Spin photovoltaic effect in magnetic van der Waals heterostructures

Spin-photovoltaic effect in 2D magnet CrI₃ exhibits interplay between magneto-excitons, photon energy, and light helicity.

Abstract

The development of van der Waals (vdW) crystals and their heterostructures has created a fascinating platform for exploring optoelectronic properties in the two-dimensional (2D) limit. With the recent discovery of 2D magnets, the control of the spin degree of freedom can be integrated to realize 2D spin-optoelectronics. Here, we report spin photovoltaic effects in vdW heterostructures of 2D magnet chromium triiodide (CrI₃) sandwiched by graphene contacts. The photocurrent displays a distinct dependence on light helicity, which can be tuned by varying the magnetic states and photon energy. Circular polarization–resolved absorption measurements reveal that these observations originate from magnetic order–coupled and, thus, helicity-dependent charge-transfer excitons. The photocurrent displays multiple plateaus as the magnetic field is swept, associated with different CrI₃ spin configurations. Giant photo-magnetocurrent is observed, which tends to infinity for a small applied bias. Our results pave the way to explore emergent photospintronics by engineering magnetic vdW heterostructures.

INTRODUCTION

Spintronics aims at manipulating the spin degree of freedom in electronic systems for novel functionalities (1). In optoelectronics, the generation and control of spins can open up emerging opportunities for spin-optoelectronics, enabling the exploration of new spin photovoltaic effects and spin photocurrents. In various magnetic heterostructures, spin photovoltaic effects can be realized by different mechanisms. For instance, a spin voltage arises from spin-dependent excitation at the interface of a nonmagnetic metal in close proximity to a magnetic insulator (2). In spin valves and magnetic p-n junctions, spin injection and accumulation can be induced by the spin-dependent injection process of the photogenerated carriers at the interfaces with ferromagnetic contacts (3–6). Alternatively, in materials without intrinsic magnetic order, circularly polarized light can generate spin photocurrents via the circular photogalvanic effect (7–10). Among these materials and heterostructures, two-dimensional (2D) materials, particularly transition metal dichalcogenides (TMDs), have established themselves as a promising system for spin-optoelectronics due to their spin valley–dependent properties and enhanced photoresponsivity from strong excitonic effects (9–17).

The recent discovery of 2D van der Waals (vdW) magnets provides a new platform for spin photovoltaic effects based on atomically thin materials with intrinsic magnetic order (18–21). Among these magnets, chromium triiodide (CrI₃) is particularly interesting because of its layered antiferromagnetism (AFM), where the ferromagnetic monolayers with out-of-plane magnetizations are antiferromagnetically coupled to each other, as shown in Fig. 1A. The spin configurations can be manipulated by an external magnetic field that switches the sample between the AFM ground states and the fully spin-polarized states via a series of spin-flip transitions (18). Multiple magnetic states become accessible as the number of CrI₃ layers increases (22), possibly enabling multiple states of the resulting spin photocurrent, defined as a photocurrent controlled by the spin degree of freedom. Moreover, given the reported strong magneto-optical and excitonic effects (23, 24), atomically thin CrI₃ should provide an ideal platform to explore spin-optoelectronic effects in the atomically thin limit (21).

Fig. 1. Photocurrent response of CrI₃ junction device.

(A) Schematic of a four-layer CrI₃ junction device in AFM ground state (↑↓↑↓), with top and bottom graphene contacts and hBN encapsulation. (B) I-V curves of a four-layer CrI₃ junction (D2) under dark condition (black curve) and with 1 μW of 1.96-eV laser excitation (red curve). Inset is a zoomed-in view of the generated photocurrent at zero bias I_ph and open-circuit voltage V_oc. (C) Differential reflectance (ΔR/R; black dots) and photocurrent (I_ph; blue squares) as a function of photon energy for trilayer (3L) CrI₃ at −2 T. The photocurrent is measured from a trilayer CrI₃ junction device (D1) with an optical power of 10 μW. (D) Optical microscopy image of the 3L CrI₃ junction device (D1). Scale bar, 5 μm. (E and F) Spatial maps of photocurrent and RMCD signal measured from the same device at 0 T with an optical power of 1 μW. Scale bars, 5 μm.

