Observing and recording transient events is crucial for understanding fundamental principles and controlling related processes, such as inertial-confinement fusion [1], interactions between the laser and materials [2], plasma physics [3], and laser surgery [4]. To investigate the underlying mechanisms of transient processes, imaging methods with high spatiotemporal resolution are required. High-speed cameras with a rate of millions of frames per second can satisfy the requirement for dynamic observation within the microsecond time range, but the limitations of on-chip storage and electronic readout speed make it challenging to record dynamic scenes occurring in shorter timescales. In contrast, framing photography captures transient events with multiple physical or virtual cameras by temporal division [5], which obtains temporal resolution up to femtoseconds, bypassing the electronic speed limitation, and therefore it has emerged as the predominant tool for observing ultrafast phenomena. However, existing framing photography techniques face one or more issues of limited sequence depth, low image quality, low temporal resolution, or fixed frame interval, hampering the precise investigation of ultrafast dynamics. For instance, an ultrafast framing camera consisting of beam splitters and several gated intensified CCD cameras [6] has high image quality but confronts the trade-off between the sequence depth and light throughput, associated with the high expense when scaling the sequence depth. Ultrafast optical framing photography techniques generally transform temporal information into other domains, such as wavelength [7–10], space [11], spatial frequency [12,13], angle [14], and polarization [15,16] and indirectly record ultrafast scenes. For the transformation from the temporal to other domains, a specially designed laser pulse train is generally used as an illumination source. However, most ultrafast optical framing photography techniques construct the illumination source by beam splitting and individual modulation, which makes it difficult to further increase the number of channels. Some methods adopt a pulse instead of a pulse train to reduce the complexity of the illumination system and increase the sequence depth, such as spectral-filter-based sequentially timed all-optical mapping photography [8] and chirped spectral mapping ultrafast photography [10]. These methods typically provide a large number of frames. However, severe crosstalk between adjacent frames can substantially diminish temporal resolution. In addition, computational reconstruction is required in some framing photography techniques, which degrades the imaging quality due to information loss and reconstruction errors. For example, light-in-flight holography requires reconstruction from holograms at different spatial positions [11]. Transient frames modulated by stripes with different orientations are compressed into a single image in the frequency recognition algorithm for multiple exposures [12], which must be recovered by cropping and shifting in the spatial frequency domain. In addition, the frame interval and time window of most ultrafast optical framing photography techniques are difficult to change, because the illumination and detection systems are generally fixed and need to match with each other, which makes it inconvenient to investigate ultrafast processes with different timescales.

To address the challenge of achieving ultrafast framing photography with large sequence depth, high image quality, ultrashort exposure time, and flexible frame interval simultaneously, here we propose a spatiotemporal shearing-based ultrafast framing photography, termed STS-UFP. A pulse train generation device based on a spectrum shuttle is employed to illuminate dynamic scenes, enabling flexible adjustment of both the number of pulses and the time interval, which determines the sequence depth and time window. The ultrashort duration of the sub-pulse also enables a precise temporal slice of dynamic scenes, avoiding spatiotemporal blurring. A streak camera is utilized to record the dynamic scenes by spatiotemporal shearing, which reassigns the images at different instants to different spatial locations. The time window of the streak camera can be easily adjusted by modifying the slope of the sweep voltage, effectively accommodating the detection requirements of a tunable time window. With its adaptable illumination source and detection system, STS-UFP achieves a flexible frame interval and time window, enabling the observation of ultrafast phenomena across various timescales. In contrast, although a flexible pulse train generation device is utilized, some spectral division framing photography techniques still encounter difficulties in adjusting the time window and incorporating additional imaging channels, because the spectral imaging devices must be compatible with the illumination source [9]. Additionally, to balance the field-of-view (FOV) and sequence depth in STS-UFP, a spatial slicing device is adopted. By duplicating the dynamic scene and reassigning the scenes in the horizontal direction with a vertical shift, the recorded area is transferred into three horizontally arranged slices after filtering through the entrance slit of the streak camera. The trade-off between the sequence depth and height of the FOV is well addressed. STS-UFP achieves high-fidelity framing imaging with a sequence depth of up to 16 and a frame interval ranging from hundreds of picoseconds to nanoseconds. The outstanding performance of STS-UFP is demonstrated through three experiments: femtosecond laser-induced plasma and shockwave in water, femtosecond laser ablation in biological tissue, and femtosecond laser-induced shockwave on a silicon surface. Featuring high image quality, ultrashort exposure time, large sequence depth, and flexible frame interval, STS-UFP will undoubtedly find extensive applications in the studies of ultrafast phenomena.

