We use CoCrPt:Oxide-based bare HDM (BM) from Hitachi Global Storage Technologies (now Western Digital). These comprise multiple layers: seed bottom layer, soft-magnetic under layers (SUL), antiferromagnetic layer (A-FM), intermediate layers, and CoCrPt:Oxide storage layer (Fig. 1a, b). We also use other HDM from Western Digital with ~2.7 nm commercial COC (CMC, Fig. 1c), and ~2.7 nm commercial COC + ~1 nm commercial PFPE lube (CMCL, Fig. 1d) as reference.

The details and nomenclature of the COCs are in Table 1. The structural properties of the FCVA and sputter-deposited COCs can be found in refs. ^50,51. HAMR-compatible FePt-based HDMs are shown in Fig. 1f. See Supplementary Notes 2 and 3 for further details.Table 1

List of samples.

Graphene is grown by chemical vapor deposition (CVD) and placed on the HMD by wet transfer^52,53, as described in the “Methods” section. The quality and uniformity of the samples are assessed by Raman spectroscopy⁵⁴. Unpolarized Raman spectra are recorded at 514.5, as well as at 785 nm, close to the 850 nm used for HAMR writing tests^25,55, with a Renishaw InVia spectrometer equipped with a Leica DM LM microscope and a ×100 objective with a numerical aperture 0.85. A 514.5 nm Raman spectrum of 1LG on Cu before the transfer is shown in Fig. 2a. The photoluminescence (PL) background due to the Cu foil is removed using baseline subtraction⁵⁶. The D to G intensity ratio I(D)/I(G) is <0.1, indicating a defect concentration n_d < 2.4 × 10¹⁰ cm^{−2 54,57–59} (Fig. 2b). The 2D peak position can be fitted with a single Lorentzian with Pos(2D) ~ 2705 cm⁻¹ and full-width-half-maximum, FWHM(2D) ~ 33 cm⁻¹, a signature of 1LG^54,60. The G peak position and FWHM(G) are ~1593 cm⁻¹ and 16 cm⁻¹. I(2D)/I(G) and A(2D)/A(G) are ~2.3 and 4.9.Fig. 2

a 514.5 nm Raman spectra of as-grown 1LG on Cu (before transfer) and 1–4LG transferred on CoCrPt:Oxide after CoCrPt:Oxide background subtraction. b 514.5 nm Raman spectra of CoCrPt:Oxide substrates and 1–4LG transferred onto CoCrPt:Oxide. c, d 785 nm Raman spectra of 1LG on (c) Si/SiO₂ and (d) FePt-based HAMR-HDM, for different laser power densities.

Representative Raman spectra of 1–4LG transferred on HDM are in Fig. 2a, with the HDM spectrum subtracted. The HDM spectrum and the spectra of 1L–4LG on HDM before background subtraction are in Fig. 2b. By subtracting the reference spectrum, using the N₂ Raman peak from the air for normalization, from the spectrum taken on 1LG-coated HDM, we can reveal the 1LG contribution. After transfer, all peaks are downshifted and broader, with Pos(2D)_1LG ~ 2672 cm⁻¹, Pos(2D)_2LG ~ 2681 cm⁻¹, Pos(2D)_3LG ~ 2687 cm⁻¹, Pos(2D)_4LG ~ 2682 cm⁻¹, FWHM(2D)_1LG ~ 61 cm⁻¹, FWHM(2D)_2LG ~ 57 cm⁻¹, FWHM(2D)_3LG ~ 49 cm⁻¹, FWHM(2D)_4LG ~ 53 cm⁻¹. Assuming 2686 cm⁻¹ as unstrained Pos(2D)⁶¹ and a rate of change in Pos(2D) with strain δPos(2D)/ δϵ ~ −64 cm⁻¹/%⁶¹, this would give ~0.22% uniaxial strain in 1LG, consistent with the large FWHM(2D)^61,62. We assume the strain to be mostly uniaxial, because it is unlikely to have biaxial, i.e., perfectly isotropic strain, in all directions, unless this is induced on purpose. There is a small increase in I(D)/I(G) to ~0.2 for 2–4LG on HDM.

