With the advent of the digital information age, the requirement to store vast amounts of data over the long term has become indispensable^{1,2,3,4,5,6,7,8}. While contemporary data storage methods, including magnetic and electric storage media such as hard disk drives and solid-state drives, address the current need for terabyte capacity, these solutions pose security risks such as susceptibility to degaussing, electric leakage and tampering, which will cause huge costs for long-term maintenance. A particular challenge is the continuous growth of energy consumption, driven by the exponentially increasing amount of information generated each year, imposing a massive barrier to the sustainability of big data^9,10. Thus, it is imperative to develop a solution for long-term data storage, distinguished by a long lifespan^{11,12,13,14,15,16} and low energy consumption^{2,16,17,18,19}, while also meeting practical demands, including high capacity^{3,4,8,13,15,18,20,21,22,23,24,25}, and rapid reading and writing capabilities^2,16.

Conventional optical data storage (ODS) technologies, such as compact discs, show relatively longer lifespan and lower energy consumption, but suffer from limited capacity. Nevertheless, the multi-dimensional nature of light endows ODS with substantial potential for high-density data storage^26,27. Recent advancements in materials enabling multi-dimensional storage have elevated optical storage density to an impressive level of terabits per cubic centimetre^{3,13,15,18,20,22,23,24}. Unsatisfactorily, these emerging schemes are not without drawbacks (Table 1). Some lifespans barely surpass those of conventional storage media, certain writing speeds and energy expenditures fall short of practical application requirements, and the fidelity in others deteriorates substantially. The quest for material that not only inherits the robustness and low-energy-consumption traits of ODS but also delivers exceptional information-carrying capacity and efficient data throughput remains a formidable challenge in meeting the demands of long-term data storage.

Wide-bandgap materials with a high Young’s modulus offer great potential for ODS media due to their exceptional robustness and transparency. One approach has focused on exploiting the charge states of point defects, such as nitrogen-vacancy centres in diamond²⁸. However, ODS systems using charge states encounter substantial obstacles in the realization of robust and high-density storage, including charge state instability during optical reading and neighbourhood crosstalk caused by optically induced charge carriers during the writing process^28,29,30,31. An alternative strategy involves birefringent nanostructures in silica glass with unequivocally long lifespan^11,14,15. Nevertheless, fabricating storage units with small sizes or high-density three-dimensional (3D) stacks proves challenging. These limitations have seriously impeded progress and cast doubt on the practical application of ODS with high capacity using these materials, as mentioned in refs. ^12,13.

In this context, we propose and develop an ODS system based on lattice structural engineering at atomic size in diamonds to address and surpass the abovementioned limitations, positioning it as a promising and practical solution for future information storage demands. We show that a single low-energy femtosecond pulse can on-demand generate Frankel defects in the diamond, functioning as individual storage units with dimensions beyond the diffraction limitation. We establish a complete ODS routine including four-dimensional (4D) multiplexing writing and multi-plane parallel optical reading techniques to achieve optimal information storage across various aspects, as detailed in Table 1.

In our scheme, the Frankel defect known as the general radiation 1 (GR1) centre stands out among hundreds of defects in a diamond^32,33 due to its exceptional structural stability, even at high temperatures, resulting in a lifespan of 10¹⁴ yr at room temperature storage and 1,000 yr at 500 K (ref. ³⁴; Supplementary Section V). Furthermore, the GR1 centre shows ultrahigh fluorescence stability as a fluorescent defect and shows resistance to photobleaching, making it an ideal candidate for long-lasting and frequent reading (Fig. 1b). As individual storage units, GR1 centres can be rapidly and precisely written using a single femtosecond laser pulse with ultralow energy consumption at the nanojoule level (Fig. 1a), enabling ultrahigh-speed data writing. Owing to the nonlinear interaction between the femtosecond laser pulse and diamond substrate, the single storage unit spot reaches a lateral feature size below 69 nm (1/12 of the wavelength; Fig. 1f) that breaks the diffraction limit for optical processing, unlocking great potentials for higher data densities.

