Spatial light modulator (SLM) has demonstrated the capability of manipulating the amplitude, phase, or polarisation state of electromagnetic wave. At terahertz (THz) frequencies, SLM has been used to address the technical challenges of THz applications, including beamforming 1, compressive imaging 2, and digital holography 3. However, the bottleneck for THz SLM development is the lack of suitable natural materials and electronic components. In recent years, metamaterials offer us a powerful tool to manipulate electromagnetic waves in desirable ways 4-6. The advances of digital and programmable metamaterials contribute to the rapid development of THz SLM7, 8. A variety of metamaterial-based THz SLM has been demonstrated and used for information procession. The reconfigurability of THz wave modulation in these designs relied on tuneable materials such as semiconductors 9, 10, liquid crystals 11, graphene 12 and phase-change materials 13 or micro-mirror arrays 14-16 and microfluidics 17.
Towards practical applications, new challenges have emerged. The large-scale integrated multi-level SLMs are highly demanding to realize highly directional multi-beam steering, which is essential for the next generation of wireless communication. They also can improve the image quality in compressive imaging. Typically, the THz SLMs are volatile, where the switched state is lost when the external stimuli are removed thus persistent bias is required for each pixel. It raises difficulties in scaling up SLM because of the exponential increase in the number of control lines and the complexity of the feed network. To resolve this difficulty, we have to look for new solutions. The SLM with the memory function can preserve the transformed state in a non-volatile manner. Therefore, it can allow each pixel to be controlled serially, enabling us to simplify the control circuit and expand the number of pixels. Moreover, the SLM with memory function is highly expected for a reconfigurable holographic plate with high-fidelity. Therefore, the programmable metamaterial with memory is a promising alternative for THz SLM.
The metamaterials having memory behaviours in response to external stimuli have been realized using phase change materials (PCM). It relies on the hysteretic behaviors of PCM to external stimuli 18-20. The memory metamaterials have exhibited many unusual electromagnetic behaviours such as dynamic resonance tuning 19, 20, reconfigurable logic-gate operation 21, 22, and greyscale data storage 23, 24. However, Programmable and pixelated memory metamaterials have not yet been realized. The main challenge for PCM based pixelated metamaterials is that the strong stimuli and thermal isolation are required for the independent control of each pixel,which induces unacceptable crosstalk between pixels.
Here we present a type of programmable memory metamaterials composed of 8×8 pixels at THz frequencies. To minimize the thermal crosstalk between adjacent pixels, we optimized our design to improve the thermal crosstalk. Each pixel is controlled under electrical excitation in a programmable manner, and the experimentally obtained modulation speed can reach ~1 kHz. This progress demonstrates its potential in applications requiring a high frame rate, such as THz beam scanning and compressive imaging. By exploiting the hysteresis effect of VO2, each pixel of the array can operate as a non-violative memory. Each pixel could store multiple THz reflection amplitudes by applying a sequence of short current pulses with different peak amplitudes. Our work offers a general approach to develop programmable devices with memory behaviours.
Principles of programmable THz memory metamaterials
As shown in Fig. 1, the programmable THz memory metamaterials consist of 8×8 pixels, each connected with an amplifier and an I/O port of field-programmable gate array (FPGA). Since the state of each pixel in the device could be switched independently, the programmable device could generate various THz spatial patterns such as ‘N’, ‘J’, and ‘U’. The VO2 patches are integrated into each pixel, and their hysteretic behaviours are utilized to realize the persistent storage of THz reflectance. Multiple states could be written and erased by applying current pulses with different amplitudes into a pixel. The designed device has the capability of multi-state THz image storage.
The optical picture of the proposed device and the microscopic image of a pixel is shown in Fig.2a,b. The pitch of each pixel is 1900 μm×1900 μm and ten rows and ten columns of unit cells are included in a pixel. The designed unit cell has a metal-insulator-metal (MIM) structure. For the top metallic layer, the VO2 patches act as the active material in response to electrical stimuli. The gold bowtie antennas connected to the VO2 patches are used to interact with the THz wave (see Supplementary Note 1 for details). The bottom metallic layer is continuous and acts as the ground plane. The 150-μm-thick c-cut sapphire substrate is sandwiched between two metallic layers.