Current commercial HDD technology uses a ~2.5–3 nm COC deposited by plasma-based processes^12,63. The thickness and roughness of 1–4LG are measured by Atomic Force Microscopy (AFM), as discussed in Supplementary Note 5. This shows that the thickness of 4LG on HDM is ~2.1 nm (Supplementary Fig. 3), well below that of the current commercial COC. Graphene is transferred without such an energetic process as that used for commercial COC deposition. Thus, it does not adversely affect the magnetic performance. Even considering that COCs slightly affect the HDM magnetic properties, the influence of graphene will be much less than commercial COCs deposited by energetic processes, where energetic atoms/ions (neutrals also) hit and participate in deposition onto the HDM. Reference ⁶⁴ reported that 1LG grown by CVD on Co (and Ni) does not affect the magnetic properties of these materials. In order to confirm this in our case, we perform polar magneto-optical Kerr effect measurements on BM, BM covered with ~2–3 nm COC, and BM covered with 1–4LG. 1–4LG coating results in a small change in the hysteresis loop. The coercivity changes from 510 mT (BM) to ~545 mT (1–4LG and COC). The remanent magnetization is unchanged at ~97–98% of the saturation magnetization when compared to BM. The coercivity and the remanent magnetization of 1–4LG-coated BM are similar to BM+COC, indicating that graphene coating does not have any detrimental effect on the media, see Supplementary Note 6 for details.

HAMR is not possible without laser heating. References ^22,25,65 reported that FePt does not degrade upon laser heating, with lubricant and COC being the main concerns when it comes to laser irradiation. Consistent with refs. ^25,55, here we use an IR laser to examine the 1–4LG irradiation stability. To test whether 1LG can withstand HAMR conditions we consider 1LG transferred on L1₀-FePt-based HDM^25,27,55 and on Si/SiO₂. Reference ⁵⁵ achieved T_c at ~1.3 mW/μm² by optimizing aperture optics. Reference ²⁵ suggested using a laser power density ~10⁷  W/cm² and ~0.35 × 10⁷  W/cm² for FePt-based HDM. We perform Raman measurements at 785 nm, the available Raman wavelength closest to that used in HAMR (~830 nm^25,55), with a spot size ~1.24 μm, as determined by the razor blade technique, see Supplementary Note 7 for details. We vary the power density from ~1.3mW/μm² (0.013 × 10⁷ W/cm²) to ~31.5 mW/μm² (0.315 × 10⁷ W/cm²) in order to examine the laser irradiation-driven evolution of the SLG Raman spectrum under HAMR power densities.

We first consider Si/SiO₂/1LG. We record spectra at different power densities for ~4 min to achieve a good signal-to-noise ratio. Figure 2c shows that 1LG on Si/SiO₂ has no D peak even for the highest power density. Pos(G) downshifts from ~1588 to 1583 cm⁻¹, indicating an increase in T ~ 312.5 K with respect to room temperature, by taking ~−0.016 cm⁻¹/K as a shift of Pos(G) with T⁶⁶. Figure 2d plots the data for 1LG on FePt HDM, with I(G) normalized to that at 1.3  mW/μm². In this case, the FePt substrate is rotated at 4100  rpm on a circular track with a diameter ~4 mm to simulate rotating HDD conditions, with a 20 min acquisition time, much larger than the total laser irradiation time expected for a 5-years life of HAMR-based HDDs²⁹. No D peak is seen for all power densities, thus confirming the stability of 1LG. The Raman data show that no visible changes occur during the irradiation test measurements.