a, The data are stored in a diamond storage medium via single femtosecond laser pulses. b, Long-term fluorescence stability under continuous 532 nm reading laser irradiation. The power density at the storage unit is 1.3 × 10⁷ kW m⁻². The yellow area is the fluorescence intensity stability required for greyscale encoding. c, Confocal microscopy imaging of a body-centred stacking data storage pattern. Laser beam wavelength at 532 nm excites the GR1 centres for fluorescence imaging. The dotted lines mark the staggered patterns between different layers. d, Schematic of thermal diffusion in the data-writing process. Owing to diamond’s extremely high thermal conductivity, the heat will spread rapidly to maintain high precision for subsequent data writing. The bottom figure shows the characteristic thermal diffusion time of different materials. The time is simulation result (‘Heat accumulation evaluation’ in Methods). e, The dot-dash line denotes the femtosecond pulse train measured by the photodiode. The blue line denotes picked single pulse under time control. The width of the electrical pulse signal generated by a single femtosecond pulse is about 4 ns limited by the photodiode bandwidth. f, The super-resolution optical fluorescence imaging of a data storage unit. The data storage unit presents a sharp, long strip. The full-width at half-maximum is about 69 nm and 160 nm, respectively. Negative signals come from the ‘negative’ ground-state depletion method (Supplementary Section II). Inset: polarization direction of the writing laser. g, The writing set-up with the self-corrected wavefront. The low-power femtosecond laser-generated plasma luminescence in diamond. After spatial filtering, the luminescence signal can reveal the laser focusing situation. On the basis of the luminescence intensity, the deformable mirror is automatically adjusted to optimize the wavefront and cancel the refractive index mismatch on the oil–diamond interface (‘Wavefront correction implementation’ in Methods). Inset: fluorescence intensity changes during wavefront correction.

Full size image

As initial and crucial progress towards realizing the proposed high-capacity diamond storage medium, we have successfully demonstrated a 4D ODS (3D spatial dimension plus fluorescence intensities multiplexing) with a remarkable storage density of 14.8 Tbit cm⁻³ and an overall writing–reading fidelity exceeding 99%. This high performance is achieved based on two key techniques: (1) developing an in situ, non-destructive and automatic wavefront correction technique that ensures precise 4D writing; and (2) demonstrating high-fidelity widefield microscopy parallel readout and multi-plane imaging technology for fast data reading. These demonstrated characteristics, including high storage capacity, ultralong lifespan, ultralow energy consumption, ultrashort exposure time, parallel recording and reading with high fidelity, combined with the diamond substrate’s ability to withstand extreme environments such as electric and magnetic fields and high pressure, make diamond storage medium an exceptional candidate for the next generation of ODS systems in various usage scenarios.

Figure 1a illustrates the fundamental conceptual scheme of the diamond storage medium. A femtosecond pulsed laser with a wavelength of 808 nm and self-corrected wavefront (Fig. 1g) is used to precisely write the storage units at different depths in the diamond. This process is facilitated by a high-numerical-aperture (NA) oil immersion objective lens (NA = 1.45, ×100), which focuses the laser beam onto the working plane (Supplementary Sections I–III). The emitted fluorescence from these storage units shows a distinct zero phonon line at 741 nm, consistent with the GR1 spectral feature³⁵ (Supplementary Sections IV and V). The long-term fluorescence stability of GR1s, even under continuous irradiation by a high-power-density reading laser (Fig. 1b), ensures their reliable reading and long-duration storage capabilities in our study (Supplementary Section VI). The diamond storage medium is fixed on a micrometre–nanometre combination stage to achieve precise writing over a large range with high accuracy. By synchronizing individual single femtosecond pulses with the movement of the diamond storage medium, a 3D array of storage units is written and imaged under confocal microscopy (Fig. 1c). Moreover, the single-pulse processing property substantially reduces the exposure time of a single storage unit from milliseconds^11,12,18 or microseconds¹³ to femtoseconds, enabling high-speed data writing at terabits per second. In addition, the ultrafast thermal dissipation time of the diamond substrate (Fig. 1d) ensures the possibility of ultrahigh-speed data writing. Failure to dissipate heat efficiently could lead to accumulated heat, especially for high-speed parallel writing, which would deteriorate the reliability and processing precision^36,37.

