Element-specific, non-destructive profiling of layered heterostructures

Abstract

Fabrication of semiconductor heterostructures is now so precise that metrology has become a key challenge for progress in science and applications. It is now relatively straightforward to characterize classic III-V and group IV heterostructures consisting of slabs of different semiconductor alloys with thicknesses of $\sim$5 nm and greater using sophisticated tools such as X-ray diffraction, high energy X-ray photoemission spectroscopy, and secondary ion mass spectrometry. However, profiling thin layers with nm or sub-nm thickness, e.g. atomically thin dopant layers ($\delta$δでるた-layers), of impurities required for modulation doping and spin-based quantum and classical information technologies is more challenging. Here, we present theory and experiment showing how resonant-contrast X-ray reflectometry meets this challenge. The technique takes advantage of the change in the scattering factor of atoms as their core level resonances are scanned by varying the X-ray energy. We demonstrate the capability of the resulting element-selective, non-destructive profilometry for single arsenic $\delta$δでるた-layers within silicon, and show that the sub-nm electronic thickness of the $\delta$δでるた-layers corresponds to sub-nm chemical thickness. In combination with X-ray fluorescence imaging, this enables non-destructive three-dimensional characterization of nano-structured quantum devices. Due to the strong resonances at soft X-ray wavelengths, the technique is also ideally suited to characterize layered quantum materials, such as cuprates or the topical infinite-layer nickelates.

With the advent of scanning tunneling microscopy (STM) lithography in the 1990s [1, 2], it became possible to fabricate dopant-based nano-electronic structures in semiconductors [3], as can now be done with phosphorus, arsenic, and boron in silicon [4, 5, 6]. In the meantime, industrially fabricated transistors have reached a 7 nm scale [7]. A number of methods to image electronic nano-structures are available. The most popular are destructive, and include transmission electron microscopy [8], atom-probe tomography [9, 8] and secondary ion mass spectrometry (SIMS) [10, 8]. Non-destructive imaging techniques include X-ray fluorescence [11, 12], X-ray diffraction [13], angle-resolved photoemission spectroscopy (ARPES) [14], ellipsometry [15, 16], as well as scanning microwave [17, 18], broadband electrostatic force [19], and single-electron probe [20] microscopy. Additionally, imaging techniques based on ion-beams, such as nuclear reaction analysis [21], medium energy ion and Rutherford scattering [22, 23, 24], can be non-destructive for conductive layers but are typically slow. Most of these non-destructive techniques only give two-dimensional lateral information, whilst X-ray methods can produce three-dimensional images by tomography [25, 26] at the expense of time. Conversely, provided strong refractive index contrast, X-ray reflectometry [27] can measure the vertical depth-profile of atomically thin dopant layers in a reasonably short time, of order 10 minutes per scan already since the 1990s [28]. But for thin layers with small dopant concentrations such as atomically thin layers of dopant atoms ($\delta$δでるた-layers), considerable modelling is required to determine the dimensions of the layers, particularly in the presence of features such as surface roughness and oxidation or other elements in the device grown on a substrate wafer.

X-ray reflectometry can be made more sensitive to a specific element by measuring resonantly at energies around the corresponding X-ray absorption edge [29]. The atomic resonance induces a large phase shift in the reflected signal, which can be used to isolate the element’s contribution to the reflectivity and obtain its distribution as a function of depth. This is particularly relevant for low concentrations, as in dopant-defined devices in silicon which are based on $\delta$δでるた-layers with 3D dopant concentrations typically below 5% [5]. Here we show that for a $\delta$δでるた-layer of arsenic donors in silicon, analysis of resonant contrast X-ray reflectometry (RCXR) can be enormously simplified, requiring no modelling of the $\delta$δでるた-layer’s host heterostructure. Furthermore, for the devices studied here SIMS gives layer thicknesses which are $>2$ nm [14], considerably larger than the typically $<1$ nm thicknesses provided by measures accessing the conduction electrons in the $\delta$δでるた-layers [5, 14, 16]; in this work reflectometry resolves this discrepancy by showing that the chemical (arsenic) layer thicknesses are considerably smaller than the SIMS results and consistent with the electronic thicknesses established by ARPES and magneto-resistance. Therefore, RCXR is well suited for non-destructive high-throughput preliminary characterizations of semiconductor heterostructures.