The HDM comprises magnetic/metallic layers, including Co-alloys-based magnetic storage layers^14,20, with a high COF ~0.6–0.8^11,15 and wear^11,15. Therefore, they can experience mechanical damage whenever intermittent contact occurs with the head^11,15, and are susceptible to corrosion^8,11,14 and oxidation^11,15, leading to HDD degradation or damage. The current ~ 2.7 nm commercial COCs have a high COF ~ 0.3–0.5^11,15 and wear in a ball-on-disk tribological environment^11,15, which can result in damage, hence durability concerns, see Supplementary Note 1 for more details.

Reference ¹¹ used FCVA to deposit ~1.5–2 nm carbon films for protection of Co-alloy-based HDM and reported low COF ~0.25, wear, and corrosion. Here, we use 1–4LG grown by CVD for Co-alloy-based and HAMR HDDs.

Since in HDDs the HDM spins during operation, COF and wear need to be examined in a setup mimicking HDD operation. AFM and other tip-based tools are used to measure COF and wear³⁹. In contrast to setups with rotating geometry, tip-based tools measure the COF based on the movement of the tip in the lateral direction³⁹. Ball-on-disk measurements (rotation-based geometry) have a similar assembly as HDDs, with samples rotating while the counterface is in contact with the surface^11,35, thus measuring  COF. We perform ball-on-disk tests using a nano-tribometer (CSM Instruments) in a cleanroom, to have a controlled environment with T = 23 ± 1 °C and a relative humidity ~55 ± 5%. A sapphire ball (Al₂O₃) of diameter ~2.0 ± 0.1 mm and surface roughness ~5.0 ± 0.1 nm is used as the counterface because the hard disk head is made of an Al₂O₃-based composite⁶. During the test, a normal load ~20 mN and a rotational speed ~100 rpm are used, corresponding to a linear speed ~1.05 cm/s for 10,000 cycles.

Since in HDDs the contact occurs occasionally⁶, the 10,000 cycles in our setup are much higher than the HDD operational lifetime. After each test, the wear track and ball images are captured using an optical microscope. To check repeatability, tests are performed 2–7 times. When two surfaces are in contact, and at least one of the surfaces starts to slide with respect to the other, friction and wear occur⁶⁷. As a result, a wear track is formed.

Figure 3 plots representative friction curves for BM, CMC, CMCL, and BML (Fig. 3a) coated with 1–4LG (Fig. 3b), and BM coated by FCVA (Fig. 3c). The average COFs are in Fig. 4, including  HDM coated by DC-sputtering. BM has the highest COF ~ 0.8, with substantial wear, as confirmed by optical images of balls and wear tracks in Figs. 5 and 6. The COF of ~2.7 nm CMC reaches ~0.4 at 10,000 cycles, but with strong fluctuations between ~0.2 and 0.6 at 1500 up to ~3500 cycles (Fig. 3a), and negligibly improves wear with respect to BM (Fig. 3a). The transfer of 1–4LG reduces COF < 0.2 for all samples, and gives non-fluctuating, smooth friction curves. The COF for 1–4LG-coated samples (without lube) is ~4 times lower than BM and ~2 times lower than CMC, despite a reduction of thicknesses ~7 times (for 1LG) to 2 times (for 4LG), using the theoretical 1LG thickness, with respect to CMC. Figures 5b, d and 6b, d reveal that the wear track width of 1–4LG-coated samples is ~2–4 times lower and debris transferred to the ball are smaller than BM and CMC, indicating higher wear resistance. All FCVA COCs with thicknesses from ~0.3 to ~1.8 nm show ~2–5 times higher COF than 1–4LG-coated samples, with and without lube, apart from 1LG without lube. Pulsed DC sputtered COCs have ~2–3 times higher COF than 1–4LG-coated samples.Fig. 3

Representative friction curves for (a) BM, commercial medium with commercial ~2.7 nm COC (CMC), commercial medium with commercial ~2.7 nm COC and ~1.0 nm PFPE lube (CMCL), BM coated with ~1.2 ± 0.2 nm PFPE lube (BML). b HDM coated with 1–4LG. c HDM with FCVA-ta-C of different thicknesses with 0.3 (3CF), 0.6 (6CF), 1.2 (12CF), and 1.8 nm (18CF) FCVA-deposited ta-C. d HDM coated with 1–4LG and ~1.2 ± 0.2 nm PFPE lubricant (L).