Owing to the excellent processability of the diamond storage medium, we have been able to achieve a 3D spatial data storage density that is close to the optical diffraction limit^2,12,17,18. Figure 1c shows a body-centred stacking data storage pattern with a horizontal and vertical separation of 450 nm and 1 µm, respectively. Under binary storage, this pattern corresponds to a storage density of 4.9 Tbit cm⁻³. The further increase of the spatial storage density is possible by exploiting super-resolution recording techniques^3,20,23, as the lateral separation is much larger than the intrinsic physical size of the storage unit (Fig. 1f). However, these techniques often entail trade-offs, impacting other aspects of the recording system’s performance, such as increased complexity, degraded fidelity in data retrieval and potential compromise of long-term stability.

To adequately address the demands of large-scale 3D data storage, rapid and automatic wavefront correction becomes imperative. As shown in Fig. 1g, optical aberrations caused by the non-uniform refractive index in the propagation path distort the focal intensity distribution and thus reduce the writing resolution and accuracy. Previous data-writing schemes utilize adaptive wavefront pre-compensation to mitigate aberrations, but in a non-automatic manner requiring extensive pre-calibration^38,39,40,41. Here we develop an appropriate correction method that utilizes the femtosecond laser-induced luminescence at the focal spot³⁹. The in situ detection of the upconversion luminous intensity allows non-destructive (Fig. 2a) and automated correction in a few seconds via maximizing the intensity using the gradient descent algorithm, which ensures the high-density 4D storage demonstrated below.

a, The fluorescence intensity of single storage units under different writing laser power. The wavefront correction is performed at an energy of around 2 nJ (the arrow) that does not interrupt the subsequent data storage. At each power, the fluorescence intensities of ten storage units were used for statistical averaging. Error bars represent standard deviation. b, Confocal microscopy imaging of storage units’ arrays written via a single femtosecond pulse with controllable laser power. A dotted grid separates the regions labelled as numbers 1–8. c, The histogram of fluorescence intensity of storage units for each region in b that presents a Gaussian distribution. The different areas are distinguished by colours and marked by numbers. d, The colour in c is encoded to 3-bit Gray code so that only 1 bit changes between 2 adjacent colours. e, Original image for storage. The image is obtained by binarizing the three primary colours of Henri Matisse’s Cat. f, Partial magnification of the image data for storage. g, The result of mapping the colour data in f to the writing power. h, Confocal microscopy imaging results of data stored in the diamond storage medium. The square localizes every storage unit. i, Fluorescence intensities distribution of all storage units from the diamond storage medium. Storage units are divided into categories according to the fluorescence intensities and mapped into the encoding space. j, Data restored from confocal microscopy imaging. Bit errors are marked with a cross. k, Total data restored from the diamond storage medium. The size of the image is 113 µm × 164 µm. Credit: e,k, Image courtesy of Gianantonio Muratori.