RESULTS

Photocurrent response of CrI₃ junction device

To investigate photocurrent response from CrI₃, a vertical heterostructure was fabricated for efficient photodetection. All measurements were carried out at a temperature of 2 K with a magnetic field in the out-of-plane direction and a linearly polarized laser excitation, unless otherwise specified. As shown in Fig. 1A, the heterostructure consists of an atomically thin CrI₃ flake sandwiched by two graphene sheets as bias electrodes, encapsulated by thin hexagonal boron nitride (hBN) to avoid degradation (see Materials and Methods). Such a structure is essentially the same as a magnetic tunnel junction, which has been used to realize large tunneling magnetoresistance via the spin-filtering effect enabled by the layered AFM in CrI₃ (22, 25–27). Using a four-layer CrI₃ device (D2) as an example, without optical illumination, the current-bias characteristics [current-voltage (I-V) curves] of the device behave as a typical tunnel junction. The tunneling current is suppressed in the low-bias regime and dominated by Fowler-Nordheim tunneling at high bias (Fig. 1B, black curve) (22, 25–27).

Compared to the dark condition case, a substantial enhancement of the current is observed with photoexcitation of carriers in the low-bias regime. The red curve in Fig. 1B is obtained with 1.96 eV (632.8 nm) of laser excitation focused to a ~1-μm spot size at normal incidence, with an optical power of 1 μW. This carrier collection process is highly efficient in the vertical junction structure of atomically thin CrI₃, due to the reduced requirement of the carrier diffusion length. At zero bias, a net photocurrent I_ph is also generated (Fig. 1B, inset). This zero-bias photocurrent can be attributed to the asymmetric potential of the junction (11, 12), which could originate from the potential difference between the top and bottom graphene/CrI₃ interfaces. Applying a bias voltage induces an external electric field, which can modulate the magnitude and reverse the direction of the photocurrent. When the applied bias compensates the built-in electric field such that the net current is zero, the system becomes equivalent to an open circuit, allowing us to measure the photogenerated open-circuit voltage (V_oc).

We investigate the spatial distribution of the photocurrent by using scanning photocurrent microscopy. Figure 1D shows the optical microscopy image of a trilayer CrI₃ device (D1), which has a large junction area, with the corresponding photocurrent map taken at zero bias, shown in Fig. 1E. We also use reflective magnetic circular dichroism (RMCD) microscopy to map out the trilayer CrI₃ flake shown in Fig. 1F. The RMCD measures the out-of-plane magnetization of the device at zero field, which is defined as (R_RCP − R_LCP)/(R_RCP + R_LCP)/2, where R_RCP (R_LCP) represents reflection amplitude for right (left) circularly polarized light (see Materials and Methods). By comparing the photocurrent map with the microscopy image and the RMCD map, the photoactive region can be identified at the junction region where the top and bottom graphene electrodes overlap.

Figure 1C shows the photon energy dependence of the photocurrent. I_ph increases sharply when the photon energy exceeds 1.7 eV. By comparison to the differential reflectance (ΔR/R) measurement of a trilayer CrI₃ on a sapphire substrate (see Materials and Methods), we attribute the strong photocurrent response to the optical excitation of ligand-to-metal charge-transfer excitons (23, 24). We do not observe photocurrent enhancement corresponding to the excitation of the lower-energy exciton at 1.5 eV. This is possibly due to its larger binding energy and more localized nature than the charge-transfer excitons (24). Notably, the photoresponsivity reaches 10 mA W⁻¹, which is already comparable to that achieved in the devices based on TMD semiconductors under similar conditions (the photocurrent map in fig. S1 shows that the photoresponsivity reaches 10 mA W⁻¹, and fig. S2 shows the laser excitation power dependence of photocurrent) (11, 12).

Magnetic order dependence of photocurrent

The photocurrent response has a strong dependence on the magnetic order. Figure 2A shows zero-bias I_ph as a function of the external magnetic field (μ₀H) in a four-layer CrI₃ with an optical power of 1 μW. As μ₀H is swept, I_ph exhibits several sharp transitions and multiple plateaus. Figure 2B shows RMCD signal with the corresponding magnetic states labeled, as identified in the previous studies (22, 28). For simplicity, only the positive magnetic field side is shown. The full-field data with magnetic states assignment can be found in fig. S3. Comparison of Fig. 2 (A and B) shows that the multiple photocurrent plateaus are associated with the distinct magnetic states. The low and high photocurrent plateaus at low and high fields can be assigned to the AFM ground states and fully spin-polarized states, respectively.

