We are reporting a new type of compact magneto-optic sensor constructed from terahertz-wave spintronic emitter and electro-optic detector. the corresponding terahertz polarization output of the emitter and the detection phase-sensitivity of the detector depend on the vector of the external magnetic field. The emitter/detector pair consists of two small and thin wafers sandwiched together and capped with a thin gold mirror. As a result, the use of bulky terahertz steering/collection optics was completely eliminated in our magneto-optic imager. With such simple on-chip generation/detection scheme for terahertz time-domain setup in reflection-type geometry, we were able to record the rasterscanned image contrast of a permanent magnet in the proximity of the sensor surface. the contrast strongly varies with the magnet orientation and its position with respect to the sensor. the imager spatial resolution depends on chip optical quality for tight femtosecond-laser pump/probe crossfocusing at detector/mirror interface and terahertz generation/detection efficiency. In this respect, the chip robustness to the pump/probe fluences is also an important factor to consider.

Recent studies in spintronics have highlighted ultrathin magnetic metallic multilayers as a novel and very promising class of broadband terahertz radiation sources. Such spintronic multilayers consist of ferromagnetic (FM) and non-magnetic (NM) thin films. When triggered by ultrafast laser pulses, they generate pulsed THz radiation due to the inverse spin-Hall effecta mechanism that converts optically driven spin currents from the magnetized FM layer into transient transverse charge currents in the NM layer, resulting in THz emission. As THz emitters, FM/NM multilayers have been intensively investigated so far only at 800-nm excitation wavelength using femtosecond Ti:sapphire lasers. In this work, we demonstrate that an optimized spintronic bilayer structure of 2-nm Fe and 3-nm Pt grown on 500 μm MgO substrate is just as effective as a THz radiation source when excited either at λ = 800 nm or at λ = 1550 nm by ultrafast laser pulses from a fs fiber laser (pulse width ~100 fs, repetition rate ~100 MHz). Even with low incident power levels, the Fe/Pt spintronic emitter exhibits efficient generation of THz radiation at both excitation wavelengths. The efficient THz emitter operation at 1550 nm facilitates the integration of such spintronic emitters in THz systems driven by relatively low cost and compact fs fiber lasers without the need for frequency conversion.

We report on efficient terahertz (THz) wave generation of Fe/Pt diabolo-shaped spintronic antennas with different Pt thicknesses fabricated on MgO substrates. Compared with the antenna-free spintronic bilayer, ∼45% and ∼98% emission amplitude improvements were obtained when using the antennas with thin and thick Pt, respectively, as THz radiation sources. The improvement can be attributed to the enhanced outcoupling of THz radiation to free space and to the enhanced THz emission with the deposition of thicker Pt layer at the displacement current direction. Our results suggest that efficient spintronic radiation sources can be obtained with proper design of these THz emitters.

No abstract

Several machine learning (ML) techniques were tested for the feasibility of performing automated pattern and waveform recognitions of terahertz time-domain spectroscopy datasets. Out of all the ML techniques under test, it was observed that random forest statistical algorithm works well with the THz datasets in both the frequency and time domains. With such ML algorithm, a classifier can be created with less than 1% out-of-bag error for segmentation of rusted and non-rusted sample regions of the image datasets in frequency domain. The degree of linear correlation between the rusted area percentage and the image spatial resolution with terahertz frequency can be used as an additional cross-validation criteria for the evaluation of classifier quality. However, for different rust staging measured datasets, a standardized procedure of image pre-processing is necessary to create/apply a single classifier and its usage is only limited to 1 ± 0.2 THz. Moreover, random forest is practically the best choice among the several popular ML techniques under test for waveform recognition of time-domain data in terms of classification accuracy and timing. Our results demonstrate the usefulness of random forest and several other machine learning algorithms for terahertz hyperspectral pattern recognition.