Full size image

Crucially, the exploration of multiplexing individual storage units is essential to further enhance the capacity of the ODS. In contrast to utilizing multiplexing wavelength^7,16,42, polarization^8,11,43 and lifetime^44,45, we exploit the dimension of fluorescence intensities to achieve high-fidelity 3-bit information multiplexing on a single storage unit. The relationship between the fluorescence intensities of the storage units and the writing energy is shown in Fig. 2a. Under a nonlinear multi-photon process (self-trapping of biexcitons)^46,47, the formation of GR1 vacancies is observed for only pulse energies above a threshold of 8 nJ (corresponding to a peak power density of 2.4 × 10¹⁰ kW cm⁻²), which ensures the stability of GR1. In the energy range of 8 nJ to 14 nJ, the writing power and fluorescence intensities show a well-defined linear dependence, suitable for fluorescence intensity encoding. The results of our batch writing test on eight different powers are shown in Fig. 2b, and the corresponding intensities distribution is depicted in Fig. 2c. At the same writing powers, there is a certain distribution of the fluorescence intensities. This reflects the combined effects of factors such as inhomogeneities of the diamond material, laser power stability and fluorescence fluctuations of the storage units. On the basis of the power dependence shown in Fig. 2a, the fluorescence intensities with different writing powers show significant distinctions beyond the fluorescence fluctuations. We use Gray code⁴⁸ to encode intensities into 3 bits. For visual distinguishability, we utilize a red–green–blue (RGB) three-primary-colour map to represent the intensity levels, as illustrated in Fig. 2d. To verify the actual performance, a typical and colour-rich RGB three-primary-colour image, Henri Matisse’s Cat (Fig. 2e), is used. Via a colour-intensity mapping (Fig. 2f,g), the femtosecond laser writing power is accordingly adjusted to implement the data storage (Fig. 2g,h). The recorded information is then retrieved from fluorescence intensities under confocal microscopy (Fig. 2h,i). On the basis of the intensity distribution, RGB data are recovered (Fig. 2i,j). The final result is shown in Fig. 2k. Here, we store 55,596 bits of data in a diamond storage medium, achieving a total fidelity (storage and readout) of 99.48%. In Fig. 3f, we store the film The Horse in Motion (the first film in the world; Eadweard Muybridge, 1887) with a space of 200 × 200 × 8 µm³ in the diamond storage medium.

a, Original data for 4D multi-layer data storage. Each image has four levels of fluorescence intensity. b, Restored data of 4D data storage in a diamond storage medium by a confocal microscope. c, Specific display of the restored data in b. d, Fidelity and transmission of difference layers. The multi-layers are written from bottom to top so that the fluorescence intensities of each layer without or with other layers’ cover are obtained and compared to get the transmission. The transmission of each layer is given by the statistical average of five randomly sampled storage units. The fidelity is given by statistical average of all storage unit points (60 × 60). Error bars represent standard deviation, and the standard deviation of fidelity is less than the markers. e, Restored data via the widefield parallel readout. The restored data are given after denoising, intensities correction and intensities decoding. f, Data restored from a diamond storage medium of The Horse in Motion (the first film in the world; Eadweard Muybridge, 1887). Credit: images in a–c,e, Pixabay.

Full size image

Multi-layer data storage is another indispensable but challenging means to achieve high storage capacity. Due to the poor axial resolution by far-field optical microscopy systems, challenges arise in reading the multi-layer data of previous works, such as crosstalk between layers and optical signal scattering⁴⁹. These problems are often reflected as a decrease in fidelity and capacity. To investigate these effects in our diamond storage medium, we demonstrate 6-layer 4D data storage (Fig. 3a,b) in the space of 50 × 50 × 9 µm³. As shown in Fig. 3c,d, we achieve a total recording fidelity of about 99.99% (the Agresti–Coull 95% confidence interval is 99.97–99.99%), and signal transmission approaching near unity, with no crosstalk and interlayer scattering kicking in. The main error contribution comes from the microscopy instabilities during the scanning of the fluorescence microscopy (Supplementary Sections VII and IIX). With fabrication at a depth of hundreds of micrometres in diamond⁴⁰, our results indicate that the diamond storage medium can achieve high-fidelity information writing and reading with more than hundreds of layers.