Resonant-contrast X-ray reflectometry

Figure 1: X-ray reflectometry. Schematic of the sample’s layer structure and the measurement geometry. The surface consists of oxide (SiO₂) and silicon with a combined thickness of $d=d_{\rm SiO_{2}}+d_{\rm Si}$. Within the silicon lattice there is an arsenic-doped layer of thickness $\delta$δでるた. X-rays shine on the sample with an incidence angle $\theta$θしーた (with respect to the surface) and the specular reflection is detected by a detector placed at an angle $\theta_{D}=2\theta$θしーた start_POSTSUBSCRIPT italic_D end_POSTSUBSCRIPT = 2 italic_θしーた (with respect to the incident beam). Constructive interference occurs at $2d\sin\theta=m\lambda$θしーた = italic_m italic_λらむだ for two interfaces separated by a distance $d$, where $m$ is an integer and $\lambda$λらむだ is the X-ray wavelength.

X-ray reflectometry was first used in 1954 [30] to measure the thickness of copper on glass and since then has become a common technique for studying a wide variety of layered materials, including surfaces, thin films and multilayers [31]. It involves measuring the specular reflection from sample surfaces (see Fig. 1). The angle $\theta$θしーた is swept to obtain a reflectometry measurement.

Light travelling through a medium is scattered upon changes in the medium’s scattering length density $\rho=r_{0}\sum_{q}N_{q}f_{q}$ρろー = italic_r start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT ∑ start_POSTSUBSCRIPT italic_q end_POSTSUBSCRIPT italic_N start_POSTSUBSCRIPT italic_q end_POSTSUBSCRIPT italic_f start_POSTSUBSCRIPT italic_q end_POSTSUBSCRIPT, where $r_{0}$ is the classical electron radius, $N_{q}$ and $f_{q}=f_{1,q}+if_{2,q}$ are the number of atoms per unit volume and the complex atomic scattering factor for an atom of element $q$, respectively. For typical semiconductors X-rays scatter weakly, hence, for a continuously varying depth profile $\rho(z)$ρろー ( italic_z ), only single scattering events need to be considered and the measured reflection is described by the kinematic Born approximation [32, 33, 27],

$$R(Q)=\frac{(4\pi)^{2}}{Q^{2}}\left|\int\rho(z)e^{-iQz}\ \mathrm{d}z\right|^{2}=r_{F}(Q)^{2}|F_{Q}(\rho)|^{2},$$πぱい ) start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG start_ARG italic_Q start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG | ∫ italic_ρろー ( italic_z ) italic_e start_POSTSUPERSCRIPT - italic_i italic_Q italic_z end_POSTSUPERSCRIPT roman_d italic_z | start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT = italic_r start_POSTSUBSCRIPT italic_F end_POSTSUBSCRIPT ( italic_Q ) start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT | italic_F start_POSTSUBSCRIPT italic_Q end_POSTSUBSCRIPT ( italic_ρろー ) | start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT , (1)

where $Q=4\pi\sin\theta/\lambda$πぱい roman_sin italic_θしーた / italic_λらむだ is the $Q$-vector, $\lambda$λらむだ the X-ray wavelength, $F_{Q}(x)$ the Fourier transform, and $r_{F}(Q)^{2}=\frac{(4\pi)^{2}}{Q^{2}}$πぱい ) start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG start_ARG italic_Q start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG the squared Fresnel reflectivity.

The scattering length density profile $\rho(z)$ρろー ( italic_z ) of a material is the sum of each of its elements’ profiles $\sum_{q}\rho_{q}(z)$ρろー start_POSTSUBSCRIPT italic_q end_POSTSUBSCRIPT ( italic_z ). Therefore, if an element $q$ is dilute, it can be considered to contribute perturbatively to the host profile $\rho(z)$ρろー ( italic_z ) and, thus, to Eq. (1). So expanding $R(Q)$ in terms of $\rho_{q}(z)$ρろー start_POSTSUBSCRIPT italic_q end_POSTSUBSCRIPT ( italic_z ), the first-order term will be due to interference between the element $q$ and the host, represented mathematically by a product of the Fourier transforms of $\rho(z)$ρろー ( italic_z ) and the dilute elemental density profile denoted $\delta\rho(z)$δでるた italic_ρろー ( italic_z ) rather than the square of Fourier amplitude of $\delta\rho(z)$δでるた italic_ρろー ( italic_z ) appearing to second order. In the first part of our Methods section, we develop an analytic theoretical description with relevant equations that describe the consequences, with the following key conclusions:

