Homoepitaxial Growth of Single Crystal Diamond Membranes for Quantum Information Processing



original image

Homoepitaxial growth of single crystal diamond membranes is demonstrated employing a microwave plasma chemical vapor deposition technique. The membranes possess excellent structural, optical, and spin properties, which make them suitable for fabrication of optical microcavities for applications in quantum information processing, photonics, spintronics, and sensing.

Fabrication of devices designed to fully harness the unique properties of quantum mechanics through their coupling to quantum bits (qubits) is a prominent goal in the field of quantum information processing (QIP). Among various qubit candidates,1 nitrogen vacancy (NV) centers in diamond have recently emerged as an outstanding platform for room temperature QIP.2–5 However, formidable challenges still remain in processing diamond and in the fabrication of thin diamond membranes, which are necessary for planar photonic device engineering. Here we demonstrate epitaxial growth of single crystal diamond membranes using a conventional microwave chemical vapor deposition (CVD) technique.6 The grown membranes, only a few hundred nanometers thick, show bright luminescence, excellent Raman signature and good NV center electronic spin coherence times. Microdisk cavities fabricated from these membranes exhibit quality factors of up to 3000, overlapping with NV center emission. Our methodology offers a scalable approach for diamond device fabrication for photonics, spintronics, optomechanics and sensing applications.

The exceptional properties of diamond, including a wide electronic and optical band gap, excellent thermal conductivity, and its biocompatibility are attractive for numerous optoelectronic and sensing applications.6 One diamond color center in particular, the negatively charged nitrogen vacancy (NV) center, is among the most promising solid state qubits due to its room temperature operation, long spin coherence time, and suitability for optical initialization and readout.7 Embedding color centers within a single crystal diamond membrane enables promising applications, including diamond-based QIP.8, 9 To this end, optically–active thin membranes are essential, as they are the fundamental building blocks of photonic components such as microcavities, photonic crystal cavities, resonators and waveguides.

Fabrication of planar devices often requires the material be undercut to achieve optical, electrical or mechanical isolation, as in the formation of mechanically-sensitive cantilever devices, optical microdisk cavities, or photonic crystal cavities.10–12 Such structures are often formed by incorporating a sacrificial layer into the initial bulk material structure. Selective etch removal of the sacrificial layer can then create an undercut structure, or a larger membrane with thickness of tens to hundreds of nanometers, depending on the device application. However, such an approach is not available for creating undercut or membrane structures in diamond, as high quality hetero-epitaxy of single crystal diamond on dissimilar substrates has not yet been demonstrated due to lattice mismatches and low surface diffusion of carbon atoms.6, 13, 14

An alternative technique has been employed for diamond: high energy (∼1–3 MeV) and high dose ion implantation (∼1 × 1017 ions/cm2) generates a thin amorphous layer below the diamond surface that can be selectively etched leaving behind a diamond membrane several hundred nanometers in thickness.15–21 Nevertheless, poor spin coherence properties, weak luminescence from its color centers, and a residual built-in strain as a result of the high ion damage significantly limit the performance of diamond membranes fabricated in this manner. Indeed, no working optical cavities, spin measurements or devices have been demonstrated from ion-damaged diamond membranes to date.

To address this challenge, we introduce an approach that utilizes the diamond membranes as templates for the epitaxial overgrowth of a thin layer of single crystal diamond. Growth is carried out in a plasma-enhanced chemical vapor deposition (PECVD) reactor and results in single crystal diamond layers with superior physical, optical, and spin properties than their original diamond templates. Furthermore, the overgrowth process allows us to intentionally dope the diamond device layer with color centers such as NV or silicon-vacancy (SiV) centers that were not present in the original diamond template. Indeed, strong luminescence, narrow Raman peaks and the appearance of NV center spin lifetimes in the membrane, are characteristic of a good quality bulk diamond. After growth, the diamond template can be completely etch-removed, leaving the highest quality overgrown diamond membrane to be further formed into device structures (see experimental methods). The schematic of the process is depicted in Figure1a. The specific homoepitaxial growth conditions were: microwave power 950 W, pressure 60 Torr under 400 standard cubic centimeters per minute of 99:1 CH4/H2. We note that our growth conditions are not typical of those used for homoepitaxial, single crystal diamond growth, which often requires higher microwave power densities (∼2–4 kW) and higher pressure (∼150–250 Torr).22, 23

Figure 1.