To further analyze the wear tracks, Auger Electron Spectroscopy (AES) imaging is performed using a JEOL JAMP Auger Microprobe, see Supplementary Note 4 and Supplementary Fig. 1 for details. Before AES, scanning electron microscope images are taken to select the AES locations. The AES images inside and outside the wear tracks show the carbon-containing sites, and indicate that the amount of carbon on the wear track increases with increasing number of graphene layers, N. The Co and Cr intensities inside the wear tracks are higher for 1LG, and decrease with increasing N, due to the increase in C and the <1–3 nm sampling depth of AES⁶⁸. The O signal in the wear track appears due to ambient oxygen as the samples are exposed to air before AES, with some contribution from the HDM oxide.

After ball-on-disk tests, the wear tracks are analyzed by recording Raman spectra across the wear track at different positions. Reference ⁵⁸ introduced a three-stage model of amorphization. Stage 1: graphene to nanocrystalline graphene. Stage 2: nanocrystalline graphene to low sp³ amorphous carbon. Stage 3: low sp³ to high sp³ amorphous carbon. For all spectra in Figs. 5 and 6, as the D peak increases, D′ and D+D′ appear, whereas the I(2D) weakens when approaching the wear track, indicating an increase in disorder according to stage 1. The broad peak between 500 and 1000 cm⁻¹ is due to the glass substrate⁶⁹. At the center of the wear track, all second-order Raman features (i.e., 2D, 2D′, and D+D′) merge, while I(D) decreases and D,G become broader, indicating an increase of disorder⁵⁸. 2–4LG is less damaged than 1LG, as for the spectra in Fig. 5a, c and I(D)/I(G) in Fig. 7a, from at least 5 positions as a function of COF, without and with lubricant (L). Overall, there seems to be no significant trend of I(D)/I(G) with COF. However, a difference is seen without lubricant: I(D)/I(G) is ~0.95 and ~1 for 3–4LG, while for 1–2LG it is ~1.33 and ~1.25. Thus, there are less defects in 3–4LG compared to 1–2LG. The peaks <1000 cm⁻¹ in Fig. 7b are due to the HDM: those ~470 cm⁻¹ (E_g), 513 cm⁻¹(F_2g), 606 cm⁻¹(F_2g), 673 cm⁻¹ (A_1g) are from Co₃O₄^70,71, while the A_g mode ~550 cm⁻¹ is from Cr₂O₃⁷².Fig. 7

a Mean I(D)/I(G) of 1–4LG-coated samples, with and without lube, after friction tests, as a function of COF. The error bars come from averaging over the intensity ratios of the peaks in the Raman spectra recorded across the wear tracks. b Raman spectrum on a wear track. The region ~400–750 cm⁻¹ shows peaks of the underlying substrate.

Current commercial HDDs use a layer of perfluoropolyether (PFPE) lubricant (lube)¹¹ on top of COCs to further reduce friction and wear, and minimize surface energy. Given the lubricating and corrosion protection properties of 1LG itself, lube may not be needed for 1LG. However, exploring the compatibility of 1LG with lube is useful from the fundamental viewpoint.

