Using both faces of polar semiconductor wafers for functional devices

Single crystals of semiconductors are macroscopic molecules with periodic repeating basis units⁷. In a perfect semiconductor crystal wafer, the atomic location and type at one point in space repeats itself hundreds of microns from one surface to the other and hundreds of millimetres from one edge to the other. In non-polar semiconductors such as silicon, the crystal orientation is chosen carefully to maximize the electronic performance of devices on the surface. Electronic devices such as transistors are made only on one (say, the top) surface of silicon and not the other. Even if the other surface was used, the parallel surfaces of a cubic crystal are identical by symmetry. This means that flipping a silicon wafer upside down does not change the chemical or electronic properties of the top surface. Thus, heterostructures and electronic devices made on the two surfaces in an identical manner exhibit identical properties.

The wide-bandgap semiconductors GaN and aluminium nitride (AlN) have a wurtzite crystal structure with an underlying hexagonal lattice⁸. This crystal structure breaks inversion symmetry along the [0001] orientation (the c axis). The two surfaces perpendicular to the c axis of a single-crystal wafer of these polar semiconductors therefore exhibit very different physical properties: flipping the crystal wafer is analogous to flipping a bar magnet. The chemical properties of the two sides are so distinct that they are used to identify metal or nitrogen surface polarity. Although the metal-polar surface is inert to most chemicals, the N-polar surface of both GaN and AlN etches vigorously in solutions with bases, such as KOH and TMAH, or acids such as H₃PO₄ (refs. ^9,10).

Large differences in the electronic properties of heterostructures are observed for the two faces^11,12. For example, if an approximately 10-nm-thin coherently strained epitaxial GaN layer is deposited on the N-polar surface of AlN, a two-dimensional electron gas (2DEG) is formed at the GaN/AlN heterojunction quantum well as a result of the discontinuity of the conduction band combined with the discontinuity of the electronic polarization across the heterojunction¹³. But when an identical, approximately 10-nm coherently strained epitaxial GaN layer is deposited on the metal-polar surface of AlN, a 2D hole gas is formed at the heterojunction quantum well¹⁴. These polarization-induced, electrically conductive 2D electron and hole gases are formed in crystals nominally free of chemical impurities such as donor or acceptor dopants. (Al,Ga)N/GaN HEMTs using such polarization-induced conductive channels demonstrate outstanding performance in high-power and high-speed applications^15,16,17,18.

So far, only a single face of c-axis-oriented GaN single-crystal wafers is used for either photonic or electronic devices^19,20,21. In this work, the two polarities on the opposite faces of a GaN single crystal are combined to realize a photonic device on one side and an electronic device on the other. A heterostructure quantum well on the N-polar side is used to generate a polarization-induced high-mobility 2DEG and a quantum well (In,Ga,Al)N p–n diode heterostructure is realized on the opposite metal-polar side. This two-sided wafer is then processed, first into HEMTs on the N-polar side, followed by blue quantum well LEDs on the metal-polar side. Successful operation of the HEMT and LED devices is observed, allowing the switching and modulation of blue LEDs by HEMTs on the other side of the wafer.

Figure 1 shows how dual-side epitaxy for dualtronics was achieved. The starting wafer is a high-transparency bulk n-type O-doped GaN substrate grown by the ammonothermal method^22,23 with mobile electron concentration of about 10¹⁸ cm^−3 and threading dislocation density of about 10⁴ cm^−2 with both the Ga-polar and N-polar sides chemo-mechanically polished to obtain atomically smooth surfaces. The N-polar surface of the GaN substrate was gallium-bonded to a GaN/sapphire carrier wafer as shown in Fig. 1a. Because the surface bonded to the carrier wafer needs to be used again, we chose gallium bonding to the GaN surface of a GaN/sapphire carrier wafer to ensure desired elements (Ga, N) at this bonding interface. Using molecular-beam epitaxy (MBE), we first grew a blue (In,Ga,Al)N LED structure consisting of a quantum well active region inside a p–n heterojunction on the Ga-polar face of GaN. The wafer was then detached from the carrier wafer and flipped as shown in Fig. 1b. The Ga was removed from the N-polar side and the metal-polar p-type GaN surface of the LED side was gallium-bonded on to a clean GaN/sapphire carrier wafer. Subsequently, a N-polar GaN/AlGaN/GaN heterostructure was grown on the N-polar face again by MBE to create a high-electron-mobility 2DEG to serve as the conduction channel of the HEMT, as shown in Fig. 1c. The chosen growth conditions for the LED and HEMT structures are described in Methods.