1. Reflectometry data collected at a single X-ray energy can be separated into components belonging to different length-scales by Fourier filtering at the appropriate frequencies. As a consequence, if the dilute layer of element $q$ is thin compared to the other layers, its signal can readily be isolated, and if it has a Gaussian profile (of standard deviation $\sigma$σしぐま) its contribution is given by:

   $$\frac{I(Q)}{\sqrt{{|F|}_{\mathrm{LF}}^2}}=2r_0{N}_{\text{2D}}|\mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}f|\exp \left(-\frac{{(Q\mathrm{<ruby><rb>\sigma </rb><rt>しぐま</rt></ruby>})}^2}{2}\right)\cos (Qd-\varphi ),$$Δでるた𝑓superscript𝑄𝜎22𝑄𝑑italic-ϕ\frac{I(Q)}{\sqrt{|F|^{2}_{\mathrm{LF}}}}=2r_{0}N_{\text{2D}}|\Delta f|\exp% \left(-\frac{(Q\sigma)^{2}}{2}\right)\cos(Qd-\phi),divide start_ARG italic_I ( italic_Q ) end_ARG start_ARG square-root start_ARG | italic_F | start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT start_POSTSUBSCRIPT roman_LF end_POSTSUBSCRIPT end_ARG end_ARG = 2 italic_r start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT italic_N start_POSTSUBSCRIPT 2D end_POSTSUBSCRIPT | roman_Δでるた italic_f | roman_exp ( - divide start_ARG ( italic_Q italic_σしぐま ) start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG start_ARG 2 end_ARG ) roman_cos ( italic_Q italic_d - italic_ϕ ) , (2)

   where $d$ and $N_{\text{2D}}$ are the dilute layer’s depth and 2D density, respectively, $\delta=2\sqrt{2}\pi\sigma$δでるた = 2 square-root start_ARG 2 end_ARG italic_πぱい italic_σしぐま the layer’s thickness, $\mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}f=f_q-f_{\text{host}}$Δでるた𝑓subscript𝑓𝑞subscript𝑓host\Delta f=f_{q}-f_{\text{host}}roman_Δでるた italic_f = italic_f start_POSTSUBSCRIPT italic_q end_POSTSUBSCRIPT - italic_f start_POSTSUBSCRIPT host end_POSTSUBSCRIPT, $\varphi =\text{arg}(\mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}f)$Δでるた𝑓\phi=\text{arg}(\Delta f)italic_ϕ = arg ( roman_Δでるた italic_f ), $I(Q)$ the first-order term of the Born approximation’s expansion in $\delta\rho(z)$δでるた italic_ρろー ( italic_z ) [see Eq. (6) and (8)], and $|F|_{\mathrm{LF}}$ the low frequency part of the reflectometry.

2. Taking the difference $\mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}R(Q)$Δでるた𝑅𝑄\Delta R(Q)roman_Δでるた italic_R ( italic_Q ) between two reflectometry measurements at energies $E$ and $E^{\prime}$ straddling a resonant edge of the dilute layer’s element species, isolates the signal from the dilute layer [see Eq. (13) and (14)]. For a Gaussian profile this leads to Eq. (2) with the substitutions $I(Q)\to \mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}R(Q)/r_F{(Q)}^2$Δでるた𝑅𝑄subscript𝑟𝐹superscript𝑄2I(Q)\to\Delta R(Q)/r_{F}(Q)^{2}italic_I ( italic_Q ) → roman_Δでるた italic_R ( italic_Q ) / italic_r start_POSTSUBSCRIPT italic_F end_POSTSUBSCRIPT ( italic_Q ) start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT and $\mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}f\to \mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}{f}_{E^{\prime }}-\mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}{f}_E$Δでるた𝑓Δでるたsubscript𝑓superscript𝐸′Δでるたsubscript𝑓𝐸\Delta f\to\Delta f_{E^{\prime}}-\Delta f_{E}roman_Δでるた italic_f → roman_Δでるた italic_f start_POSTSUBSCRIPT italic_E start_POSTSUPERSCRIPT ′ end_POSTSUPERSCRIPT end_POSTSUBSCRIPT - roman_Δでるた italic_f start_POSTSUBSCRIPT italic_E end_POSTSUBSCRIPT, which directly yields the depth $d$ and thickness $\delta$δでるた of the layer, as opposed to conventional fitting of the full reflectometry data.

This first-order perturbation theory has the benefit of measuring the dimensions of a dilute layer within a heterostructure of other elements without the need to model the full structure. An ideal test-bed for this are the substitutional, nm-thin dopant $\delta$δでるた-layer in silicon, with typical 2D density $N_{\text{2D}}\approx$10¹⁴ cm^-2 corresponding to $N_{\text{3D}}=N_{\text{2D}}/\delta\approx$δでるた ≈10¹⁷ cm^-3 (for $\delta\sim$δでるた ∼1 nm) representing $<$5% of the total silicon atoms [5].