The experimental configuration of STS-UFP is illustrated in Fig. 1(a). A Ti:sapphire femtosecond laser (Coherent, Legend Elite USP-HE) with a central wavelength of 800 nm, pulse width of 35 fs, and repetition rate of 1 kHz provides the source for pulse train generation. The repetition rate is first reduced to 100 Hz using an optical chopper (Thorlabs, MC2000B), and then a single pulse is obtained through an electronic shutter (Newport, 76993). This single femtosecond laser pulse is then temporally shaped into a pulse train using the spectrum shuttle method [17]. The femtosecond laser is incident on a grating (G1, LBTEK, BG25-1200-750) and dispersed horizontally after passing through beam splitter BS1. It is then incident on another grating (G2, Spectrogon, 715.700.550), continuing its horizontal propagation. The horizontally dispersed laser enters between the two parallel mirrors with a slight vertical shift, M1 and M2 (Thorlabs, BBSQ2-E03), and is reflected back and forth repeatedly, as shown in Fig. 1(b). Each time it is reflected, a specific spectral component of the laser leaks from the edge of mirror M2 and is incident on mirror M3 (Thorlabs, BBSQ2-E03), thus forming a series of sub-pulses. The pulse train is then reflected by mirror M3 and returns along its original path. After reflection by BS1, the pulse train illuminates the dynamic scene (DS). The spectrum shuttle method can generate laser bursts with time intervals ranging from hundreds of picoseconds to nanoseconds, and the number of pulses can be adjusted. In addition, the spatial profiles of each pulse can be shaped individually.

After illuminating the dynamic scene, the pulse train is collected by the objective lens (OB) and imaged through the lens (L1) with a focal length of 200 mm. The beam passes through a 750 nm long-pass filter (F1) to eliminate the pump laser and a half-wave plate (HP, Thorlabs, AHWP10M-980) to adjust the polarization state. It is subsequently imaged onto the slit (S), which limits the horizontal boundaries of the imaging area. The intermediate image is relayed by a lens pair (LP) consisting of two lenses with focal lengths of 200 and 500 mm and is then directed to an image duplication and shearing system. The image is split through a polarized beam splitter (PBS, LBTEK, MPBS642), and the energy ratio of the split beam can be controlled by rotating the half-wave plate. The vertically polarized beam first passes through a quarter-wave plate (QP1, Thorlabs, AQWP10M-980), is reflected by mirror M4, and then passes through QP1 again. As a result, the beam becomes horizontally polarized and transmits through the PBS and is finally imaged onto a streak camera (SC, Hamamatsu, C7700). The beam transmitted through the PBS passes through a quarter-wave plate (QP2, Thorlabs, AQWP10M-980) and is then split into two beams by a beam splitter (BS2). These beams are reflected by mirrors M5 and M6, which are oriented at specific angles. This orientation results in a spatial shift of the images from the streak camera (SC), as illustrated in Fig. 1(c). The image is replicated as three images, with one-third of each image cropped by the entrance slit of the streak camera, as shown in Fig. 1(d). This configuration allows the streak camera to capture more temporal images vertically, thereby increasing the sequence depth.

A schematic diagram of the image acquisition in STS-UFP is illustrated in Fig. 2. A dynamic scene I(x,y,t) is illuminated by a pulse train to capture image sequences with long intervals and short exposure times, with the pulse train illumination operation denoted as D. The discrete images are duplicated into three copies through a duplication operation R, and each copy is then processed via cropping operation C to obtain three slices, with each slice containing one-third of the discrete images. These slices are subsequently spatially shifted by a spatial offset operation O to ensure the horizontal alignment of the images. The R and O operations are executed by the image duplication and shearing system depicted in Fig. 1(a), whereas the C operation corresponds to the cropping process at the entrance slit of the streak camera, as shown in Fig. 1(d). We use A to represent the process of manipulating an image in space, denoted by the equation A=OCR. A measurement image M(x′,y′) is then obtained through a time shearing operation S and time integration operation T of the streak camera. The forward image acquisition process can be mathematically expressed as M(x′,y′)=TSAD·I(x,y,t).