Beyond writing and reading one by one with high fidelity, multi-channel parallelism is an important requirement for high-efficiency data throughput. We demonstrate that the diamond storage medium is compatible with parallel techniques, accelerating data reading and writing. Figure 3e shows a multi-channel readout using widefield fluorescence microscopy (Fig. 4a). At 1,024 parallel channels, the reading fidelity is maintained at 99.8%, in which the main error comes from the light intensity non-uniformity of the illumination laser beam. This non-uniformity induces more errors for reading 3,600 parallel channels with fidelity degraded to 85.19%, which is improved to 99.24% after reconstructing the beam shape and correcting the intensity data (Supplementary Sections IX and X). These results verify the feasibility of high-fidelity parallel readout for high storage density. With the aid of other solutions to overcome uniformity issues, such as beam-reshapers⁵⁰, multi-mode fibres⁵¹ and adaptable scanning for tunable excitation regions⁵², achieving a field of view of 100 µm with 95% light uniformity becomes feasible. These solutions can be seamlessly integrated into our system to ensure high reading fidelity for practical applications.

a, Microscope layout for the multi-plane data readout. A 532 nm laser is used for widefield illumination. The fluorescence is collected by an objective lens, either imaged on a complementary metal–oxide–semiconductor (CMOS) for two-dimensional widefield imaging or going through a multi-plane prism, then imaged on the CMOS for 3D widefield imaging. Top inset: data stored in the diamond storage medium. Bottom inset: fluorescence image mapped on the CMOS. b, Image from CMOS fluorescence signal after interlayer denoising. c, Schematic illustration of parallel writing via the TEM01 optical mode. Inset: light intensities distribution of TEM01 optical mode in the x–z plane at the focal point. d, Confocal microscopy imaging of two storage units simultaneously written by the TEM01 mode. Inset: intensities distribution of the TEM01 optical mode in the x–y plane at the focal point.

Full size image

Here we introduce a method to further increase the readout dimension in diamond via 3D multi-layer microscopy⁵³. It is used to augment the number of parallel readout channels (Fig. 4a), by using a multi-plane prism. The multi-plane prism is capable of imaging different layers of objects and reading them out simultaneously. After the interlayer denoising, a piece of three-layer data is restored at once with a fidelity of 99.45% (Fig. 4b). Considering that readout shot noise and storage units maximum count rate (14 million counts per second (Mcps)), an exposure time of 56 µs is appropriate for the optical readout (Supplementary Section IV) and the parallel readout can easily reach more than 200,000 channels⁵⁴, a data-reading rate about 14.4 Gbit s⁻¹ is within reach (Supplementary Section XI).

Meanwhile, parallel writing can be realized with multiple writing beams. As a proof of principle demonstration, a TEM01 (transverse electromagnetic mode, a single bump along y and two bumps along x) optical mode generated by a Hermite–Gauss optical mode converter with two beam spots is applied (Fig. 4c). The spacing of the two spots is around 500 nm, restricted by the optical diffraction limit. As expected, Fig. 4d shows the confocal microscopy imaging of two storage units. By utilizing the state-of-the-art beam array⁵⁵, one can expect 1,000 channels of parallel writing for high-speed data storage. Furthermore, the diamond’s excellent mechanical strength will avoid the deformation and cracking problems of traditional optical discs under high-speed rotation. The ultrashort single-unit exposure time means that traditional high-speed rotation and writing techniques can be naturally integrated. The common repetition frequency of commercial femtosecond pulse lasers is 80 MHz. Position adjustment through rotation requires a linear speed of 40 m s⁻¹ (500 nm × 80 MHz), which corresponds to a rotation speed of approximately 12,000 rpm for the size of the compact disc. During the single-pulse exposure time (1 ps), the disc movement distance is 0.04 nm, which allows single-pulse writing under high-speed rotational writing not to be affected in any way and achieve a writing speed of 240 Mbit s⁻¹. This is beyond the capabilities of other multi-pulse writing schemes. We expect its ultimate writing speed to reach 3 Gbit s⁻¹ (1 GHz × 3 bits), corresponding to the fastest photoelectric switch speed (1 ns) and femtosecond repetition frequency (1 GHz).

Several potential strategies for higher storage density are routine for the diamond storage disc. First, from the super-resolution pattern represented in Fig. 1e, one can find that the morphology of data storage is polarization dependent, which means the polarization multiplexing is feasible for an additional storage domain (2 bits). Second, diamond is a wide-bandgap material with plenty of luminous colour centres resided³³. Previous experiments⁵⁶ have found that weak femtosecond pulses can anneal the local lattice to produce colour centres in other bands. This property enables wavelength multiplexing on a single storage unit by applying appropriate weak laser annealing⁷ (2 bits). Integrating all these techniques, the diamond disc is promising for an ultrahigh storage density of 150 Tbit cm⁻³ (7 bits at single storage units).