The grey arrows indicate the chronological order of the growth procedure. a, Ga bonding of the GaN substrate and MBE growth of the (In,Ga,Al)N LED along the metal-polar direction. b, Unloading, cleaning, flipping, Ga bonding and reloading of the sample. c, MBE growth of the GaN/AlGaN HEMT along the N-polar direction.

Figure 2a shows the precise layer thicknesses and compositions of the HEMT on the N-polar side on the top and the LED on the metal-polar side on the bottom of the single GaN wafer achieved after the dual-side epitaxy. Figure 2b–g shows the corresponding scanning transmission electron microscopy (STEM) images.

a, Schematic showing the HEMT-LED heterostructures grown on both faces of a single-crystal c-plane n-GaN substrate. b, HAADF-STEM image showing the GaN/Al_0.40Ga_0.60N/GaN HEMT. Scale bar, 10 nm. c, Atomic-resolution image corresponding to the uppermost GaN/Al_0.40Ga_0.60N heterojunction interface that hosts the 2DEG. Scale bar, 2 nm. d, iDPC image in the uppermost GaN layer of the HEMT indicating that the nitrogen polarity follows that of the substrate to the surface. Scale bar, 1 nm. e, HAADF-STEM image showing the LED quantum well, electron blocking layer, cladding and contact layers. Scale bar, 100 nm. f, Atomic-resolution image corresponding to the LED In_0.07Ga_0.93N/In_0.17Ga_0.83N/In_0.07Ga_0.93N single quantum well. Scale bar, 2 nm. g, iDPC image of the p-InGaN contact layers of the LED indicating that the metal polarity follows that of the substrate. Scale bar, 1 nm.

The MBE-grown HEMT on the N-polar side started with the growth of a 500-nm-thick unintentionally doped (UID) GaN buffer layer as shown in Fig. 2a, followed by 5-nm GaN:Si and 5-nm Al_0.40Ga_0.60N:Si layers with a Si concentration of 3 × 10¹⁸ cm^−3 capped with undoped 15-nm Al_0.40Ga_0.60N and 10-nm GaN. Silicon doping below the 15-nm AlGaN layer was used to prevent the formation of a polarization-induced 2D hole gas at the bottom N-polar AlGaN/GaN interface²⁴. The bulk energy bandgap of GaN is E_g1 = 3.4 eV and that of Al_0.40Ga_0.60N is E_g2 = 4.5 eV at room temperature. For the N-polar HEMT, the high-angle annular dark-field (HAADF) STEM image in Fig. 2b shows the 20-nm AlGaN layer near the surface with sharp interfaces with GaN on both sides. Figure 2c zooms in to show the lattice image of the top GaN/AlGaN heterojunction, in which the 2DEG resides mostly on the GaN side. A sharp heterointerface between GaN and AlGaN is observed, with the atomic planes clearly resolved. Figure 2d shows an integrated differential phase contrast (iDPC) image of the GaN HEMT channel layer, proving the N-polar crystal structure along the growth direction: the N atoms located vertically below the brighter Ga atoms.