Fig. 2. Dependence of photocurrent on magnetic order of four-layer CrI₃.

(A) Photocurrent as a function of external magnetic field (μ₀H) measured from the four-layer (4L) CrI₃ junction device (D2) with an optical power of 1 μW. Green (orange) curve corresponds to decreasing (increasing) magnetic field. (B) RMCD as a function of μ₀H for the same device. Insets show the corresponding magnetic states and the optical microscopy image of the device (D2). (C) Tunneling current (I_t) as a function of μ₀H measured from the same device at 80-mV bias under dark condition. Insets are schematics of the device with laser excitation and under dark condition. (D) I_ph-V curves for the four-layer CrI₃ in the AFM ground state (↑↓↑↓, 0 T, black curve) and the fully spin-polarized state (↑↑↑↑, 2.5 T, red curve). (E) Magnitude of the photo-magnetocurrent ratio as a function of bias extracted from the I_ph-V curves in (D). The red shading denotes the bias range where |MC_ph| tends to infinity. Inset is a zoomed-in view of the I_ph-V curves in (D).

The intermediate magnetic states (either ↑↓↑↑ or ↑↑↓↑) result in a lower photocurrent than the AFM ground states. This nonmonotonic photocurrent response to the magnetic states is distinct from the monotonic increase of the tunneling conductance due to the spin-filtering effect (22, 29). As a comparison, Fig. 2C shows the tunneling current of the same device measured as a function of μ₀H at 80-mV bias under dark condition. In sharp contrast to I_ph, the tunneling current increases monotonically and markedly as the spins in each layer are aligned from ↑↓↑↓ to ↑↑↑↑, because the current-blocking antiparallel interfaces are removed. Note that the tunneling current varies by two orders of magnitude for different magnetic states (22, 25–27), while there is only a twofold difference in the photocurrent. For tunneling under dark condition, the electron energy is below the CrI₃ conduction bands. In contrast, the optical excitation generates photoexcited carriers in the conduction bands, and their asymmetric extraction by the top and bottom graphene electrodes results in the measured photocurrent. The spin configurations in CrI₃ determine the layer distribution of the wave function of photoexcited carriers, through which the extraction efficiencies at the top and bottom electrodes can be affected, accounting for the nonmonotonic magnetic state dependence of photocurrent (see text S1) (11, 12, 30, 31). Note that the determination of precise magnetic state–dependent band alignment between graphene and CrI₃ will require future theoretical and experimental efforts.

In analogy to giant magnetoresistance and tunnel magnetoresistance (1, 32–35), which are of great importance for spintronics, our spin-optoelectronic device exhibits a novel photo-magnetocurrent effect (3). Figure 2D shows the I_ph-V curves corresponding to the AFM ground state (0 T, black curve) and fully spin-polarized state (2.5 T, red curve), respectively. For the short-circuit condition, the fully spin-polarized state generates a higher photocurrent, whereas the AFM ground state gives a larger open-circuit voltage magnitude. To quantify this magnetic state dependence, we define the photo-magnetocurrent ratio as MC_ph = (I_ph^p − I_ph^ap)/I_ph^ap, where I_ph^p and I_ph^ap are the photocurrents corresponding to the fully spin-polarized state (parallel) and AFM ground state (antiparallel). Figure 2E shows the absolute value of MC_ph as a function of bias extracted from the I_ph-V curves in Fig. 2D. A giant MC_ph is observed within a range of bias voltage indicated by the red shading. This can be attributed to the magnetic state–dependent open-circuit voltage, where, at certain bias, I_ph^ap goes to zero while I_ph^p is still finite, leading to a giant MC_ph ratio tending to infinity (3). This demonstrates a proof-of-concept photomodulated magnetocurrent effect. We further investigate the excitation power dependence of the photo-magnetocurrent effect (see fig. S4), which demonstrates optical control of the photo-magnetocurrent effect with high sensitivity to excitation power. Achieving such a giant and tunable photo-magnetocurrent could be useful for optically driven magnetic sensing and data storage technologies.