In conclusion, we have developed and demonstrated a 4D diamond storage medium using atomic-level lattice structures engineering to meet the pressing requirements for long-term and high-capacity data storage. Our approach is compatible with well-established ODS techniques and could be extended to other substrates with bright fluorescent centres. With the fast development in low-cost synthesis of inch-scale diamond⁵⁷, an important step would be to develop wafer-scale diamond disc for practical applications, opening up the prospect of next-generation ODS.

Experimental set-up and sample information

The diamond-storage-medium data storage is demonstrated on a home-built integrated optical system. The diamond storage medium was a commercial single-crystal diamond plate from Element Six, with a nitrogen concentration of <5 ppb. The diamond storage medium was mounted on a nano stage (Physik Instrumente P-562.3CD) and entirely on a self-made micro stage. A 1.45 NA objective lens (Olympus UPLXAPO100XO) was used for the storage units’ direct writing, excitation and fluorescence collection.

Readout design and methods

GR1 centres, inherent fluorescence defects within diamond, show a zero phonon line close to 741 nm. With excitation sidebands spanning 525 nm to 775 nm, and a fluorescence distribution ranging from 675 nm to 875 nm, GR1 centres offer versatile optical properties. Opting for a 532 nm laser for excitation, we captured the resulting fluorescence and generated a fluorescence intensity map of the target area through laser scanning or widefield imaging techniques. Subsequently, fluorescence intensities were compared against a calibrated threshold in greyscale storage, facilitating the conversion of fluorescence intensities into bit codes, as shown in Fig. 2c,d.

Laser writing and confocal microscope set-up

A pulse picker (Eksma UP2) combined with a titanium:sapphire laser (Coherent Chameleon Ultra II) was used to select a single 808 nm femtosecond pulse for the laser writing. A half-wave plate (LBTEK AHWP10-SNIR) and polarizing beam splitter (LBTEK PBSFR-24-808) controlled the intensity of the femtosecond laser pulse. The wavefront of the femtosecond laser pulse was adjusted by a deformable mirror (Alpao DM97-15) and imaged to the objective lens back focal plane through a 4-f system. The 4-f system matches the clear aperture of the deformable mirror and the objective lens to achieve maximum actual NA. In the parallel write demonstration, we put a mode converter (LBTEK LMC25-780-01) before the objective lens to achieve the TEM01 mode of the writing laser beam. In the confocal microscopy part, a 532 nm laser (CNI Optoelectronics Technology MGL-III-532) for non-resonant excitation of the GR1 centres was coaxial with the writing laser beam. Both beams were focused to the same point by the immersion objective lens. The fluorescence of the GR1 unit is spatially filtered by a 30 µm pinhole and collected by an avalanche photodiode (APD; Excelitas SPCM-AQRH-44-BR1). A series of filters (Semrock BLP01-633R-25 ×1, Semrock FF01-790/SP-25 ×3) were mounted before the APD to exclude laser signals. Fluorescence was coupled into a multi-mode fibre via a flipped mirror and sent to a spectrometer (Ocean Insight QEPRO-FL) for spectral measurement.

Super-resolution microscope set-up

A home-built ‘negative’ ground-state depletion super-resolution microscope⁵⁸ was used to image the topography of storage units. This system utilizes the doughnut intensities modulation and the saturation excitation characteristics of GR1 to achieve super-resolution and show a negative imprint. A high-power 532 nm laser (CNI Optoelectronics Technology OEM-V-532-5W) as an excitation laser was used as a depletion laser, passing through a vortex phase plate (VIAVI U9271E-A44) to form a doughnut shape. In addition, a quarter-wave plate (Thorlabs AQWP10M-580) was used to convert the polarization to circular polarization to ensure the lowest centre power of the doughnut laser. The fluorescence detection was the same as that in the confocal microscope set-up mentioned above.

