元面积制造从化学蒸气沉积生长并转移到熔融石英底物上的大面积单层石墨烯开始（补充注释1）。用双层聚（PMMA）（PMMA 495底层和PMMA 950上层）在3,000 rpm上旋转底物，持续30 s，然后在180°C下进行1分钟的前烘烤。然后使用导电聚合物层（放电），以确保电子束暴露期间充电的充电耗散。然后使用100 kV JEOL JBX-6300FS系统通过电子束光刻进行阵列。为了 <30 nm features, interparticle proximity effect correction is used with an optimized total dosage around 1,650 µC cm−2. Under these conditions, writing of a 1-mm2 array takes approximately 1 h. Following a 30-s rinse in isopropanol for DisCharge removal, samples are then developed in a 3:1 isopropanol:methyl isobutyl ketone solution for 1 min. A 30-nm gold layer is then deposited by means of electron-beam or sputter deposition without any adhesion layer. Liftoff is performed after overnight acetone soaking, using gentle acetone wash bottle rinsing to remove the residual PMMA/gold. For most samples, no particle delamination is observed.

　　Devices for simultaneous ultrafast THz emission and average photocurrent electrical readout studies are prepared using maskless photolithography (Heidelberg MLA 150). We use AZ 5214 E photoresist in image-reversal mode, precoating the samples with a hexamethyldisilazane layer to promote photoresist adhesion. Excess graphene is etched to obtain desired geometries by means of 2-min exposure to O2 plasma (100 W, 10 sccm; Anatech RIE). Electrodes consist of 50-nm Au with a 3-nm Ti adhesion layer.

　　For back-gated devices, the back-gate electrode with minimal overlap with the top electrodes (precluding any shorting through defects in the dielectric spacer layer) is prepared through photolithographic patterning, deposition of 3 nm/30 nm Ti/Pt and a liftoff process. Various spacer-layer materials (SiO2, Al2O3 and HfO2), deposition methods (atomic layer deposition and physical vapour deposition) and thicknesses (5–100 nm) are tested for optimal electrostatic gating of large-area devices. We find that 30-nm SiO2 offers a good trade-off between high gating capacitance and relatively modest nanoantenna resonance redshifting owing to the image charge oscillation within the Pt and the dielectric environment of the spacer layer. Subsequent device preparation follows as described above.

　　The quality of metasurface fabrication is characterized by means of optical microscopy, scanning electron microscopy (SEM) and white-light transmission spectroscopy. Optical bright-field and dark-field imaging under 50× magnification is sufficient for resolving individual nanostructures to verify successful liftoff and large-scale sample uniformity while also resolving the monolayer graphene edge in etched devices. A more detailed view of the graphene and nanostructure morphologies (Supplementary Figs. 1 and 2) is provided by means of SEM micrographs collected using FEI Magellan, Nova NanoSEM 450 and Nova NanoLab 600 systems. The fabricated nanoantennas closely reproduce the design profile down to the tens-of-nanometres scale, with extra-sharp approximately 15 nm radius of curvature tips.

　　metasurface optical properties are verified by means of white-light transmission spectroscopy using a tungsten-halogen white-light source passed through a broadband polarizer for linear polarization control. Two 20× microscope objectives are used to focus the light through the sample and collect the transmitted light into both visible (Acton SP2300) and near-infrared (Ocean Optics NIRQuest) spectrometers. The transmittance is given by the ratio of signal (on the metasurface) to reference (off the metasurface but on the same substrate), with the ambient background collected with the white-light source turned off and subtracted from each.

　　THz emission experiments are performed across several systems with different functionality, with overlapping datasets verifying reproducibility. All measurements except wavelength-dependent THz emission (see below) are performed in the low-fluence regime (<1 µJ cm−2) using a Ti:sapphire oscillator (Chameleon Vision-S) with 800-nm, roughly 100-fs pulses at the sample location at 80 MHz repetition rate. As seen in Fig. 1e, these low fluences ensure linear responses and also preclude thermally induced nanoantenna deformation or photochemical degradative effects51 that can occur in the intense hotspot regions. The gradual onset of a sublinear regime beyond about 0.8 µJ cm−2 in Fig. 1e is attributed to competing temperature-dependent thermodynamic contributions (Supplementary Notes 3 and 4). For metasurfaces operating at 800 nm, the THz radiation is measured through electro-optic sampling52 using a 1-mm 〈110〉 ZnTe crystal. Three wire-grid polarizers are used to measure the x and y field components of the THz radiation, with the first polarizer at 0° or 90°, the second polarizer at 45° and the third polarizer at 0°. The THz beam path is contained within a dry-air-purged environment (<2% relative humidity).

　　Pump-wavelength-dependent THz emission experiments are performed using a noncollinear optical parametric amplifier to generate pump pulses across 725–875 nm (pulse width ≤ 50 fs) and an optical parametric amplifier to generate pump pulses across 1,450–1,650 nm (pulse width ≤ 200 fs). Both amplifiers are seeded by a 1-MHz fibre-amplified solid-state laser producing approximately 270-fs pulses at 1,032 nm. In all of these measurements, we tune the pump wavelength while the incident fluence is maintained at a constant value using a neutral-density attenuator. The direct output of the seed laser is used to gate the electro-optic sampling with a 0.5-mm 110 GaP crystal. This enables wavelength-dependent pumping without affecting the detection.