Dependence of photocurrent on light helicity

The broken time-reversal symmetry of our system should also enable a light helicity dependence of the spin photocurrent. Here, we use the trilayer CrI₃ device (D1) with 1.96 eV of excitation as an example. The magnetization is set in the fully spin-polarized state, either ↑↑↑ (2 T) or ↓↓↓ (−2 T). As the light helicity is switched between σ⁺ and σ⁻, the photocurrent exhibits a clear circular polarization dependence. As shown in Fig. 3A, the ↑↑↑ state (red dots) gives a higher photocurrent for photon helicity σ⁻ (135°) than σ⁺ (45°). In contrast, the ↓↓↓ state (black dots) exhibits the exact opposite helicity dependence, consistent with the time-reversal operation that connects the two fully spin-polarized states.

Fig. 3. Helicity dependence of photocurrent in trilayer CrI₃.

(A) Photocurrent as a function of quarter-wave plate angle for ↑↑↑ state (2 T, red dots) and ↓↓↓ state (−2 T, black dots) measured from the trilayer CrI₃ junction device (D1) with an optical power of 10 μW. Vertical arrows represent linearly polarized light. (B) The change in photocurrent [ΔI_ph [σ⁺ − σ⁻] = I_ph(σ⁺) − I_ph(σ⁻)] as a function of μ₀H measured from the same device with an optical power of 10 μW. The degree of helicity ΔI_ph [σ⁺ − σ⁻]/(I_ph(σ⁺) + I_ph(σ⁻) given on right axis. Insets show the corresponding magnetic states and schematic of the device with circularly polarized light excitation. (C) RMCD as a function of μ₀H for the same device. Insets show the corresponding magnetic states and the optical microscopy image of the device (D1). Scale bar, 15 μm.

We define the difference of photocurrent between σ⁺ and σ⁻ excitation as ΔI_ph [σ⁺ − σ⁻] = I_ph(σ⁺) − I_ph(σ⁻). The degree of helicity is then denoted by ΔI_ph [σ⁺ − σ⁻]/(I_ph(σ⁺) + I_ph(σ⁻)). To fully understand the interplay between the helicity-dependent photocurrent and the underlying magnetic order, we measure ΔI_ph [σ⁺ − σ⁻] and the degree of helicity as a function of μ₀H shown in Fig. 3B (see Materials and Methods). Four distinct plateaus are observed, which behave essentially the same as the RMCD signal versus μ₀H measured from the same device with the same 1.96-eV laser (Fig. 3C). We can thus assign the corresponding magnetic states to each ΔI_ph [σ⁺ − σ⁻] plateau. There is notable magnetic hysteresis of ΔI_ph [σ⁺ − σ⁻] centered at zero field due to switching between the ↑↓↑ and ↓↑↓ AFM-coupled ground states. In addition, the four-layer CrI₃ device (D2) shows a similar magnetic state–coupled helicity dependence of the photocurrent (fig. S3).

Optical selection rules of magnetic order–coupled charge-transfer excitons

We find that this unique spin photovoltaic effect originates from the helicity dependence of charge-transfer excitons in CrI₃, which couple to the underlying magnetic order. Figure 4A shows circular polarization–resolved differential reflectance (ΔR/R) measurements of a trilayer CrI₃ on a sapphire substrate. Data from all four magnetic states, {↑↑↑ (2 T), ↓↓↓ (−2 T), ↑↓↑ (0 T), ↓↑↓ (0 T)}, are shown. Evidently, the σ⁺/σ⁻ (red/blue dots) absorption peaks split in both energy and intensity and are determined by the magnetic state. This observation is consistent with the magnetic order–coupled charge-transfer excitons calculated by the many-body perturbation theory (24). The helicity-dependent absorption reveals the optical selection rules of the charge-transfer transitions between the spin-polarized valence and conduction bands (24, 36) and, thus, results in the observed helicity-dependent spin photovoltaic effect.

Fig. 4. Interplay between magnetic order and photon helicity in absorption and photocurrent of 3L CrI₃.

(A) Helicity-dependent ΔR/R spectra for all four magnetic states of 3L CrI₃ at selected magnetic fields. Red (blue) dots correspond to σ⁺ (σ⁻) photon helicity. Insets show the corresponding magnetic states and the optical microscopy image of a trilayer CrI₃ on sapphire. (B) Photocurrent as a function of quarter-wave plate angle for ↑↑↑ state (2 T, red dots) and ↓↓↓ state (−2 T, black dots) measured with three selected photon energies indicated by the dashed lines in (A). (C) ΔR/R helicity difference [(ΔR/R(σ⁺) − ΔR/R(σ⁻), curve] and the overlaid change in photocurrent [ΔI_ph [σ⁺ − σ⁻] = I_ph(σ⁺) − I_ph(σ⁻), squares] as a function of photon energy for ↑↑↑ state (2 T, red) and ↓↓↓ state (−2 T, black).