(a)

(b)

Figure 2: Dopant contribution to reflectometry across the arsenic $L_{3}$-edge. (a) X-ray reflectometry of sample #1 with photon energies below (red, 1300 eV) and above the arsenic $L_{3}$-absorption edge (black, 1335 eV), obtained by rotating the sample and the detector to measure the specular reflection. The horizontal axis shows the wave-vector change on scattering $Q$, to directly compare the interference patterns of both energies. The inset depicts the fast Fourier transform (FFT) of the reflectometry curves on a logarithmic scale, directly yielding the depth $d$ of the arsenic $\delta$δでるた-layer. The right axis of the inset shows the FFT phase and the phase difference (blue) at both energies. (b) The black line shows the $R$(1335 eV) data from (a) multiplied by $Q^{2}$. A Fourier transform filter was used to separate frequencies corresponding to reflections from depths less than 10 nm ($R$(Si), green) and greater than 10 nm ($R$(As), blue). The gray line represents the difference $\mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}R$Δでるた𝑅\Delta Rroman_Δでるた italic_R between the reflectometry measured above (1335 eV) and below (1300 eV) the arsenic $L_{3}$-edge. Both $R$(As) and $\mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}R$Δでるた𝑅\Delta Rroman_Δでるた italic_R have been divided by $\sqrt{R(\text{Si})}$ to isolate the arsenic contribution. The inset shows the arsenic scattering length density $\rho_{\text{As}}$ρろー start_POSTSUBSCRIPT As end_POSTSUBSCRIPT, obtained from $R$(As) and $\mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}R$Δでるた𝑅\Delta Rroman_Δでるた italic_R as described by Eq. (11) in the Methods. Dotted curves denote fits using the DYNA program and Eq. (2) in (a) and (b), respectively.

Extracting the arsenic signal from reflectometry

The samples used to demonstrate our method are Si(001) wafers with arsenic $\delta$δでるた-layers (Si:As) at varying depths $d=15-75$ nm below the surface and an oxide (SiO₂) surface layer of width $d_{\rm SiO_{2}}\approx 1$ nm [34] (see Fig. 1). They were prepared as described in the Methods, with resulting two-dimensional dopant density $N_{\mathrm{2D}}\approx 10^{14}$ cm^-2, as determined by SIMS, STM, magneto-resistance, and X-ray fluorescence [5, 11]. In addition, one control sample features a buried oxide layer instead of arsenic.

In the doped region, considering a $d_{\rm As}\approx 1$ nm thick layer, $<5\%$ of silicon atoms are replaced by arsenic, while the oxide contains twice as many oxygen atoms as silicon. Therefore, the optical contrast between the silicon and Si:As is small compared to that between oxide and silicon. Figure 2a shows typical reflectometry data for our samples. The black data are for an energy of 1335 eV, above the arsenic $L_{3}$-absorption edge at 1324 eV [35], and the red ones for 1300 eV, below the edge. Both datasets contain distinct fast oscillations due to interference between reflections from the $\delta$δでるた-layer and those from the surface region. The periodicity of the oscillations is related to the dopant layer depth $d$ through the Bragg condition $d=m\lambda/2\sin\theta$λらむだ / 2 roman_sin italic_θしーた (corrections from the refractive index $n$ can be omitted for X-rays), i.e., a smaller $d$ implies a larger period. In the inset of Fig. 2a, a depth of $d=18$ nm is obtained from the Fourier transform. The fast oscillations are modulated by an envelope that varies slowly with $Q$ on a scale inversely proportional to the thin arsenic layer thickness $\delta$δでるた and broadened also by the surface roughness.

At very small angles (i.e. for $Q<1.5$ nm^-1) the signal contains no meaningful structure on account of a combination of total external reflection and the leakage of light directly to the detector due to growth of the beam footprint to larger than the sample. At higher angles we can analyze the data quantitatively: the oscillations decay because the overall signal intensity decreases by the factors $r_{F}(Q)^{2}\propto Q^{-2}$ and $H(Q)$ [see Eq. (9) in Methods]. Surface roughness reduces the intensity by an additional factor which can be modelled by $e^{-Q^{2}\sigma_{r}^{2}}$σしぐま start_POSTSUBSCRIPT italic_r end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_POSTSUPERSCRIPT, where $\sigma_{r}$σしぐま start_POSTSUBSCRIPT italic_r end_POSTSUBSCRIPT is the root mean square roughness [36, 33]. This roughness was measured with STM and atomic force microscopy (AFM, see Methods), and found to be 0.1 nm, consistent with other studies [37].