For a N-polar heterostructure, a mobile 2DEG is expected to form on the GaN side inside a triangular quantum well formed by the conduction band offset ΔE_c ≈ 0.8 eV at the sharp GaN/AlGaN heterojunction seen in Fig. 2c. A Hall-effect transport measurement described in Methods showed that a 2DEG is formed with room-temperature electron density n_s ≈ 1.26 × 10¹³â€‰cm^−2 and electron mobility of μ ≈ 1,970 cm²â€‰(V s)^−1, resulting in a sheet resistance of R_sh = 1/(en_sμ) ≈ 252 Ω sq^−1, in which e is the magnitude of the electron charge. The observed sheet resistance is the lowest and the electron mobility is one of the highest among all N-polar HEMT heterostructures reported in the literature, resulting largely from low interface roughness scattering of the quantum-confined electrons at the sharp interface owing to the high-quality epitaxy on the double-side polished bulk GaN substrate^25,26,27. A benchmark comparison with state-of-the-art N-polar III-nitride HEMTs reported in the literature is shown in Extended Data Fig. 1.

The MBE-grown LED heterostructure on the metal-polar side seen on the bottom side of Fig. 2a consists of a 150-nm GaN:Si buffer layer, followed by a 50-nm In_0.07Ga_0.93N/40-nm In_0.07Ga_0.93N active region with a single 2.5-nm-thick In_0.17Ga_0.83N quantum well embedded in between.

This photonic active region was followed by p-type doped layers starting with a 20-nm Al_0.09Ga_0.91N electron blocking layer with a Mg concentration of 3 × 10¹⁹ cm^−3, followed by a 150-nm-thick GaN layer with a Mg concentration of 6 × 10¹⁸ cm^−3. This LED heterostructure was capped with p-InGaN contact layers consisting of 40-nm-thick In_0.02Ga_0.98N and 5-nm-thick In_0.15Ga_0.85N doped with Mg at the levels of 1 × 10²⁰ cm^−3 and 5 × 10²⁰ cm^−3, respectively, to form low-resistance ohmic contacts for hole injection into the LED active region, while simultaneously being transparent to the light emitted from the active region. Figure 2e shows a HAADF-STEM image of each of these LED layers, indicating a uniform quantum well and the surrounding active region along the plane of the wafer. The atomically resolved image in Fig. 2f shows the 2.5-nm-thick In_0.17Ga_0.83N quantum well and the InGaN cladding layers. The group III element fractions and doping profiles for both the LED and HEMT structures are presented in the time-of-flight secondary ion mass spectrometry (ToF-SIMS) data shown in Extended Data Fig. 2, which indicate that the measured values are as intended. Notably, there is no noticeable interdiffusion of dopants or group III broadening of the metal-polar LED layers, indicating that interface and doping control is not affected by the second growth step.

Figure 2g shows an iDPC image of the p-InGaN contact layers on the metal-polar LED side. The LED is indeed metal-polar viewed in the growth direction with the substrate out of view above. This confirms that the polarity of the crystal substrate locks the polarity of the epitaxial layers and that the crystalline registry goes through the entirety of the GaN wafer over hundreds of microns. This is why the 2DEG forms on the N-polar side of the wafer, whereas the metal polarity is retained throughout the active regions, quantum well and electron blocking and contact layers of the LED. The high-crystalline perfection of the LED heterostructure also proves that the growth of the HEMT heterostructure on the N-polar side at a high temperature and with nitrogen plasma does not destroy or degrade the LED structure during the rather harsh thermal and chemical conditions. The surface of the as-grown heterostructures after the completion of both epitaxial steps as shown in Extended Data Fig. 3 exhibit sub-nanometre roughness atomic steps characteristic of step-flow growth, indicating the successful dual-side epitaxy. X-ray diffraction (XRD) measurements indicate high crystallinity, sharp heterointerfaces and coherently strained layers to the GaN substrate seen in Extended Data Fig. 4.