Starting with the ↑↑↑ state, the 1.96 eV (632.8 nm) of excitation indicated by the red dashed line is near the resonance of the σ⁻ polarized charge-transfer exciton, while the σ⁺ resonance is at 2.07 eV, about 110 meV higher. The stronger absorption for σ⁻ than σ⁺ results in higher photocurrent for the σ⁻ excitation versus σ⁺, as shown in Fig. 3A. For the ↓↓↓ state, the absorption peaks are switched between σ⁺ and σ⁻ compared to the ↑↑↑ state, which agrees with the opposite helicity dependence of the ↓↓↓ state photocurrent (Fig. 3A). The magnetic ground states at zero magnetic field, ↑↓↑ and ↓↑↓, also give notable but opposite splitting between the σ⁺ and σ⁻ absorption peaks, due to their opposite net magnetizations. This splitting between σ⁺ and σ⁻ vanishes above the critical temperature of trilayer CrI₃ (fig. S5). Note that, for even-number layers, the vanishing net magnetization at the AFM ground states leads to vanishing helicity dependence of the charge-transfer excitons (see fig. S6 for the ΔR/R measurement of a six-layer CrI₃ on a sapphire substrate). All these observations confirm the underlying magnetic order as the origin of the helicity dependence of the charge-transfer excitons.

The circularly polarized optical selection rules of charge-transfer excitons also enable the control of photocurrent helicity dependence by tuning the optical excitation energy. We choose three selected photon energies, indicated by the dashed lines in Fig. 4A for the magnetic-state ↓↓↓ panel. These three energies represent stronger σ⁺ absorption than σ⁻ (1.88 eV), nearly equal absorption (2.01 eV), and stronger σ⁻ absorption than σ⁺ (2.13 eV), respectively. Figure 4B shows the corresponding helicity-dependent photocurrent at these photon energies. For the ↓↓↓ state, the σ⁺ excitation at 1.88 eV gives a higher photocurrent than σ⁻ excitation, and this scenario is reversed for 2.13 eV of excitation. The helicity dependence nearly vanishes for the 2.01 eV of excitation, consistent with the observed equal absorption of σ⁺ and σ⁻ polarized light. Figure 4C shows ΔI_ph [σ⁺ − σ⁻] at several photon energies. The helicity dependence exhibits a change in sign around 2.01 eV. This ΔI_ph [σ⁺ − σ⁻] as a function of photon energy matches well with the overlaid ΔR/R helicity difference [ΔR/R(σ⁺) − ΔR/R(σ⁻)]. As expected, the photocurrent behavior of the magnetic state ↑↑↑, the red dots and curves in Fig. 4 (B and C), is the time reversal of ↓↓↓ state. The circular polarization–resolved absorption measurements reveal a strong correspondence between the photocurrent and RMCD measurements in Fig. 3. For a given magnetic state (↑↑↑), the splitting between σ⁺ and σ⁻ in the ΔR/R spectra leads to circular polarization–dependent absorption and reflection for a fixed excitation energy, which results in the helicity-dependent photocurrent and RMCD, respectively. As the photon energy is tuned, the σ⁺/σ⁻ difference in the absorption can be switched, causing the photocurrent helicity dependence to also be reversed.