(a)

(b)

Figure 3: Resonant reflectometry across the arsenic $L_{2,3}$-edges. (a) Sharp reflectometry edge at the arsenic $L_{3}$-absorption edge resonance of 1324 eV in sample #1, at an incidence angle of $\theta=10^{\circ}$θしーた = 10 start_POSTSUPERSCRIPT ∘ end_POSTSUPERSCRIPT, normalised by the incident photon flux $I_{0}$. The blue line is a fit assuming $\delta$δでるた = 1.5 nm (see Methods and Fig. 5). The top inset depicts the resonance process where an X-ray photon is absorbed and excites an electron from the $L$-shell to the unoccupied $M$- or $N$-shells. The bottom inset depicts the energy dependence of the arsenic, silicon, and oxygen scattering factors in full, dashed, and dotted lines, respectively [38]. Only the arsenic scattering factor changes noticeably in this energy range. (b) Simulated (black) and experimental (red) relative reflection change $\mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}R/R$Δでるた𝑅𝑅\Delta R/Rroman_Δでるた italic_R / italic_R, assuming $N_{\text{2D}}=2.77\times 10^{14}$ cm^-2. The black line is a guide to the eye used to estimate the error of the arsenic layer thickness $\delta$δでるた (see Methods). The intersection of the data and the simulation gives $\delta=1.6\pm 0.5$δでるた = 1.6 ± 0.5 nm for sample #1.

The phase between the two signals, shown in the inset of Fig. 2a is shifted at the arsenic resonance edge, as described by Eq. (6), where the second term is phase sensitive and proportional to the arsenic scattering length density. The expected phase shift is given by the scattering factor change $\mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}f=f_{\text{As}}-f_{\text{Si}}$Δでるた𝑓subscript𝑓Assubscript𝑓Si\Delta f=f_{\text{As}}-f_{\text{Si}}roman_Δでるた italic_f = italic_f start_POSTSUBSCRIPT As end_POSTSUBSCRIPT - italic_f start_POSTSUBSCRIPT Si end_POSTSUBSCRIPT, as shown in equation Eq. (8). Thus, reflectometry data taken at 1300 eV and 1335 eV will be shifted by $\text{arg}\left[\mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}f(1335\text{eV})\right]-\text{arg}\left[\mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}f(1300\text{eV})\right]\approx (0.33\pm 0.10)\mathrm{<ruby><rb>\pi </rb><rt>ぱい</rt></ruby>}$Δでるた𝑓1335eVargdelimited-[]Δでるた𝑓1300eVplus-or-minus0.330.10𝜋\text{arg}\left[\Delta f(1335\text{eV})\right]-\text{arg}\left[\Delta f(1300% \text{eV})\right]\approx(0.33\pm 0.10)\piarg [ roman_Δでるた italic_f ( 1335 eV ) ] - arg [ roman_Δでるた italic_f ( 1300 eV ) ] ≈ ( 0.33 ± 0.10 ) italic_πぱい, using scattering factors from [38]. This is in agreement with the measured phase shift of $(0.27\pm 0.03)\pi$πぱい at 18 nm indicated by the peak in the inset of Fig. 2a.

Conventional fits to the data using the DYNA [39] software in Fig. 2a are shown with dotted lines, where the three-dimensional density $N_{\text{3D}}$ and the $\delta$δでるた-layer thickness are obtained from fitting the resonance in Fig. 3, to reduce the number of free fitting parameters. Direct fits to the data for a single photon energy are unable to give reliable results for the $\delta$δでるた-layer thickness $\delta$δでるた, due to the large fitting parameter space containing each layer’s thickness, density, and roughness metrics, highlighting the need for a dedicated method to extract $\delta$δでるた.

To analyze the data at an X-ray energy of 1335 eV, above the arsenic $L_{3}$-edge (black data in Fig. 2a), the data are multiplied by $Q^{2}/(4\pi)^{2}$πぱい ) start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT to remove the $Q^{-2}$ divergence expected from Eq. (1). Thereafter, a Fourier transform filter is used to separate oscillations corresponding to reflections from depths less than 10 nm that we attribute to the silicon layer ($R$(Si), shown in green in Fig. 2b), and oscillations that stem from depths greater than 10 nm, which after division by $\sqrt{R(\text{Si})}$ we attribute to the arsenic layer ($R$(As), shown in blue in Fig. 2b). Under the assumption that the arsenic layer has a Gaussian profile, $R$(As) can be fitted to Eq. (2) (dotted blue line, Fig. 2b) to obtain the arsenic layer depth and thickness, here $d=18.1\pm 0.1$ nm and $\delta=0.9\pm 0.2$δでるた = 0.9 ± 0.2 nm, respectively.

