area), ion milling, and lift-off patterning for the antennas. Array formation and integration are compatible with backend-of-line CMOS processing and can be scaled via step-and-repeat lithography (Singh et al., 7 Oct 2025).

2. Statistical Detection Paradigm

Spintronic Poisson bolometers operate fundamentally in a statistical regime. Device readout does not measure analog resistance changes directly; instead, it counts discrete, thermally-activated switching events of the MTJ free layer between two easy-axis magnetization states. Let $\lambda_0$ be the baseline Poisson event rate at equilibrium temperature $T_0 \approx 300\,\text{K}$. Upon IR absorption, the increase in local temperature $\Delta T$ leads to a higher switching rate $\lambda(T) = \lambda_0 + \Delta \lambda$.

Thermal switching follows Arrhenius–Néel kinetics: $\lambda(T) = f_0\, \exp\left(-E_b / (k_B T)\right)$ where $f_0$ is the attempt frequency ($10^{8}$–$3–14\,\mu \text{m}$0). The probability of registering $3–14\,\mu \text{m}$1 switching events in a time window $3–14\,\mu \text{m}$2 is strictly Poissonian: $3–14\,\mu \text{m}$3 Maximum-likelihood estimation yields the observed $3–14\,\mu \text{m}$4 as $3–14\,\mu \text{m}$5 and, for small $3–14\,\mu \text{m}$6, the temperature estimation: $3–14\,\mu \text{m}$7 (Mousa et al., 16 Jan 2026, Yang et al., 13 Dec 2025, Yang et al., 16 Dec 2025).

Interarrival times between transitions are exponentially distributed ($3–14\,\mu \text{m}$8), and illumination leads to a pronounced increase in event rate (up to 153% in representative measurements) and commensurate reduction in mean waiting times (Yang et al., 16 Dec 2025).

3. Noise, Sensitivity, and Bandwidth Analysis

Instead of suppressing thermal noise, the event-counting paradigm leverages it as the primary information carrier: both signal and noise originate from the same Poissonian process. The RMS count noise in integration time $3–14\,\mu \text{m}$9 is $<0.1\,\mu\text{m}^2$0, leading to shot-noise–limited performance.

The NETD (or NEDT for some works) derives from propagation of count variance to temperature uncertainty: $<0.1\,\mu\text{m}^2$1 with NETD typically expressed as $<0.1\,\mu\text{m}^2$2, with BW the measurement bandwidth. At $<0.1\,\mu\text{m}^2$3 K and $<0.1\,\mu\text{m}^2$4, SP-bolometers achieve NETD $<0.1\,\mu\text{m}^2$5; devices with optimized plasmonic absorbers and thermal engineering report best NEDT $<0.1\,\mu\text{m}^2$6 at 50 Hz (Mousa et al., 16 Jan 2026, Yang et al., 13 Dec 2025). The bandwidth is set by the faster of thermal ($<0.1\,\mu\text{m}^2$7) or magnetic switching timescales; 3dB bandwidths up to tens of MHz are realizable, and event-rate streams up to GHz are technically feasible (Singh et al., 7 Oct 2025).

Comparison to conventional technologies reveals significant advantages: uncooled VO$<0.1\,\mu\text{m}^2$8 microbolometers operate at NETD $<0.1\,\mu\text{m}^2$9 (300 K), while cooled InSb photodiodes ($\lambda_0$0 K) reach $\lambda_0$1, albeit requiring cryogenics. Spintronic Poisson bolometers offer competitive NETD at room temperature, event-based digital output, and immunity to $\lambda_0$2 and readout noise (Mousa et al., 16 Jan 2026, Yang et al., 16 Dec 2025).

4. Array Architecture, Scalability, and Enhancement Techniques

Multipixel arrays of spintronic Poisson bolometers are fabricated using row-column multiplexing, enabling scalable readout in focal-plane architectures (Singh et al., 7 Oct 2025). Proof-of-concept $\lambda_0$3 arrays demonstrate high-speed digital readout (up to $\lambda_0$4 counts/s) and sub-micron pixel pitches. The small active area of each pixel (<$\lambda_0$5) yields a low fill factor (typically $\lambda_0$6), which limits photon collection and overall SNR in imaging modes (Yang et al., 13 Dec 2025).

Microlens arrays (MLAs), such as plano-convex Al$\lambda_0$7O$\lambda_0$8 microlenses, are deployed to concentrate incident infrared flux onto the submicron MTJ pillar: finite-difference time-domain (FDTD) modeling and full-scene radiometric–stochastic simulation demonstrate up to 15$\lambda_0$9 increase in collection efficiency and a 3–4$T_0 \approx 300\,\text{K}$0 reduction in NEDT (from 30 mK to $T_0 \approx 300\,\text{K}$110 mK) for MWIR imaging when properly matched to pixel pitch and absorber size (Yang et al., 13 Dec 2025).