DISCUSSION

We explore spin photovoltaic effects in atomically thin CrI₃ vdW heterostructures. The photocurrent exhibits distinct responses to the spin configurations in CrI₃, together with a giant photo-magnetocurrent effect. The combination of our helicity-dependent photocurrent and circular polarization–resolved absorption measurements reveals the emergent interplay between the spin photocurrent and the underlying excitons, intrinsic magnetic order, photon energy, and helicity. Our work demonstrates a proof-of-concept 2D spin-photovoltaic device incorporating the intrinsic magnetic order in few-layer CrI₃. This study also establishes atomically thin CrI₃ as an archetypal 2D magnet for studying the photocurrent generation in a vertical junction device. This device structure can be adapted using other 2D magnets with varied magnetic order and coercive fields—such as CrBr₃, CrCl₃, and CrSBr—and holds promise for potential applications in magnetic sensing and data storage. Furthermore, our results show that the generated photocurrent can probe the magnetic order in CrI₃ and exhibit distinct responses to photon energy and helicity, which originates from magnetic order–coupled charge-transfer exciton states. This demonstrates the efficacy of photocurrent as a new means to probe magnetic order, charge-transfer exciton states, and magnetoexciton-photon coupling. This approach could be useful for exploring other 2D magnetic systems, for instance, probing the zigzag antiferromagnetic order–coupled excitons in NiPS₃ (37–41) and the charge-transfer process at the α-RuCl₃/graphene interfaces (42–45).

MATERIALS AND METHODS

Device fabrication

The monolayer/few-layer graphene and 15 to 25 nm of hBN flakes were mechanically exfoliated onto either 285 or 90 nm of SiO₂/Si substrates and examined by optical and atomic force microscopy under ambient conditions. Only atomically clean and smooth flakes were used for making devices. V/Au (5/50 nm) metal electrodes were deposited onto a 285-nm SiO2/Si substrate using standard electron beam lithography with a bilayer resist [A4 495 and A4 950 poly(methyl methacrylate)] and electron beam evaporation. CrI₃ crystals were exfoliated onto 90 nm of SiO₂/Si substrates in an inert gas glovebox with water and oxygen concentration of less than 0.1 parts per million. The CrI₃ flake thickness was identified by optical contrast with respect to the substrate using the established optical contrast model (18). The layer assembly was performed in the glovebox using a polymer-based dry transfer technique. The flakes were picked up sequentially: top hBN, top graphene contact, CrI₃, bottom graphene contact, and bottom hBN. The resulting stacks were then transferred and released on the prepatterned electrodes. In the resulting heterostructure, the CrI₃ flake is fully encapsulated on both sides, and the top/bottom graphene flakes are connected to the prepatterned electrodes. Last, the polymer was dissolved in chloroform for less than 1 min to minimize the sample exposure to ambient conditions.

Photocurrent measurement

The photocurrent measurements were performed in a closed-cycle cryostat (attoDRY 2100) at a temperature of 2 K and an out-of-plane magnetic field of up to 9 T. A 632.8-nm HeNe laser was focused to a ~1-μm spot size at normal incidence to generate photocurrent. Figure 1A shows the schematic of CrI₃ junction devices. For DC measurement, a bias voltage (V) was applied to the top graphene contact with the bottom contact grounded. The resulting photocurrent (I_ph) or tunneling current (I_t) was amplified and measured by a current preamplifier (DL Instruments, model 1211). For AC measurement, a standard lock-in technique was used to measure the change in photocurrent with Stanford Research Systems SR830. For the photon energy dependence measurement, a SolsTiS continuous-wave widely tunable laser was used to generate photocurrent. For the photon helicity dependence measurement, a motorized precision rotation mount was used to rotate an achromatic quarter-wave plate with respect to the linear polarized incident laser beam.

Optical measurements

The RMCD and Kerr rotation measurements were performed in two similar cryostats (attoDRY 2100 and Quantum Design OptiCool) under the same experimental conditions. A 632.8-nm HeNe laser was used to probe the device at normal incidence with a fixed power of 1 μW. The AC lock-in measurement technique used to measure the RMCD and Kerr rotation signal closely follows the previous RMCD and magneto-optical Kerr effect measurements of the magnetic order in atomically thin CrI₃ (18, 22). For the differential reflectance measurements, we spatially filtered a tungsten halogen lamp and focused the beam to a ~3-μm spot size on the CrI₃. The reflected light was deflected with a beam splitter and detected by a spectrometer and a liquid nitrogen–cooled charge-coupled device, which enabled signal measurement from 1.4 to 3 eV. To obtain the differential reflectance, we subtracted and normalized the CrI₃ reflectance by the reflectance of the sapphire substrate. The absorbance of CrI₃ is proportional to the differential reflectance, which can be determined as $\frac{1}{4}(n^2-1)\mathrm{\Delta }R/R$ (23, 46).

Supplementary Materials

This PDF file includes:

Supplementary Text

Table S1

Figs. S1 to S6