In the Methods section, we show that for a thin arsenic layer, the slow $Q$ oscillations (see Fig. 2b, green) are largely due to the amplitude $|F_{Q}(\rho)|$ρろー ) | for the unperturbed host. Thus, we obtain the inverse Fourier transform of the arsenic factor $F_{Q}(\delta{\rho}(z))$δでるた italic_ρろー ( italic_z ) ) by first subtracting $|F_{Q}(\rho)|$ρろー ) | from the raw data divided by $Q^{2}/(4\pi)^{2}$πぱい ) start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT and then dividing by $\sqrt{|F_{Q}(\rho)|}$ρろー ) | end_ARG (see Fig. 2b, blue). Finally, taking the Fourier transform of $|F_{Q}(\delta{\rho}(z))|$δでるた italic_ρろー ( italic_z ) ) | results in the scattering length density profile $\rho_{\text{As}}(z)$ρろー start_POSTSUBSCRIPT As end_POSTSUBSCRIPT ( italic_z ) [Eq. (11)] shown in the inset of Fig. 2b, with a large density at the expected arsenic layer depth. The resolution in $z$ is $\sim$1.5 nm and dictated by the sampling range in $Q$; too low to resolve the shape of a $\delta<1$δでるた < 1 nm dopant-layer profile.

RCXR isolation of the $\delta$δでるた-layer signal

To quantify the modulation of the fast oscillations in the reflectivity, we look at the difference measured below and above the arsenic $L_{3}$-edge resonance energy at 1300 eV and 1335 eV, respectively, similarly to how magnetism can be extracted from the difference between left- and right-handed circularly polarized X-rays [40, 41].

Figure 2b shows in gray the difference $\mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}R$Δでるた𝑅\Delta Rroman_Δでるた italic_R of the reflectivities at the two energies divided by square root of the silicon contribution. Since the quantity of interest is the Fourier transform of the density profile $\rho(z)$ρろー ( italic_z ), the data were multiplied by $Q^{2}/(4\pi)^{2}$πぱい ) start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT consistent with Eq. (6). The subtraction removes the zeroth-order term due to the oxide and silicon layers and keeps the first-order term due to the arsenic-doped layer, because the scattering factors of the former are almost constant in this energy range, while they change drastically for the latter (see inset of Fig. 3b). The division by the Si/SiO₂ contribution leaves a fast oscillation, with a period corresponding to the arsenic $\delta$δでるた-layer depth, modulated by a slow envelope with a decay corresponding to the arsenic $\delta$δでるた-layer thickness (described by Eq. (13)). The same slow modulation of the fast oscillation is present in the data for $R(\text{As})$ in Fig. 2b. Assuming a Gaussian arsenic density profile, the data are fitted to Eq. (2) yielding a thickness of $\delta=0.6_{-0.2}^{+0.7}$δでるた = 0.6 start_POSTSUBSCRIPT - 0.2 end_POSTSUBSCRIPT start_POSTSUPERSCRIPT + 0.7 end_POSTSUPERSCRIPT ($0.4-1.3$) nm and depth of $d=18.4\pm 0.1$ nm in agreement with the result from a single energy (see Fig. 2b, gray), where $\delta$δでるた is taken to be the full width at half maximum of the Gaussian profile and the uncertainty indicates the fit’s 95% confidence interval. We emphasise that the value extracted here represents an upper bound on the arsenic layer thickness due to the high-angle cutoff due to noise and interface roughness [35, 42, 43]. Surface roughness contributions add in quadrature to the result, here roughness was found to be of the order of 0.1 nm by STM and AFM measurements (see Methods) and is, therefore, negligible.

The Fourier transform of the $\mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}R$Δでるた𝑅\Delta Rroman_Δでるた italic_R data in Fig. 2b also leads to the arsenic profile $\rho_{\text{As}}(z)$ρろー start_POSTSUBSCRIPT As end_POSTSUBSCRIPT ( italic_z ) (see inset of Fig. 2b). The obtained profile is in good agreement with that obtained from a single scan of the reflectivity, suggesting that our Fourier filtering procedures for single X-ray energies, not required when we are taking differences between data collected at different X-ray energies, are adequate. Furthermore, the resonance subtraction method suppresses fluctuations on the sides of the $\delta$δでるた-layer peak.

RCXR experiments were performed on five different samples, of which four contained Si:As $\delta$δでるた-layers and one reference sample a buried oxide layer. They were also measured with SIMS (see Methods). The results for both techniques are shown in Fig. 4, where the data in red are obtained from Fourier filtered data at a single energy above the arsenic $L_{3}$-edge and the data in black from the resonant contrast method.