Friction measurements on lube-coated samples (Fig. 3) show that in certain cases the COF is higher with L, than without. Taking the error bars into account, a significant difference in COF is only observed for 1LG and 1LGL. 1LG has wetting transparency⁷³, i.e., it is thin enough that its introduction does not affect the wettability of an underlying substrate. Without overcoat, the L containing HDM, as in BML, shows very high COF ~ 0.55 and wear, due to the metal (medium) L-induced catalysis⁷⁴. 1LG is not enough to avoid the interaction of L with medium, hence higher COF ~ 0.45 and wear is observed in 1LGL. Without L, 1LG shows very low COF ~ 0.15 and wear. A minor difference in case of 2LG versus 2LGL can be seen. However, for 3LG versus 3LGL and 4LG versus 4LGL the difference in average COF is marginal. L on 1LG results in higher and inconsistent COF and wear, as some measurements show lower COF ~ 0.15, some higher ~0.78, with COF increasing after few thousands cycles. The COF of 1LG, on an average, is ~0.45, i.e., ~3 times its non-lubricated counterpart. Thus, COFs of 2–4LG are more or less similar to those of the non-lubricated counterparts, suggesting that 2–4LG without L are lubricious enough, and that L does not improve lubricity. The D and G peaks in the Raman measurements in Figs. 5 and 6 on 1–4LG confirm the presence of carbon on the wear tracks, and transfer of debris to the balls, similar to the non-lubricated counterparts.

Co-alloys have a great propensity to corrode, mainly due to Co oxidation⁷⁵. This results in the loss of magnetic properties⁷⁵, hence this is one of the major concerns for the long-term functionality and durability of HDDs^14,51,75. To examine the corrosion protection efficiency of 1–4LG and compare their performance with state-of-the-art COCs, the corrosion of different uncoated- and coated-HDM, exposed to an electrolyte solution ~0.1 M NaCl similar to that used in refs. ⁷⁵ on a ~0.24 cm² area, is investigated using an electrochemical corrosion method^51,75. The measurements are performed with a 3-electrode setup with a Pt wire as counter electrode, Ag/AgCl as reference electrode and HDM as working electrode, to which the potential is applied⁵¹. Each test consists of anodic and cathodic sweeps, where the potential is varied and the corresponding current measured⁵¹. Every sweep is conducted at different locations, with at least 3 sets of 6 sweeps on each sample. The so-called Tafel’s analysis⁷⁵ is done by plotting anodic and cathodic curves on a semi-logarithmic scale of potential versus log current. The linear part of the logarithmic anodic and cathodic currents are extrapolated⁷⁵, and the intercept of these lines gives the corrosion current⁷⁵, and the corrosion current density J_corr when divided by the contact area.

The corrosion protection efficiency (CPE) defines how efficient the COC is in protecting against HDM corrosion. This is defined as⁵¹:1CPE(%)=Jcorr0−JcorrJcorr0×100,\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{CPE}}\,( \% )=\frac{{J}_{\mathrm{corr}}^{0}-{J}_{\mathrm{corr}}}{{J}_{\mathrm{corr}}^{0}}\times 100,$$\end{document}where Jcorr0\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${J}_{\,\text{corr}\,}^{0}$$\end{document} is the BM corrosion current density and J_corr that of coated HDM.

When a metal is exposed to a corrosive solution, it releases ions that leave behind electrons, which can be observed in an anodic reaction as⁷⁵:2M→Mn++ne−,\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{M}}\to {\mathrm{M}}^{n+}+n{e}^{-},$$\end{document}where M represents the metal and n the number of electrons released by it. For Co-alloy-based systems, the metal dissolution, which decreases the anode conductivity⁷⁵, can be written as⁷⁵:3Co→Co2++2e−,\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{Co}}\to {\mathrm{Co}}^{2+}+2{e}^{-},$$\end{document}As the corrosion reaction involves the transfer of electrons and ions between metal and solution⁷⁵, the corrosion rate varies with corrosion current⁷⁵, hence J_corr varies inversely with corrosion resistance, i.e., the COC ability to reduce the HDM corrosion^51,75.