Discrete frames of the dynamic scene can be extracted and recombined from measurement M, represented as the P operation in Fig. 2. To enhance the image quality, the Laplacian blending method [18] is employed to minimize intensity mismatches at the splice positions, while a denoising algorithm based on DRU-Net [19] is utilized to suppress noise.

The streak camera is a key component of the STS-UFP system. Typically, a streak camera operates in a one-dimensional mode to record the temporal evolution at a specific line position. A narrow entrance slit is employed to minimize spatiotemporal crosstalk [20]. However, in the STS-UFP system, the slit of the streak camera is opened to enable two-dimensional imaging within a designated area. When the duration of the sub-pulses is less than or equal to the temporal resolution of the streak camera, the captured images can be considered non-sheared. Spatial slice method is utilized to address the trade-off between the sequence depth and the height of FOV in STS-UFP. Given the fixed size of the streak camera sensor, more spatial slices allow larger sequence depth but narrower width of FOV. Besides, the spatial slice is conducted by duplicating and then cropping the scene, which means that larger spatial slice number leads to lower light throughput. Considering the balance between sequence depth and signal-to-noise ratio, spatial slice number of 3 is selected.

We characterize the parameters of the experimental setup using an STS-UFP system equipped with a 20× objective (Mitutoyo, M Plan Apo 20×) to observe the USAF 1951 test target, and the results are shown in Fig. 3. By stitching the three sub-images together, we obtain the image shown in Fig. 3(a). The orange dashed lines indicate the stitching boundaries. The 7-6 pattern (228 lp/mm) on the USAF 1951 test target is clearly visible, indicating a high spatial resolution fidelity. The intensity distribution curves along the horizontal and vertical lines are both extracted, as shown in Fig. 3(b). It is worth mentioning that this is not the upper resolution limit of STS-UFP. A higher spatial resolution can be achieved using an objective lens with a higher numerical aperture (NA). By adjusting the positions of the mirror pair in the illumination system of STS-UFP, two types of pulse trains are presented: 12 sub-pulses with a 2 ns time interval and 16 sub-pulses with a 500 ps time interval. It is worth noting that not all the pulses generated by the spectral shuttle are used in STS-UFP. Considering the size of the streak camera sensor and the uniformity of the pulse train intensity, we only selected parts of sub-pulses for illumination. The temporal distributions of the pulse trains are characterized using a streak camera operating in one-dimensional mode, as illustrated in Figs. 3(c) and 3(d). Here, the time windows of the streak camera are set to 20 ns and 5 ns, respectively. The vertical pixel count of the working area of the streak camera is 1016, corresponding to a maximum temporal resolution of approximately 19.7 and 4.9 ps. Since the duration of the sub-pulses is shorter than the temporal resolution of the streak camera, the images collected by STS-UFP exhibit high fidelity without blurring from spatiotemporal shearing. Meanwhile, the duration of individual sub-pulses within each of the two pulse trains is approximately 4.8 ps and 3.5 ps, respectively. This estimation is derived from the temporal stretching caused by the double-grating setup and the spectral characteristics of the sub-pulses within the spectrum shuttle system. The duration of the laser pulse prior to spectrum shuttle is measured using a streak camera operating in one-dimensional mode. The spectral bandwidth of both the laser pulse and the sub-pulses is assessed with a spectrometer. Assuming no higher-order dispersion, the duration of the sub-pulses is estimated based on their spectral weight. The sub-pulse duration directly influences the exposure time of the captured transient images.