(a) Schematic illustration of the regrowth process of the diamond membrane using a PECVD reactor. (b) SEM side view and (c) top view of the original diamond membrane before the regrowth. (d) SEM side view and (e) top view showing a ∼300 nm overgrowth of a single crystal diamond. The pits observed on the surface are transferred from the original membrane. (f) AFM scan of a 5 μm strip demonstrating the smoothness of the top membrane surface. The surface roughness is measured to be 4 nm.

Figure 1b,c show the top view and the cross sectional scanning electron microscope (SEM) images of the original membrane, 1.7 μm in thickness. Figure 1d,e illustrate the overgrown material after 5 minutes of growth. The single crystal material, with a top (100) facet, is clearly observed on top of the original membrane. No indications of polycrystalline material, grain boundaries or surface defects (e.g., hillocks) are observed. The pits in the regrown layer are transferred from the original membrane, which was pitted. In principle, this can be avoided through the use of smooth, dislocation free, single crystal diamond as a starting material. Importantly, the undamaged regions of the regrown material are very smooth. Atomic force microscope (AFM) scans reveal that the root mean square (RMS) surface roughness is only 4 nm (Figure 1d). In comparison, the RMS surface roughness of a typical commercially available CVD single crystal diamond (Element Six, Inc) is on the order of 2–3 nm.

Raman spectroscopy can provide detailed information about the structural quality of a material (see experimental methods). The best diamond templates show Raman peaks at 1331 cm−1 with a FWHM of 9.9 cm−1 (after removing the most damaged region but before any annealing or regrowth treatment). For comparison, the Raman peak we measure for bulk single crystal diamond is centered at 1333.5 cm−1 with a FWHM of 2.3 cm−1. Remarkably, the Raman signature of the composite overgrown membrane-plus-template gave a peak value of 1332 cm−1 with a FWHM of 6.6 cm−1. These values are a convolution of the Raman peak from the overgrown material with that of the membrane template, and accounts for the best fit to two peaks, shown in Figure2a.

Figure 2.

(a) Diamond membrane grown on a lifted-off diamond slab with the damaged side etched away. The double Voigt fitting of the Raman line (red curve) reveals a FWHM of ∼4 cm−1 for the regrown material (blue curve) and ∼7 cm−1 for the original membrane (green curve). (b) Diamond membrane growth on the damaged side of the lifted-off membrane. Despite the damaged original material with a wide Raman line (green curve, FWHM of ∼17 cm−1), the regrown material shows a much narrower FWHM of ∼3 cm−1 for the regrown material (blue curve).

To further gauge the influence of the template quality on the properties of the overgrown diamond, material was overgrown on the most heavily-damaged surface of the template (which was not removed by etching) (Figure 2b).19 Even with a heavily damaged template material, the epitaxially grown material exhibits superior Raman characteristics with linewidth of ∼ 3 cm−1, approaching the quality of a bulk single crystal diamond, despite the shifted and much broader Raman signature of the template, as seen in the green curve of Figure 2b. This is convincing evidence that the overgrown diamond film has much improved structural properties over its highly-damaged, highly-strained diamond template. This is contrary to conventional epitaxial growth of other semiconductors (e.g., GaAs) where the strain in the template is manifested in the properties of the overgrown material and often reduces its quality.24 However, strain fields do exist in our grown material, as is evident by the optically detected magnetic resonance (ODMR) measurements (shown below).

To further investigate the optical properties of the regrown material, photoluminescence (PL) measurements were recorded under 532 nm excitation at room temperature. The regrown membranes showed bright fluorescence even though the original templates were not optically active. The appearance of PL clearly indicates that the formation of color centers occurs during the CVD growth. We have demonstrated this by engineering two types of different membranes–one with primarily NV centers and the other with SiV centers. The PL of these membranes is presented in Figure3. The impurity incorporation can be controlled by placing the template on various substrates, e.g., silicon or other bulk diamond. We believe that during the growth the plasma slightly etches the substrate, whose elements are subsequently incorporated into the growing diamond lattice.25 No nitrogen gas was intentionally introduced during the growth; the nitrogen source is due to residual nitrogen in the chamber.

Figure 3.