Detector Type	Pixel Area ($T_0 \approx 300\,\text{K}$2)	Fill Factor (%)
SNSPD (Oripov 2023)	$T_0 \approx 300\,\text{K}$3	4.8
SNSPD (Wollman 2019)	$T_0 \approx 300\,\text{K}$4	36
SPB (Leif 2025)	$T_0 \approx 300\,\text{K}$5	8.2

5. Experimental Performance and Application Domains

Spintronic Poisson bolometers reproducibly demonstrate event-rate scaling with temperature, Poissonian statistics confirmed via count histograms and interarrival time analysis, and high agreement between observed and theoretical switching distributions (Yang et al., 16 Dec 2025, Mousa et al., 16 Jan 2026). NETD values between 80–100 mK (broadband, $T_0 \approx 300\,\text{K}$6) and best-in-class room-temperature NEDT $T_0 \approx 300\,\text{K}$7 at 50 Hz have been reported (Yang et al., 13 Dec 2025).

Applications include autonomous vehicle and robotics thermal imaging, spectroscopic gas sensing, environmental monitoring, biomedical IR thermometry, heat-assisted detection and ranging (HADAR), and edge computing for event-driven IR vision. Real-time streaming of Poisson event data enables integration with neuromorphic and spiking neural network processors for low-latency inference (Mousa et al., 16 Jan 2026, Mousa et al., 26 Jan 2026).

6. Modeling and Estimation Theory

Temperature estimation is carried out via either maximum likelihood (MLE) or Bayesian frameworks, directly from the Poisson count data. For observed counts $T_0 \approx 300\,\text{K}$8 in interval $T_0 \approx 300\,\text{K}$9, the likelihood function $\Delta T$0. The MLE solution yields $\Delta T$1, with temperature increment estimated as: $\Delta T$2 Bayesian approaches optionally incorporate prior distributions on $\Delta T$3 to encode environmental or scene statistics (Mousa et al., 16 Jan 2026).

Realtime thermal imaging streams are processed via causal Kalman or particle filters to reconstruct temperature sequences, providing robust video imaging even under background-limited or noisy illumination. In array mode, cross-correlation and event coincidence analysis enables multiplexed, temporally resolved, and sparsity-tuned acquisition (Singh et al., 7 Oct 2025).

7. Comparative Advantages, Limitations, and Future Directions

Spintronic Poisson bolometers bypass the limitations of analog thermal detectors, including Johnson and $\Delta T$4 noise, frame-rate bottlenecks, and complex micromachining requirements. Their digital output is directly compatible with standard CMOS logic, and their noise statistics are fundamentally shot-noise limited.

Key advantages:

• Ultra-broadband spectral coverage ($\Delta T$5).
• Room temperature operation, no cryocooling required.
• Pixel dimension $\Delta T$6, enabling megapixel dense arrays.
• Power consumption per pixel $\Delta T$7, kHz-class timing resolution (Mousa et al., 26 Jan 2026).

Limitations include active area constraints, dynamic range saturation at high flux, and need for plasmonic or metasurface-enhanced absorption for optimal SNR. Pathways for improvement focus on materials engineering (alternate ferromagnets, energy barrier tuning), array integration, bandwidth extension, MLA optimization, and direct on-chip event processing for edge AI (Mousa et al., 16 Jan 2026, Yang et al., 13 Dec 2025).

A plausible implication is the emergence of real-time, event-driven IR vision systems with sub-millisecond latency and enhanced sensitivity for environmental, industrial, and biomedical applications.

• "Ultra-broadband Mid to Long-wave Infrared Spintronic Poisson Bolometer" (Mousa et al., 16 Jan 2026)
• "Uncooled Poisson Bolometer for High-Speed Event-Based Long-wave Thermal Imaging" (Mousa et al., 26 Jan 2026)
• "Long-Wave Infrared Spintronic Poisson Bolometers with High Sensitivity" (Yang et al., 13 Dec 2025)
• "Long wave infrared detection using probabilistic spintronic bolometer arrays" (Singh et al., 7 Oct 2025)
• "Optical Response in Spintronic Poisson Bolometers" (Yang et al., 16 Dec 2025)
• "Design of Microlens Arrays for Thermal Imaging with Spintronic Poisson Bolometers" (Yang et al., 13 Dec 2025)