Widefield microscope set-up

The widefield imaging was performed on a home-built widefield microscope. We used a 532 nm laser (CNI Optoelectronics Technology MGL-III-532) to excite the storage units, and the laser was focused at the back focal plane of the objective lens (Nikon Plan APO DM Lambda 100× Oil). The fluorescence of storage units was collected by the objective lens and imaged to CMOS, Dhyana 400BSI V2 by an eyepiece (Thorlabs ACT508200-B-ML). The excitation laser and fluorescence were separated by a dichroic mirror (Thorlabs DMLP550L). A filter (Edmund 575LP) was used to isolate the excitation laser.

Three-dimensional widefield microscope set-up

A 3D widefield microscope was applied for high-speed multi-plane data readout. The concept of multi-plane readout is based on the correlation between the tube lens and sensor, where the distance between the tube lens and sensor determines the focused sample plane at the detector. The depth of the image can be adjusted by altering the optical path length. By splitting the imaging beam into multiple beams and capturing the resulting images simultaneously, a 3D multi-plane data readout can be obtained. A 532 nm laser (CNI Optoelectronics Technology MGL-III-532 150 mW) was used to excite the storage units and the laser was focused at the back focal plane of the objective lens (Zeiss EC Plan-Neofluar 40×/1.30 Oil M27). To enable rapid 3D image capture, we used an image splitter described in ref. ⁵³. In the light collection path, a custom prism was positioned along the converging beam path of the eyepiece. This prism effectively split the volumetric readout acquisition into eight distinct beams, with four planes on each side, each showing a slightly shifted optical path length. The spacing between consecutive planes is roughly 1 µm, while the distance between adjacent images captured by a camera is approximately 2 µm. Consequently, the axial range covered by the multi-plane imaging is estimated to be around 8 µm and a maximum sample volume measuring about 10 µm × 10 µm × 8 µm. The camera’s frame rate dictates the primary speed constraint.

Heat accumulation evaluation

The impact of the heat accumulation from hundreds of laser pulses on the local area is notable. However, the overall rise in temperature of the diamond storage medium is quite limited. To provide specific dependencies related to thermal diffusion, we disregard the diamond–air boundary condition and carry out analytical calculations. Furthermore, we consider a centrally symmetric Gaussian-shaped initial laser intensity distribution. Under these conditions, we can express the temperature diffusion analytically as follows

where T(r, t) is the temperature at a distance r from the centre of the initial temperature distribution at time t; U₀ and ω₀ are the central energy density and the broadening of initial energy distribution, \({U}_{0}{\omega }_{0}^{3}\) is proportional to the total energy; k_a, c_p and ρ are the energy absorption coefficient, heat capacity and density of the material, respectively. D is diffusion coefficient defined as D = k_c/(c_pρ), where k_c is thermal conductivity. We can find that the thermal diffusion characteristic time is \({\omega }_{0}^{2}/(4D)\).

In the case of parallel writing, we consider the spot to be a larger distribution of writing light. Replace ω₀ in the above formula with N^1/3ω₀, where N is the number of parallel channels. The characteristic times in Fig. 1e are obtained by taking ω₀ to 130 nm and N to 100.

Wavefront correction implementation

Before wavefront correction, the single-pulse energy is adjusted to around 2 nJ to generate plasma luminescence in the diamond storage medium. The wavefront correction process is based on the relationship between the plasma luminescence signals detected by an APD and the focusing tightness of the femtosecond laser. Spatial filtering ensures that only plasma luminescence signals in close proximity to the focal point contribute to the final signal. While the deformable mirror eliminates the optical aberration and improves the focusing tightness of the femtosecond laser, the APD’s signal increases. The gradient descent algorithm is performed by monitoring the plasma luminescence signal and randomly adjusting the surface of the deformable mirror. When the plasma luminescence signal converges to the maximum fluorescence intensity, the wavefront correction process is completed.