The working principle of the designed device is similar to Salisbury screen 25. The resonance frequencies mainly depend on the thickness and permittivity of the spacer. The VO2-metallic hybrid structures on the top metallic layer functions as the lossy screen. The THz reflection amplitude around the resonance frequency depends primarily on the conductance of VO2 patches. We measured the THz reflection spectra at various temperatures in x-polarised normal incidence using THz time-domain spectroscopy (THz TDS). As shown in Fig. 2d, the reflection amplitude around the resonance frequencies is sensitive to the change of the conductivity of VO2. As the conductivity of the VO2 film increases, the reflection coefficient decreases, indicating it gradually approaches the perfect absorption condition. Meanwhile, the resonance frequency experiences a blueshift, mainly due to the rise of the coupling between the two antennas 26. Besides, we altered the conductivities of VO2 film and simulated the THz reflection spectra as shown in Fig. 2e. The simulation results validate our theoretical analysis and experimental results. Utilizing the VO2 conductivity change during the phase transition, we realized the dynamic THz reflection with a modulation depth (MD) of more than 65%.
Thermal management is crucial for pixelated VO2 devices. When external electrical stimuli trigger a pixel, lateral thermal diffusion can heat the surrounding pixels. To optimise the thermal conduction, we established a model in which the thermal conduction from a pixel to the surrounding pixels, substrate and stages are considered (see Supplementary Note 4 for the thermal model). We took the numerical calculation and found that the thermal conduction between the MIM structures and the stage is critical to suppress the thermal crosstalk. To achieve good thermal contact between the two objects, a layer composed of liquid metal, i.e. gallium-based alloy, with a thermal conductivity as high as 73 W/(m·K), was used to fill their gap.
To verify our thermal management design, we used a thermal infrared camera to measure the temperature distribution. To study the thermal crosstalk between pixels, we selected two pixels (labelled as P1 and P2), one pixel away as schematically shown in Fig. 3a. P1 and P2 were triggered by a voltage bias of 50 V and 30 V, respectively. Infrared thermal images of the device before and after applying voltage bias were captured. The temperature variation distribution was calculated based on the two images (see Supplementary Note 3 for the temperature calibration). As displayed in Fig. 3b, the temperature rise is remarkable in the regions of P1 and P2. A line scan along the blue dashed line in Fig. 3b is displayed at the bottom. The temperature increment at P1 and P2 was approximately 10 ℃ and 7 ℃ respectively. In contrast, the temperature increment at the pixel which bridges P1 and P2 was about 1 ℃ It indicates that we suppressed the thermal crosstalk from the neighbouring pixels.
Meanwhile, we conducted the electrothermal simulation of the temperature distribution in the VO2- device using the finite element method (see Methods for details). We simulated the distribution of the temperature variation with the same voltage bias as the experiment. As shown in Fig. 3c, the simulation results indicate good agreement with the experimental results. In the corresponding line scan, there are sharp oscillations in the regions of P1 and P2. The fluctuation reflects the temperature contrast between the VO2 patch and the surrounding bare substrate, as illustrated in the close-up view of P1 at the top of Fig. 3c.On the contrary, the oscillation was not observed in Fig. 3b because of the limited resolution of the thermal camera. In both experimental and simulated results, there is an evident temperature rise in the bias lines of P1 and P2. Since the resistance of the bias lines is non-negligible compared with the resistance of VO2 patches in the metallic state, the ohmic loss results in a remarkable increase in temperature.
Spatial THz modulation
The bias voltage could trigger the phase change of VO2 and result in a remarkable change of conductivity. Correspondingly, the reflectance of the pixel experiences a noteworthy change. Therefore, the device can work as an electrically reconfigurable THz SLM. For SLM, the MD is defined as MD = |Ri,on - Ri,off|/Ri,off, where Ri,on and Ri,off denote the reflection amplitude of the ith pixel in ‘on’ and ‘off’ states, respectively. To verify its function as SLM, we generated THz patterns of ‘N’, ‘J’, and ‘U’ by switching on a combination of pixels optionally. We applied a voltage bias of 26 V to switch on a pixel. The collimated THz beam was used to shine the device, and the spatial distribution of the reflected THz beam was measured (see Supplementary Note 7 for experimental setup). A spatial map of MD could be obtained by measuring the spatial distribution before and after switching on the required pixels.
Figure 4a shows the spatial map of MD at 0.479 THz when the pattern of ‘N’ was generated by the device. The letter of ‘N’ is evident in the obtained map despite the variance of MD in the range of 23–65%. We also generated other patterns such as ‘J’ and ‘U’, and the obtained spatial maps of MD are displayed in Fig. 1. Meanwhile, we measured the temperature change when the pattern of ‘N’ was generated using a thermal camera (assuming an emissivity of 1). As shown in Fig. 4b, the letter of ‘N’ is also obtained since there is a remarkable temperature rise in the triggered pixels. In Fig. 4b, there is a bright vertical line in the middle, where the ground electrode is positioned. Since a large amount of current flow through the ground electrode, the non-negligible ohmic loss generates excess heat. The diffusion of the heat brings in thermal crosstalk to the neighbouring pixels. That is why we could see a bright spot at the bottom of Fig. 4a.