The BM shows the highest J_corr, indicating the greatest propensity to corrode (Fig. 8). Introduction of 1–4LG reduces HDM corrosion. J_corr decreases with N, from ~5.3 nA/cm² for 2LG to ~4.4 nA/cm² for 4LG, Fig. 8, and CPE increases. The COC defects and pinholes are the active corrosion sites⁷⁵, and their electrical conductivity could further add galvanic corrosion⁷⁵. The COC should be defect- and pinhole-free, smooth, and with excellent barrier properties to minimize HDM oxidation and corrosion. Reference ⁴³ showed that 1LG can be used as a corrosion barrier for metallic surfaces, and suggested that this should be uniform and defect-free to achieve corrosion protection. In our case, 1LG reduces J_corr by ~2.5 times with respect to BM, but it is twice that of BM with ~2.7 nm COC. For 2LG, J_corr is ~3.8 times smaller than BM, and marginally higher than COC. This is remarkable as the 2LG thickness ~0.7 nm, is ~4 times lower than ~2.7-nm-thick COC. Beyond 2LG, J_corr and CPE remain similar, adding marginal anti-corrosion improvement (Fig. 8). The higher corrosion protection in 2–4LG with respect to 1LG is mainly due to the fact that the increase in N improves the HDM coverage, leading to the reduction of active corrosion sites. 2–4LG display corrosion protection similar to COC, attributed to better barrier properties. Figure 8 indicates that, for a similar thickness, graphene-based overcoats have lower J_corr than amorphous COCs, indicating greater protection. Figure 8 shows that the corrosion resistance performance of 1LG is ~2.5 times better than BM, but ~2 times higher than COCs. By adding graphene layers the corrosion resistance is improved. The difference in corrosion current density between COC and 2–4LG is 30–10%. Thus, the corrosion resistance of 2–4LG approaches that of COCs. Taking into account that the theoretical thickness of 1LG is ~8–9 times lower than state of the art COCs, and that a COC having the same thickness as 1LG would provide no protection and be full of holes¹⁹, 1LG comfortably beats any COC of the same thickness (i.e., same storage density).Fig. 8

Corrosion current density J_corr and corrosion protection efficiency (CPE) for different samples. The corrosion resistance varies inversely with J_corr⁵¹. 2–4LG have similar CPE to ~2.7 nm CMC and ~1.8 nm COC in sample 18CF. The dotted line is a guide to the eye for J_corr of different samples. The arrow compares various 1–4LG-COCs pairs, such as 1LG → 3CF, 2LG → 8CS, 3LG → 12CS, and 4LG → 18CF, as the 1–4LG-based sample in each pair is always slightly thinner than the amorphous COC. The red shaded region highlights the bare media and media with a commercial overcoat, the yellow region highlights media with graphene overcoats and the blue region highlights media with FCVA-deposited ta-C. The horizontal red dashed line is a reference at J_corr = 4 nA/cm². The error bars stem from taking the average of multiple repeated measurements.

1LG is grown by CVD on a 35 μm Cu foil. The substrate is loaded into a hot wall tube furnace evacuated to ~1 mTorr. The Cu foil is annealed in hydrogen at 1000 °C for 30 min. This reduces the copper oxide surface and increases the Cu grain size. The growth process starts when 5 sccm CH₄ is added to H₂. After 30 min the substrate is cooled down for several hours in vacuum (~1 mTorr) to room temperature and unloaded.

We use a polymethyl methacrylate (PMMA)-based wet transfer^52,53. First, ~500 nm PMMA is spin-coated on the sample. The PMMA/1LG/Cu stack is then placed in an aqueous solution of ammonium persulfate to etch Cu⁵². When Cu is fully etched, the graphene/PMMA-stack is placed into a de-ionized (DI) water bath to rinse any acid residuals and subsequently fished out using the HDM substrate. After drying for one day at room temperature, the sample is placed in acetone to remove PMMA, leaving 1LG on the HDM. By repeating the steps described above, several 1LGs are transferred to create an MLG stack. The same procedure is used to place 1LG and MLG onto FePt.