(a) PL spectrum recorded from an overgrown membrane, with dominating NV concentration. (b) PL spectrum recorded from a different overgrown membrane, showing primarily the SiV defects.

Some of the most promising scientific and technological applications of diamond are spin-based QIP algorithms3, 4 and detection of weak magnetic fields with high sensitivity.26–29 For these, an important prerequisite is characterizing the coherent dynamics of the NV center electron spin. We note that no electron spin resonance signal was detected from the original diamond templates prior to the regrowth process. After the epitaxial regrowth of membranes containing NV centers, an ODMR signal was observed, motivating further investigation of the NV center spin properties within the grown membranes.

Figure4 plots various measures of ensemble spin coherence as functions of time at room temperature and with no applied magnetic field. More details regarding the timing sequences to determine these measures can be found in the Supporting Information. In Figure 4a, the NV center longitudinal spin coherence as a function of time is plotted. Fitting to an exponential decay reveals a spin-lattice relaxation time of T1 = 5 ms for the NV center ensemble. The transverse homogeneous spin coherence, plotted in Figure 4b, was measured with a Hahn echo sequence and was found to have a value of T2 = 3.5 μs when fit to a single exponential decay. Figure 4c plots the transverse inhomogeneous spin coherence, taken with a Ramsey sequence (off resonance by ∼25 MHz). The pronounced beating in the Ramsey data suggests several distinct spin resonance frequencies which may be due to crystal strain energetically splitting the SX and SY spin sublevels as well as a small stray magnetic field. A fit consisting of a single exponential decay envelope around four beating frequencies reveals T2* = 158 ns for this ensemble of NV centers. It should be noted that transverse spin coherence is sensitive to the surrounding spin bath, originating from the presence of other electron and nuclear spins as well as magnetic fluctuations on the surface. The measured values of T2 and T2* for the regrown material are comparable with both natural NVs in type Ib diamond28 and shallow implanted NVs in a high purity bulk diamond30, 31 (T2 ∼1–20 μs) and are higher than typical coherence times found in nanodiamonds (T2 ∼1 μs).32 Nevertheless, it is still shorter than the isotopically pure single crystal diamond (T2 ∼ 2 ms).7

Figure 4.

(a) Spin lattice relaxation time, yielding T1 ∼ 5 ms. The red curve is a single exponential fit of the data. (b) Hahn-echo sequence of the transverse homogeneous spin coherence time resulting with a T2 ∼ 3.5 μs. The red curve is a single exponential fit of the data. (c) Ramsey sequence to determine transverse inhomogeneous spin coherence precession decay time, yielding T2* = 158 ns. Error bars are smaller than data markers. The sequences applied to measure the spin properties are schematically depicted. All the data are recorded at room temperature with no applied magnetic field.

Remarkably, good spin behavior exists in the regrown material despite the fact that the original template did not show any ODMR signal and the regrown membrane is only a few hundred nanometers thick. The spin properties of the regrown membrane further confirm the good quality of the thin, single crystal diamond membrane and indicate promise for applications in nano-magnetometry and QIP.

The spin properties of the regrown diamond membranes can be significantly improved by growing material in a purer environment, resulting in fewer paramagnetic defects such as substitutional nitrogen. Additionally, growing with 12C isotopically purified methane has the promise to achieve millisecond T2 coherence times, as the nuclear spin of 13C can significantly reduce the NV spin coherence.7 Furthermore, additional experimental steps such as high temperature annealing31 may also improve the coherence times of NV centers in the grown material.

Having at our disposal an optically-active, thin, single crystal diamond membrane, we demonstrate its suitability for QIP by fabricating an optical microcavity. Figure5a shows PL spectra recorded from a microdisk (black curve) and the substrate (red curve) at room temperature. The NV zero phonon line (ZPL) is visible at 637 nm. The peaks decorating the PL recorded from the microdisk are the modes with quality factors (Q) ∼1500. Q's of up to 3000 were measured from the microdisk cavities. Figure 5b shows a typical SEM image of a microdisk cavity (diameter of 2.5 μm and a thickness of ∼800 nm), fabricated from the epitaxially-grown membrane.

Figure 5.