We found that the current bias is more suitable as the control signal of this device compared with the voltage bias. During the phase change process, there are several orders of magnitude drop in the resistance of VO2 patches. If we used voltage bias as the control signal, the power would increase dramatically during the phase change process, making the thermal management challenging to implement. On the contrary, if we choose the current bias, the driving power drops continuously until the sample reaches thermal equilibrium. In the following, we adopt the current bias as the control signal.
We applied a square wave current with a peak amplitude of 60 mA to a pixel and measured the modulated reflection signal. Figure 4c shows the modulation amplitude of a pixel (ΔRi) (the measured signal difference between ‘on’ and ‘off’ states for the ith pixel) as a function of frequency. The 3 dB cut-off frequency was more than 1 kHz. That is much higher than previous reports 20. As shown in Fig. 4d, when the modulation frequency increases to 1195 Hz, the THz reflection signal is periodically modulated with the same frequency. The increase of modulation speed is mainly attributed to the highly thermally conductive layer, which improves the thermal contact between the sample and the stage.
Multi-state THz image storage
In Fig. 5a, the resistance–temperature curves of a pixel during the heating and cooling processes are shown. The phase transition process of VO2 film is hysteretic, making the sample capable of retaining a ‘memory’ of the transformed state. When the external stimuli are removed, the resistance of VO2 patches will stay at another steady state rather than returning to the initial state. Our device can, therefore, be treated as a memory array. In the following, we verify the multi-state memory operation of the device in response to current pulses with different amplitudes and pulse widths.
The current signal applied to one pixel for memory operation is shown in Fig. 5b. The stage temperature (Tstage) was set to 50 ℃, far below the Tc. A bias current of 40 mA was applied to achieve the maximum hysteresis. This bias is referred to as a ‘read’ signal. A current pulse signal with an amplitude higher than 40 mA is referred to as a ‘write’ signal. The pulse width was 10 s. Four different types of ‘write’ signals, with currents of 50 mA, 60 mA, 70 mA, and 80 mA, were applied. Besides, a 20-s-wide pulse with zero current was used as an ‘erase’ signal. When the ‘erase’ signal was applied, the pixel cooled down to Tstage and returned to its original state. We obtain the reflection amplitude of a pixel with a sequence of ‘read’, ‘write’ and ‘erase’ signal and then repeat the cycles with different ‘write’ signals. The corresponding normalised reflection amplitude as a function of time is shown in Fig. 5c. After applying four kinds of “write’ signals, the THz reflection amplitude would stay at the levels of 0.59, 0.55, 0.53, and 0.51 correspondingly during the “reading” process. The above results indicate that the THz reflectance of each pixel is non-volatile after application of a current pulse, i.e., the MD of each pixel is stored in the memory
To verify multi-state storage capability in a programmable manner, we applied a ‘write’ signal of a 100-ms-wide current pulse to every pixel in a serial way. In this step, the Tstage was held close to the Tc of 56 °C to maximise the hysteresis effect. The letter ‘N’ was written into the 8×8 array by injecting different “write” current pulses into specific pixels. The 100-ms-wide current pulses with amplitudes of 20 mA, 30 mA, 40 mA, and 60 mA (Fig. 6a) were used to write the letter ‘N’. Maps of the MD at 0.429 THz obtained by raster scanning is shown in Fig. 6b. In all four pictures, the letter ‘N’ is clear, but there are remarkable differences in the brightness, which corresponds to the MD. We averaged the MD in the black box areas of the four spatial maps. Four different values in MD were obtained. As shown in Fig. 6c, the MD increase monotonously as the current increases
Utilising the four-state memory of the programmable device, we could spatially modulate the THz beam with greyscale. In the following, we selected four groups of pixels, and each group consisted of two pixels to form a loading-icon-like greyscale map as schematically illustrated in Fig. 6d. We injected the above four types of pulses into the four groups of pixels, respectively. A loading-icon-like greyscale image with four different greyscales was obtained, as shown in Fig. 6e. The various greyscales correspond to different MD. To measure the retention time of the state storage, we remeasured the spatial map of the MD, after 5 hours, as shown in Fig. 6f. We did not observe any apparent change in this figure, suggesting that the multi-state can be well stored for at least several hours.