For successful dualtronics, the survival of the thin LED heterostructure layers during the epitaxial growth conditions of the HEMT side needs to be replicated for the steps of device fabrication. The N-polar surface is particularly difficult to protect chemically because it reacts strongly with several solutions that are typically used for surface cleaning and conditioning during the processing steps. For example, the natural etching of the N-polar surface of GaN in solutions of KOH, TMAH or H₃PO₄ form pyramidal facets that substantially roughen the pristine, flat surface⁹. This property of surface roughening is put to good use for efficient light extraction in blue LEDs. But such roughening would destroy the N-polar HEMT desired here, which requires sub-nanometre smoothness to preserve the high-mobility 2DEG and for the fabrication of source, drain and gate electrodes. Figure 3 shows the procedure used for the fabrication of the HEMT and LED on the opposite sides of the same wafer, along with the final dualtronic device. A more detailed overview schematic of the fabrication of the dualtronic device is shown in Extended Data Fig. 5.

a–d, Device processing flow for the double-sided HEMT-LED. Starting from the as-grown heterostructures, the grey arrows follow the independent processing steps chronologically, with the metal-polar LED being processed after the N-polar HEMT. e, A 3D representation of the complete device. f, Optical microscope images of the as-processed sample, focused on the LEDs (right) and on the HEMTs (left). The N-polar HEMTs are oriented upwards, forming the uppermost surface. For scale, the diameter of the large LED anode contact is 140 μm. g, Scanning electron microscope images of the HEMTs on the N-polar GaN surface (bottom) and the LEDs on the metal-polar GaN surface (top).

Figure 3a shows the starting layer structure, which is first processed into a HEMT on the N-polar side shown in Fig. 3b. Ti/Au low-resistance source and drain ohmic contacts are formed on regrown GaN layers that are heavily n-type doped with silicon. A thin gate SiO₂/Al₂O₃ dielectric layer was deposited, followed by a Ni/Au gate metal stack to complete the HEMT fabrication. The HEMT layer was then protected by a photoresist coating and flipped upside down to the metal-polar side up for LED fabrication for each photolithography step.

Figure 3c shows that, for the fabricated LED, Pd/Au metallization was used for the p-type anode contact. The high-work-function metal palladium aligns the metal Fermi level close to the valence-band holes of the p-type doped InGaN cap layer of the LED side. The thin p-InGaN reduces the contact resistance by enhancing the free hole concentration and by lifting the valence-band energy maximum. To form the n-type cathode contact, mesa regions were etched by removing the entire active region and terminating the etch on the buried n-type GaN layer to reveal a flat surface. Ti/Au metal stacks were deposited on this flat n-GaN surface to form a low-resistance contact between electrons at the Fermi surface of the metal and the electrons in the GaN conduction band.

Figure 3d shows the completed dualtronic structure after removal of the protective photoresist coating and Fig. 3e shows a cross-section of it flipped upside down. This schematic emphasizes the single-crystal nature of the whole structure, across the bulk of the wafer indicated by the wavy break in the thick GaN layer between the HEMT on the top and the LED on the bottom. Figure 3f shows an optical microscope image focused on the top HEMT surface on the left and the bottom LED focused on the right. The room-temperature energy bandgap of the bulk wafer between the HEMT and LED mesas is E_g1 = 3.4 eV of GaN, which is optically transparent to visible light, allowing for imaging of the LEDs through the wafer from the HEMT side as seen in Fig. 3f. This also emphasizes that the transistor layers are transparent to visible wavelengths and are therefore attractive for dualtronic integration with LEDs as transparent thin-film transistors (TFTs) but, as seen later, with much higher performance than existing TFTs made of oxide or organic semiconductors. Figure 3g shows the scanning electron microscope images of the HEMTs on the N-polar GaN surface (bottom) and the LEDs on the metal-polar GaN surface (top).