　　For spatial mapping of THz vector beams, we use a custom Menlo Systems apparatus consisting of a 100-MHz mode-locked 1,560-nm erbium-doped all-fibre laser oscillator that seeds two separate amplifiers. The first line uses a high-power erbium fibre amplifier driven in the linear pulse propagation regime, followed by free-space second-harmonic generation, outputting 1 W of 780-nm, 125-fs pulses and serving as the pump beam for generation of THz radiation from metasurfaces. The second amplifier line is used to gate a broadband, fibre-coupled photoconductive antenna detector53 (TERA 15-RX-FC), which is mounted on a 2D stage for automated xy spatial scanning. Along with the high-repetition integrated delay stage (>45 Hz对于20-PS的时间扫描范围），该系统可以快速对THZ矢量梁的快速高光谱成像，这些成像是由两个TPX透镜（50 mM焦距）精加工和重新聚焦的。光电传导天线位于第二镜头的焦点之前，在图3中显示的结果中，校准并校正了平面相位前部的偏差。

　　使用锁定放大器（以最高的信号噪声比）和皮科疗法计（直接测量极性），将时间平均的光电电气读取。使用计算机控制的双通道源表（Keithley 2614b）同时应用栅极电压并读数设备电阻（使用源仪）或光电流（使用picoammammeter或lock-In放大器）进行门控研究。尽管石墨烯设备中的非局部光电经常通过冲击 - 雷莫型响应进行评估1,54，但此处的方形跨面和电极几何形状导致整体x和y光电流成分的整体X和y光电成分进行了简化，直接读数。通过直接调整80-MHz Ti：Sapphire激光器的输出波长，同时保持恒定功率来测量波长依赖性光电流响应。尽管电极金甘烯连接连接也可以有助于光热电流，但通过纳米antennans诱导的电流通过：（1）聚焦束的空间位置；（2）更强的纳米反胶依赖性；（3）在无纳米and的设备中缺乏类似的响应（有关更多详细信息，请参见图5参见图5）。

　　使用有限元法（Comsol多物理6.1）求解耦合的电磁，热力学和流体动力方程。首先，使用由空气超隔酸盐，金纳米antennas（约翰逊和克里斯蒂55给出的介电功能），导电石墨烯边界层和融合的石英底物56组成的3D矩形跨侧面单位细胞结构域模拟纳米级光场。周期性（Floquet）边界条件在横向X和Y方向上应用，沿Z方向的输入/输出端口由完美匹配的层支持。石墨烯光导率在近红外频率范围内大约是恒定的57，实际分量σR= 6.1×10-5Ω-1和虚构的组件σi= -2.1×10-5Ω-1。与实验相一致，从平面波激发下的模拟S参数中获得了谐振传输光谱。相同的纳米annna几何文件用于光刻和模拟。

　　2D石墨烯电子和晶格的完整热演化（特别是光电子31）系统是通过耦合热方程计算的

　　in which ce is the volumetric electronic specific heat described in Supplementary Note 3, is the in-plane electronic thermal conductivity approximating Wiedemann–Franz law behaviour (approximately preserved in the Fermi liquid regime of graphene58), gsub = 3 MW m−2 K−1 is used as the out-of-plane electronic thermal conductance owing to coupling with the SiO2 substrate phonons32 (corresponding to acooling length at 3,000 K), Top is the optical phonon temperature, cop is the optical phonon specific heat based on time-resolved Raman studies of graphite31,59 and κl ≈ 1.7 × 10−7 W K−1 is the lattice thermal conductance based on thermal transport measurements of supported graphene60 (1:1 device aspect ratio here).通过瞬态反射率测量值估计能量弛豫电子 - 光学声子耦合常数，GER≈2×107 W K -1 M -2（补充注释3）。源项是瞬态激光脉冲场的吸收功率密度，其中E0是在频域电磁模拟中确定的峰场，而τp= 100 fs是激光脉冲持续时间。忽略了较长的皮秒时标和辐射损失，与声音声子浴和辐射损失相结合。发现Peltier冷却的作用（将显示在热方程的右侧）很小。补充注释3中提供了有关热力学建模的更多详细信息，包括石墨烯高度限制的热力学量以及电子和晶格温度演化的简化两温模型。

　　公式（1）描述的电子温度演化（能流）通过光热力fpte te驱动动量流动，如时间依赖性线性化navier-Stokes方程所建模

　　假设带电的流体不可压缩流动

　　这里u是2D电子速度场（电荷密度的电流密度j = ENEU），ν是依赖温度的运动学电子粘度。正如电子流动流量的典型情况一样，雷诺数数很小，并且发现非线性对流项的贡献可以忽略不计。电子运动粘度线性取决于电子 - 电子散射时间ντEE，因此在局部缩放为大约（补充注释3）。扩展项的写入为与热方程中的扩散项相似，以正确地说明运动粘度的空间变化（扩展数据图4）。力项中s的温度依赖性（fpte = -qste，其中q =±e）通过​​求解广义的莫特关系来确定（补充注释3）。焦耳加热的少量贡献是四次依赖于当前密度的，因此未包括在当前处理中。驱动力FPTE在空间上高度不均匀，因此，速度流的空间曲线不仅由由动量弛豫和粘度或Gurzhi长度确定的长度尺度来控制。在扩展数据中显示了电子温度，力场，电荷流量和τee/τmr的纳米级时空演化，并在扩展数据中显示了进一步讨论补充注释4中提供的流体动力建模。