(a) PL spectra recorded from the diamond membrane (red curve) and from a 1 μm microdisk (black curve, shifted for clarity) under 532 nm excitation at room temperature. The ZPL of the NV center is clearly seen at 637 nm. The spectrum from the microdisk is decorated with modes with typical Q ∼1500. (b) A typical SEM image of a microdisk cavity with a diameter of 2.5 mm and a thickness of ∼800 nm.

In conclusion, we have described a process that produces thin, optically-active, high-quality single crystal diamond membranes. The regrowth results are intriguing for a number of reasons. First, an improved Raman signature indicates a much smaller strain distribution in the regrown membranes. This is despite the fact that the original templates are structurally imperfect with high strain fields. Second, the regrown material exhibits good spin properties, unlike the original template which did not exhibit any spin signature. Third, the regrowth methodology allows engineering of other color centers within the regrown material. Further exploration of the growth conditions could lead towards thin single crystal diamond membranes with a low distribution of color centers that can emit single photons on demand (e.g., NV, SiV, Cr) – an important prerequisite for single photon device engineering.

The successful homoepitaxial growth of thin diamond membranes is a pivotal step in advancing diamond processing. Our results provide important insights that can potentially further the understanding of the complicated diamond CVD growth, particularly its nucleation steps. The demonstration of good quality epitaxial diamond growth even on damaged diamond template opens new avenues towards engineering of thin, single diamond membranes. The scalability of this process can propel the use of diamond from laboratory setting into a plethora of devices spanning high power electronics, photonics, sensing and QIP.

Experimental Section

Generation of initial diamond templates: A type IIA CVD diamond crystal (Element Six) was implanted with He ions (1 MeV, 5 × 1016 He/cm2) to create a damage layer and annealed for one hour at 900 °C.19 The sample was then immersed in water, and using electrochemical etching carried out under constant voltage, the membranes were lifted off and placed on different substrates. The original templates have (100) crystallographic orientation, same as the original type IIA diamond crystal.

Regrowth process– the original single crystal diamond membranes were first cleaned using a mixed (1:1:1 sulfuric–perchloric–nitric acid) and put into a Seki Technotron AX5010-INT PECVD reactor. The chamber was evacuated below 0.1 mTorr and flushed with H2 gas. After the plasma ignition (only hydrogen ambient), it took approximately 8 minutes for the temperature and the pressure to stabilize, at which time the methane was introduced. The homoepitaxial growth conditions were: microwave power 950 W, pressure 60 Torr, 400 SCCM of 99:1 CH4 (99.999%)/H2(99.999%) for 5 minutes. No external heating source was used and the temperature was ∼850 °C as read by a pyrometer.

Device fabrication: the regrown membranes were flipped and thinned down using oxygen inductive coupled plasma reactive ion etching (ICP-RIE) to their final thickness of ∼500 nm. A silicon dioxide hard mask was deposited on top of the membranes and the microdisk cavities were patterned using e-beam lithography and etched down using ICP-RIE.

Spectroscopy: the Raman data was collected using 532 nm laser excitation with a conventional confocal Raman microscope (LabRAM ARAMIS, Horiba Jobin-Yvon) at room temperature. The typical spatial resolution of the microscope is ∼1 μm.

The PL measurements were recorded using a custom built confocal microscope using a 532 nm continuous wave (CW) diode laser through a 100×, 0.9 numerical aperture objective. The emission was collected through the same objective and directed into a spectrometer. The laser light was filtered using a dichroic mirror.

Spin coherence measurements: The samples were placed in a confocal microscopy setup with a 100× objective where a 532 nm laser for photoexcitation, microwave magnetic fields for spin manipulation, and the sample PL measured by an avalanche photodiode were all gated in time for time-domain spin coherence measurements. Timing sequences for the measurements of T1, T2, and T2* (see Supporting Information) were generated by an arbitrary waveform generator.

Supporting Information

Supporting Information is available from the Wiley Online Library or from the author.


The authors thank Prof David Clarke for the access to the Raman facilities, F.J. Heremans and A.L. Falk for the assistance with the optical measurements, M. Huang for the assistance with ion implantation. The authors also acknowledge A. Yacoby, M. Grinolds, S. Hong and P. Maletinsky for the assisting with the initial ESR measurements and thank T.M. Babinec for useful discussions. The financial support from DARPA, QuEST and AFOSR is gratefully acknowledged.