Figure 4a shows the measured transfer characteristic and transconductance of the N-polar HEMT. A threshold voltage of V_T ≈ −3 V is measured with an on-current drive exceeding 1 A mm^−1 at a drain voltage of V_DS = 5 V and gate voltage of V_GS ≈ 3 V. The transistor transconductance g_m = ∂I_D/∂V_GS peaks at V_GS ≈ −1.6 V. Figure 4b shows the measured I_D–V_DS output characteristics of the HEMT for various gate voltages overlaid with the LED current–voltage characteristics. The transistor delivers enough output current to drive the LED with variable drive currents controlled by the gate voltage up to 50 mA or more. Figure 4c shows the LED current and the HEMT current for V_DS = 5 V in the logarithmic scale, indicating that the HEMT can control the LED current and hence its optical output intensity over several orders of magnitude. Figure 4d shows the measured output spectra of the LED for injection current densities ranging from 1 to 140 A cm^−2. The emission intensity increases with the injection current density and the peak emission wavelength blueshifts from λ ≈ 470 nm at 1 A cm^−2 to λ ≈ 450 nm at 140 A cm^−2 in the blue regime of the visible spectrum. The blueshift is caused by screening of the internal polarization fields, which result in a reduced quantum-confined Stark effect, as well as the Burstein–Moss effect. The inset in Fig. 4d shows the bright blue electroluminescent glow of the blue LED slightly below the centre of the wafer on the metal-polar side. The output and transfer characteristics of the HEMTs before and after the fabrication of the LEDs are shown in Extended Data Fig. 6. It was confirmed that, after the LED processing, there was minimal degradation of the HEMT with a less than 0.3 V threshold voltage shift and negligible change in the output and gate leakage current. This minimal degradation highlights the feasibility of using double-side epitaxy and processing for reliable heterogeneous device integration on both crystal faces.

a, Normalized drain current (black line) and transconductance (grey line) of an N-polar HEMT as a function of gate-source voltage, operating at a drain-source voltage of 5 V. b, Linear plot showing the family of curves for a HEMT (black lines) for a gate-source voltage ranging from 1.75 (on) to −3.25 V (off). On the right axis, the linear current–voltage characteristics of a 400-μm-diameter LED (blue line) and unnormalized output characteristics of a HEMT are shown as well. The dimensions of the measured HEMTs are L_SD = 4 μm, L_G = 1.5 μm and W_G = 50 μm. c, Semi-log plots showing the unnormalized drain current (solid black line) and gate current (dashed black line) versus gate-source voltage, for a drain-source voltage of 5 V. Here the transistor current corresponds to the left vertical axis. The horizontal dashed black line indicates a normalized channel sheet current density of 1 A mm^−1. Similarly, corresponding to the right vertical axis is the LED current (blue line) as a function of forward bias for a 400-μm-diameter device. d, Electroluminescence spectra of a metal-polar, 400-μm-diameter LED. The injection current density ranges from 1 to 140 A cm^−2. The inset shows a camera image of the sample with an LED in the on state. a.u., arbitrary units.

Source Data

Figure 5a shows the layer structures and corresponding energy-band diagrams of the HEMT on the left and the LED on the right when they are connected in the circuit format shown in Fig. 5b. The source of the HEMT is connected to the anode of the LED to inject current to light up the LED. The light emitted from this dualtronic device is measured with a photodiode. A customized probe setup was fabricated to probe both HEMTs and LEDs without the need to flip the sample, shown schematically in Extended Data Fig. 7. When the gate voltage of the HEMT is lower than the transistor threshold (V_GS < V_T), the triangular quantum well is lifted above the Fermi level in the energy-band diagram of the HEMT as shown and the transistor is off. The corresponding energy-band diagram of the LED is shown on the right. When the HEMT gate voltage is above threshold, the bottom of the triangular quantum well is pulled below the Fermi level, as seen in Fig. 5a, flooding the GaN/AlGaN interface with electrons that form a 2DEG channel. The output current flowing from the drain of the HEMT is then injected into the cathode of the LED, forward biasing it to the energy-band diagram shown and indicated by the split quasi-Fermi levels of electrons E_fn and holes E_fp. The electrons injected from the cathode radiatively recombine with the holes injected from the anode into the In_0.17Ga_0.83N quantum well, emitting light that is sensed by the photodiode.