The hydrodynamic effect of laser-induced breakdown in water has been extensively investigated in recent years [21]. When an intense laser pulse is focused on water, plasma is generated at the focal point, and the accumulated energy leads to micro-explosions that produce shockwaves in the water. The generated shockwave gradually decays into acoustic waves during transmission. Laser-induced acoustic sources have applications in various fields, including underwater communication [22], particle removal [23], and biomedicine [24]. STS-UFP is used to observe laser-induced plasma and shockwave in water. The experimental configuration is shown in Fig. 4(a). An 800 nm femtosecond laser is frequency-doubled by a β-BBO crystal to produce a 400 nm femtosecond laser. The 400 nm femtosecond laser with a pulse energy of 200 μJ is transversely focused into a cuvette filled with distilled water through an objective (Mitutoyo, M Plan Apo 20×). The laser-induced dynamic scene is then observed through a shadowgraph configuration using STS-UFP equipped with a 10× objective (Mitutoyo, M Plan Apo 10×).

A pulse train consisting of 12 sub-pulses, each with a 2 ns time interval, is employed as the probe beam, resulting in the capture of 12 transient images at 2 ns intervals, as illustrated in Fig. 4(b). The observation time window for the pulse train is 22 ns. The red arrow in the 0 ns image in Fig. 4(b) indicates the direction of femtosecond laser propagation. As shown in the 0 ns image, the high input laser intensity and tight focus of the objective lens lead to a self-focusing phenomenon near the laser focal point, which rapidly ionizes the molecules and creates a plasma channel [21]. The image at 4 ns reveals the formation of a shockwave layer outside the plasma channel, with the separation between the shockwave and plasma channel indicated by two red triangles in the image. After 4 ns, the contours of the shockwave and plasma become distinctly recognizable, spreading outward from the plasma channel. In the image at 6 ns, “S” and “P” denote the positions of the shockwave and plasma, respectively. We extract the evolution of the shockwave and plasma radius from 4 to 20 ns, as shown in Fig. 4(c). The results indicate that the shockwave expands at a relatively stable velocity of approximately 1.5 km/s. Meanwhile, plasma expansion occurs slowly within this timescale, resulting in separation between the shockwave and plasma. The spatiotemporal evolution videos corresponding to Fig. 4(b) are provided in Visualization 1.

Femtosecond lasers have a wide range of biomedical applications, including ophthalmic surgery [25], high-precision processing of bone tissue [26], and ablation of neural tissue [27]. Visualizing the interaction between ultrashort pulses and biological tissues enhances our understanding of the ablation mechanism, which in turn provides the guideline for femtosecond laser surgery. Due to the inhomogeneity of biological tissues, the femtosecond laser ablation process is not reproducible, rendering the conventional pump-probe method [28] unsuitable. In contrast, STS-UFP offers a more effective alternative. We conduct experiments using the configuration illustrated in Fig. 5(a) to validate the potential of STS-UFP in observing the interaction of ultrashort pulses with biological tissue. A 400 nm femtosecond laser with a pulse energy of 110 μJ is reflected by a 650 nm long-pass dichroic mirror (DM), transmitted through the objective (Mitutoyo, M Plan Apo 10×), and then focused on the surface of the onion epidermis. In this configuration, 12 probe pulses with a 2 ns time interval are generated for illumination, and the ablation scene information is collected by the same objective and detected by the imaging system.

The obtained image is presented in Fig. 5(b). A femtosecond laser is directed onto the surface of the upper epidermal cell, resulting in the formation of a black region at the center of the cell, which corresponds to the laser-induced plasma generated [29]. The time-dependent transmittance in the ablated region is illustrated in Fig. 5(c). The transmittance decreases rapidly after ablation because of the absorption of the pump light by the generated plasma. The relative transmittance is defined as ΔT/T0(ΔT=T−T0), where T and T0 represent the transmittances with and without pump laser excitation, respectively. The sharp decline in transmittance indicates an increase in the plasma density induced by femtosecond laser ablation. Furthermore, the ablated area expands gradually, and we measure the radius of the expanding black region at the position indicated by the red arrow in Fig. 5(d). The black plasma produced by ablation expands slowly over 0–20 ns, contributing to the formation of cavitation bubbles. The size of laser-induced cavitation significantly influences the extent of tissue disruption during laser surgery [4]. The spatiotemporal evolution movies corresponding to Fig. 5(b) are provided in Visualization 2.

