Optical Response in Spintronic Poisson Bolometers

TL;DR

The paper addresses the limitation of Gaussian noise in uncooled infrared detectors by exploring a probabilistic spintronic Poisson bolometer based on an MTJ. It experimentally characterizes the optical response under illumination using 405 nm CW, 808 nm CW modulated, and 405 nm ps lasers, and models the response with Poisson statistics tied to thermally activated switching described by the Néel-Arrhenius law, $τ = τ_0 \, exp(E_b/(k_B T))$ with $E_b = \frac{M V (H_k \mp H)}{2}$. Key results show Poissonian dark and bright counts with light-induced increases in mean count rate (e.g., ~17% at 808 nm and ~153% at 405 nm) and ns-scale diffusion-dominated response, complemented by pulsed-laser experiments that reveal hot-electron effects and non-probabilistic photocurrent. Overall, the work establishes spintronic Poisson bolometers as fast, room-temperature, probabilistic infrared detectors with potential for time-resolved imaging and sensing, addressing the speed and noise limitations of conventional VOx-based detectors.

Abstract

Analog bolometers based on temperature-dependent phase-transition materials such as vanadium oxide (VOx) and barium titanate (BTO) represent the state of the art in uncooled infrared detectors. Recently, the first room-temperature spintronic Poisson bolometer based on magnetic tunnel junctions (MTJs) was proposed and demonstrated as a promising infrared detector. Unlike conventional bolometers, the spintronic Poisson bolometer operates in a probabilistic regime dominated by Poissonian noise, where the response is governed by resistance fluctuations arising from thermally activated magnetization transitions. Spontaneous transitions between two metastable magnetic states occur even in the absence of incident light, and the transition probability increases under illumination. In this work, we experimentally study the statistical properties of the optical response of the spintronic Poisson bolometer under illumination. We demonstrate that transitions in spintronic Poisson bolometers, both in the absence and presence of light, exhibit Poissonian behavior, with transition rates and interarrival times modulated by incident radiation. Under illumination, we observe a 153% increase in the count rate accompanied by a 70% reduction in interarrival time. These results establish spintronic Poisson bolometers as a promising platform for probabilistic, high-speed, and high-sensitivity infrared detection at room temperature.

Figures (4)

• Figure 1: (a) In analog detectors, thermal noise and Johnson noise follow a Gaussian distribution as they stem from thermal agitation of a large number of electrons. (b) In digital detectors, dark counts follow a Poisson distribution as they stem from discrete independent events happening at a constant mean rate, such as thermal excitations and quantum tunneling effects in single photon detectors (SPDs), or magnetization flips governed by exponentially distributed interarrival times in spintronic Poisson bolometers (SPB). (c) Device structure. The main structure has five layers: a transduction layer, a magnetic free layer, a tunnel barrier, a magnetic fixed layer, and a synthetic antiferromagnetic layer (SAF). The magnetization M of the readout layer is pinned by the SAF. The magnetization of the sensing layer flips between M1 and M2 probabilistically due to heat. (d) When no light is incident, natural heat in the device causes discrete transitions between these two magnetization directions, which we refer to as dark counts. When light is incident, a hot spot is formed in the transduction layer, which thermalizes through a magnetic tunnel junction (MTJ). This increases the probability of transition in the MTJs sensing layer, leading to an increased count rate, which is read out through the device’s resistance. (e) Left: Optical image of the device. Right: SEM image of the MTJ nanopillar before Au deposition.
• Figure 2: (a)Schematic of the power-time relationship of 3 laser sources. We study the device response to 3 types of laser sources. The first laser is a continuous-wave (CW) 405 nm laser. The second laser is a modulated 808 nm CW laser. The power output of this laser is modulated by an arbitrary waveform generator (AWG). The third laser is a 405 nm pico-second laser that has a pulse width as small as 20 ps. (b) Raw readout signal. The orange curve is the laser trigger signal. (c) Time correlation. We divide each period into 100 bins (100 ns per bin) and calculate the probability of a count happening in each bin. The red dashed line marks the time of the trigger of one of the periods when it turns the laser on, which occurs at t = 3233.6 ns. The system has around 80 ns latency between the laser trigger and light incidence on the device. We can see a stabilized count probability increase at t = 3366.7 ns, which is around 53 ns after the light is incident when considering the latency. (d) Count statistics. We divide the data into 20 $\mu s$ time bins and calculate the number of counts happening in each time bin. For both laser-on and off regions, the number of counts follows a Poisson distribution. With the fitted mean number of counts $\lambda$, we can calculate the mean count rates for laser off and on regions, which are 3.373285 Mcps and 3.95123 Mcps, respectively. This shows around a 17% increase in the count rate due to incident light.
• Figure 3: (a) Probability histogram of detector counts over $3.8~\mu\text{s}$ intervals under 405 nm CW laser illumination. The count statistics follow a Poisson distribution, as shown by the black fit curve. When the laser is off, the mean count is $\lambda$= 17; under illumination, the mean increases to $\lambda$ = 43, indicating light-induced switching events. (b) Voltage state distributions under dark and illuminated conditions show two distinct peaks corresponding to the parallel (low-resistance) and antiparallel (high-resistance) magnetization states. (c) Histogram of relaxation times spent in the parallel (low-voltage) state before transitioning to the antiparallel (high-voltage) state. The distribution follows an exponential decay, consistent with Poissonian statistics. Illumination reduces the mean relaxation time by 66%, increasing the switching rate. Inset: Bright count rate versus laser intensity shows a significant increase in counts in response to higher optical power. (d) Histogram of relaxation times in the antiparallel state before switching to the parallel state. Like the parallel case, the distribution follows an exponential profile, with mean relaxation time decreasing by 70% under illumination. Inset: Field dependence of both dark and bright count rates shows a maximum near H=1 mT, corresponding to the field strength where the energy barrier between magnetic states is minimized and switching is most probable.
• Figure 4: (a) For a 405 nm picosecond laser, the laser-induced response is observed in the spintronic Poisson bolometer readout when light is incident on or near the device. (b) We observe different behavior with the 808 nm laser, where the photocurrent response does not reduce further away from the device. Inset is the reflection spectrum of the Au top contact. (c)The pulse width dependence is studied by varying the duty cycle of the 808 nm modulated laser. We observe that for larger pulse widths, the photocurrent is reduced, which signifies power dependence. (d) The peak power dependence of the laser-induced voltage is linear at low powers and saturates at high powers.