Figure 5c shows that the photovoltage signal follows the gate voltage as it is swept between −3.0 ≤ V_GS ≤ 3.6 V with the drain voltage set at a voltage of 5.9 V and the back-gate voltage set at a voltage of 1 V, over the millisecond timescale across its threshold voltage from the on state to the off state. Figure 5d shows that, when the gate voltage of the HEMT is modulated in the range 2.0 ≤ V_GS ≤ 3.0 V when it is in its on state with V_GS > V_T, the light emitted by the LED follows this gate modulation, translating the electronic signal in the gate of the HEMT into a photonic bright and dim signal output from the LED on the other side of the wafer. The on–off modulation of the dualtronic device can be improved by using similar concepts as used for traditional LEDs^28,29.

a, Energy-band diagrams of the HEMT and LED indicating the off and on states. b, Circuit schematic of the monolithic HEMT-LED, taking into account the back-gating effect of the conductive GaN substrate. c,d, Monolithic switching measurements, modulating between on and off (c) and between bright and dim modes (d). The gate-source modulation voltage is shown in red, the photodiode voltage while modulating the LED in solid blue and the background (LED off state) photodiode voltage in dashed blue. a.u., arbitrary units.

Source Data

Owing to the dualtronic structure, the cathode of the LED can also serve as a back gate for the HEMT, which was taken into account in the monolithic switching measurements. With a separate contact to the n-GaN substrate, we observed that the cathode voltage could exponentially control the drain current when the top gate was electrically floating. This back-gating effect is available as a new functionality or may be eliminated if undesirable by replacing the conductive substrate with a semi-insulating substrate. The back-gating effect of the HEMT with the LED is shown in Extended Data Fig. 8.

The observations described here thus prove that the concept of dualtronics is feasible, opening the paths for many interesting possibilities. This work realized an electronics fabric on the N-polar face and a photonic fabric on the metal-polar face of the same wafer. This dualtronic combination is of immediate interest for combining microLEDs on the photonic side with transparent TFTs on the same GaN wafer on the other side. This monolithic convergence of devices reduces the number of components required for microLEDs, with the potential for substantial area and cost savings owing to efficient use of the substrate real estate. Moreover, monolithic integrating schemes combining transistors and LEDs on a single substrate face rely on several epitaxial layers that need to be either selectively removed and/or regrown to expose the LED heterostructure buried underneath the transistor structure or vice versa^{30,31,32,33,34,35,36,37}. Exposing the buried heterostructure by dry etching induces plasma damage, leading to diminished light emission and degradation of contacts. Similarly, with a selective growth method, the ex situ nature of regrowth results in a poor growth interface, increasing leakage pathways. These issues are all avoided with the dualtronics integration scheme.

The dualtronics concept extends to several exciting new opportunities. The metal-polar face of the substrate, to take advantage of high emission efficiencies, can be used for any optoelectronic device such as laser diodes, semiconductor optical amplifiers and electro-optical modulators, while transistors or photodetectors are fabricated on the N-polar face. This full use of the substrate markedly decreases the number of components and chips needed in photonic integrated circuits. For other applications, both GaN polarities can be used. For instance, RF transistor power amplifiers for the transmit part of communication systems can be realized on one polarity and low-noise amplifiers for the receive end of communication systems can enable integrated transceivers in a combined transmit/receive module of smaller form factor than existing systems. The combination of n-channel transistors on one polarity with p-channel transistors on the opposite polarity can enable new forms of complementary transistor circuit topologies connected by through-vias through the substrate. Such dualtronic devices can take advantage of the wide-bandgap nature of the polar nitride semiconductors for new forms of power electronics and RF electronics. These, and several allied possibilities, can allow for the creation and manipulation of electrons and photons on the opposite faces of the same wafer to achieve new functionalities.

The ultrawide-bandgap polar semiconductor AlN boasts large electro-acoustic coupling, which makes it the preferred material today for acoustic-wave RF filters. Dualtronics can thus take advantage of this property of the polar nitride semiconductors to combine sonar (by sound waves), radar (microwaves) and lidar (light) on the same platform.

The efficient use of the substrate surfaces eliminates wasted space, reduces the energy and material costs of producing several wafers and thus should be of great interest for future technologies well beyond the particular polar semiconductor materials discussed here.