To verify the applicability of STS-UFP in various dynamic scenarios, we further record femtosecond laser-induced shockwaves on a silicon surface. Silicon is a fundamental material in the semiconductor industry, and examining the dynamics of laser-induced ablation on silicon surfaces is essential for understanding the interaction mechanisms between femtosecond lasers and silicon materials, ultimately enhancing the laser processing efficiency. The experimental setup is shown in Fig. 6(a). The observations are conducted using an STS-UFP system equipped with a 10× objective (Mitutoyo, M Plan Apo 10×) in a shadowgraph configuration. Here, 16 probe laser pulses are generated at 0.5 ns intervals, resulting in 16 transient images, as shown in Fig. 6(b). The observation time window is 7.5 ns. A 400 nm femtosecond laser with a single pulse energy of approximately 200 μJ is focused through a lens (LBTEK, MCX10610-A) to ablate the silicon surface.

The image at 0 ns illustrates the formation of a long laser filament, which results from the high-energy laser-induced plasma channel generated in the air due to the focusing of the laser with a long-focal-length lens [30]. After the femtosecond laser hits the silicon surface, a mass of plasma is produced in the ablation region, as evidenced by the images taken at 0.5–1.5 ns, where the black regions indicate laser-excited electrons within the plasma absorbing the probe laser [31]. The plasma gradually becomes transparent, pushing the surrounding air outward, which gradually forms a shockwave. Between 2 and 7 ns, the shockwave front expands rapidly and exhibits a bulging effect. The focused femtosecond laser pulse ionizes the air, creating a narrow plasma channel and generating a cylindrical shockwave. As the time progresses, the channel expands, resulting in the formation of a low-pressure region within it. This low-pressure area accelerates the propagation of the shockwave front, contributing to the bulging effect observed in the hemispherical shockwave [32]. Here, we employ a long-focal-length lens to focus the laser to generate long filaments that enhance the prominence of the bulge in the shockwave front. Additionally, we measure the horizontal and oblique 45° directional expansion distances of the shockwave over time, as indicated by the blue and green arrows in the image at 4 ns, as shown in Fig. 6(c). It can be observed that the shockwave expands rapidly in the initial stages and then progressively slows down. We fit the shockwave expansion distance in the oblique 45° direction using the Sedov–Taylor theory [33], which remains unaffected by the presence of air plasma. According to the Sedov–Taylor theory, the radius R of the shockwave with respect to time t is expressed as R∝Bt2/5, where B is a constant related to energy and air density. The fitted curve is also illustrated in Fig. 6(c), with a coefficient of determination of 0.98. The high agreement between the experimental and theoretical values demonstrates the exceptional accuracy of STS-UFP. The average expansion velocities of the shockwave in the horizontal and oblique 45° directions are calculated to be 16.5 and 10.39 km/s, respectively. The spatiotemporal evolution movies corresponding to Fig. 6(b) are provided in Visualization 3.

In summary, we have developed an ultrafast framing imaging photography, STS-UFP, which combines time-discrete illumination with spatiotemporal shearing imaging. This method offers large sequence depth, high image quality, ultrashort exposure time, and flexible frame interval. To validate the impressive ultrafast imaging capabilities of STS-UFP, we capture three types of ultrafast events: laser-induced plasma and shockwaves in water, laser ablation of biological samples, and laser-induced shockwaves on a silicon surface. The shockwave propagation, plasma behavior, and structural damage resulting from femtosecond laser interactions with matter are visualized and analyzed. STS-UFP enables in situ observation with long high-fidelity image sequences, significantly facilitating further research on ultrafast phenomena, such as exploring femtosecond laser ablation mechanisms [34], optimizing laser manufacturing parameters [35], and guiding laser surgery [25–27]. With these advantages, STS-UFP serves as a powerful tool for observing the intricate details of ultrafast dynamics and holds substantial potential to advance the study of ultrafast phenomena. The rapidly developed burst-mode laser serves as a potential illumination source for STS-UFP, offering high pulse power and flexibility in interval adjustment [36]. Furthermore, by integrating with other imaging paradigms, STS-UFP can achieve enhanced imaging capabilities. For instance, it can provide transient references [37] for compressed ultrafast photography, enabling high-fidelity imaging with a higher frame depth and achieving ultrafast complex-amplitude imaging through integration with off-axis holography [38]. To reduce the system costs, a combination of an electro-optical deflector and a camera can be utilized instead of a streak camera [39].