Our isolation of the $\delta$δでるた-layer signal based on fitting the amplitude modulation of the dopant layer has the advantage that it relies neither on extensive multi-variable fitting, nor on prior knowledge or hypotheses concerning the sample composition other than assuming Gaussian profiles. Additionally, for thin layers, the amplitude modulation occurs over a large $Q$-range, such that the estimation of the layer thickness has high accuracy.

(a)

(b)

Figure 4: Comparison of dopant layer thicknesses and depths. The thickness $\delta$δでるた (a) and depth $d$ (b) of the Si:As layer below the surface, obtained by RCXR (black) and a single energy above the $L_{3}$-edge (E⁺, red), as well as SIMS are plotted on the vertical and horizontal axes, respectively. The inset in panel (b) shows the dependence of $\delta$δでるた on $d$ as measured with X-ray reflectometry (XRR). The shaded regions in (a) denote the thickness range obtained by ARPES and magneto-resistance (MR) of samples prepared in the same way. SIMS depth errors are from the uncertainty in the sputter rate (see Methods). XRR errors for E⁺ and RCXR are taken from the 95% confidence interval of fits to Eq. (2).

Arsenic $L_{2,3}$-edge resonance measurements

The previous section shows that upper bounds on $\delta$δでるた-layer thicknesses are readily obtained from examination of the rapid oscillations associated with their depth. On the other hand, short wavelength disorder can introduce noise at high momentum transfers $Q$ and, thereby, complicate the determination of lower bounds of $\delta$δでるた. However, it is possible to obtain alternative thickness estimates by relying on photon energy scans through the dopant resonance at low $Q$, provided that the two-dimensional dopant density $N_{\text{2D}}$ is known precisely. The arsenic $L_{2,3}$-edge resonances are visible when measuring the reflected intensity as a function of X-ray energy at a fixed angle (see Fig. 3a), where the arsenic complex atomic scattering factor changes abruptly, while the silicon and oxygen atomic scattering factors are smooth functions of energy (see inset of Fig. 3a) [38]. The scattering length density of the layers and the resonance intensity are determined by the effective three-dimensional dopant density $N_{\text{3D}}=N_{\text{2D}}/\delta$δでるた and the Born equation Eq. (1), such that the thinner the dopant layer ($\delta$δでるた), the stronger the resonance (because $N_{\text{3D}}$ is larger) at any non-zero $Q$ (described by Eq. (8)). The relative change in reflection, i.e., $\mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}R/R=[R(1330\text{eV})-R(1320\text{eV})]/R(1320\text{eV}),$Δでるた𝑅𝑅delimited-[]𝑅1330eV𝑅1320eV𝑅1320eV\Delta R/R=[R(1330~{}\text{eV})-R(1320~{}\text{eV})]/R(1320~{}\text{eV}),roman_Δでるた italic_R / italic_R = [ italic_R ( 1330 eV ) - italic_R ( 1320 eV ) ] / italic_R ( 1320 eV ) , depends mostly on $N_{\text{3D}}$ of the layer that undergoes the resonance, here the arsenic $\delta$δでるた-layer, but its absolute value also hinges on the characteristics of all other layers [apparent from Eq. (6), see Methods]. As a consequence, if $N_{\text{2D}}$ and the layer depth are known, it is possible to obtain its thickness $\delta$δでるた directly from the resonance spectrum.

The two-dimensional dopant density $N_{\text{2D}}$ is readily obtained by X-ray fluorescence [11]. In principle, it is straightforward to record the X-ray fluorescence and reflectivity simultaneously, as the fluorescence photons have an isotropic distribution. $N_{\text{2D}}$ was measured for one sample in this work (sample #1 shown in Fig. 2 & 3), and was found to be $N_{\text{2D}}=(2.77\pm 0.14)\times 10^{14}$ cm^-2 by the same method as in [11]. The relative change in reflection $\mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}R/R$Δでるた𝑅𝑅\Delta R/Rroman_Δでるた italic_R / italic_R at the arsenic $L_{2,3}$-edges as a function of the dopant layer thickness was also calculated with the DYNA program [39] (see Fig. 3b, black). The layer depth used for this calculation was obtained in Fig. 2a. As expected, decreasing the dopant layer thickness and/or increasing $N_{\text{3D}}$ at fixed $N_{\text{2D}}$, increases the relative change in resonance intensity $\mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}R/R$Δでるた𝑅𝑅\Delta R/Rroman_Δでるた italic_R / italic_R. In Fig. 3b, the experimental value of $0.09\pm 0.01$ for $\mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}R/R$Δでるた𝑅𝑅\Delta R/Rroman_Δでるた italic_R / italic_R extracted from Fig. 3a is shown in red. An arsenic $\delta$δでるた-layer thickness of $1.6\pm 0.5$ nm is deduced. Within errors this agrees with the value from the previous section ($\delta=0.6_{-0.2}^{+0.7}$δでるた = 0.6 start_POSTSUBSCRIPT - 0.2 end_POSTSUBSCRIPT start_POSTSUPERSCRIPT + 0.7 end_POSTSUPERSCRIPT nm), but relies on fitting the resonance with each layer’s thickness, density, and roughness, resulting in a large number of fitting parameters, and prior knowledge about all layers in the sample.

Discussion

RCXR and SIMS experiments were performed on four samples containing an arsenic $\delta$δでるた-layer. Figure 4 shows that the depths $d$ of the shallower samples agree within error bars, whereas SIMS underestimates the depth of deeper Si:As $\delta$δでるた-layers. This discrepancy might originate from the variability of the SIMS sputter rate during a measurement [44]. The $\delta$δでるた-layer thickness (see Fig. 4a) is lower when measured with X-ray reflectometry, as is expected since the SIMS resolution is $\approx 2$ nm [45]. The values for the layer thickness measured with RCXR denote an upper bound, in our case determined by the maximum momentum transfer $Q_{\rm max}$=5 nm^-1. In the present experiment where $\lambda$λらむだ=0.94 nm, $Q_{\rm max}$ is fixed by $\theta_{\rm max}=22^{\circ}$θしーた start_POSTSUBSCRIPT roman_max end_POSTSUBSCRIPT = 22 start_POSTSUPERSCRIPT ∘ end_POSTSUPERSCRIPT, the maximum angle of incidence, which can easily be raised in experiments with experiments with next-generation synchrotron sources and instruments. Nevertheless, the RCXR results show that our arsenic $\delta$δでるた-layer samples are as thin as 0.6 nm. Also, the upper-bound thicknesses measured here are in good agreement with results from ARPES [14], where Si:As samples fabricated in the same way were measured to be 0.4 to 0.7 nm thick (see shaded area in Fig. 4a) from the point of view of the 2D electron liquids hosted by the $\delta$δでるた-layers. Additionally, one of the samples in this study ($\delta_{RCXR}=1.4\pm 0.5$δでるた start_POSTSUBSCRIPT italic_R italic_C italic_X italic_R end_POSTSUBSCRIPT = 1.4 ± 0.5 nm) was etched into a Hall bar geometry and contacted for low-temperature magneto-resistance [46], from which a conductive layer thickness of $0.97\pm 0.02$ nm was extracted [11]. Magneto-resistance of ten equivalent arsenic $\delta$δでるた-layers yielded a thickness range from 0.4 nm to 1.8 nm (see shaded area in Fig. 4a), further corroborating our method and results.

Dopant $\delta$δでるた-layers in silicon have been extensively studied with magneto-resistance [46], SIMS [8], and ARPES [14]. The agreement between data in Fig. 4 measured at a single energy above the arsenic $L_{3}$-edge (red) and by RCXR (black) shows that RCXR, treated within the first-order perturbation theory described here, is able to extract the arsenic dopant layer depth and thickness reliably without extensive structural modelling. The fact that the difference in phase shift of the oscillations on and off resonance matches the tabulated atomic phase shifts certifies the applicability of perturbation theory.

While RCXR is used here to enhance the sensitivity to a single two-dimensional dopant layer in silicon, naturally it also increases the sensitivity to specific layers in any multilayer material, including group IV and III-V semiconductors, as well as quantum materials where electron correlations are strong. For example, owing to the strong soft X-ray resonances of the transition metal $L$-edges and the oxygen K-edge, RCXR is ideally suited to non-destructively characterize oxide heterostructures [47, 48]. An ideal venue for our technique are the recently discovered superconducting infinite-layer nickelates [49, 50], for which the exact structure of the infinite-layer phase and other topotactically related phases are currently a matter of discussion [51, 52]. In particular, there is the possibility of an oxygen-ordered impurity superlattice [53], for which RCXR can be used to isolate and amplify the oxygen signal.

In conclusion, X-ray reflectometry can be made sensitive to specific elements in layered samples by performing a differential measurement above and below a resonance absorption edge of the respective element. We show that with this technique and a simple perturbation expansion of the differential reflectivity, it is possible to isolate the signal from one specific element without needing to model the material, yielding an upper bound as low as 0.9 nm for the thickness of arsenic-doped $\delta$δでるた-layer buried in silicon with a 3D arsenic density $<5\%$ of the total silicon density. In principle, the upper bound can be extended by increasing the X-ray fluence to scan to higher reflection angles, as well as by reducing the sample surface roughness. Combination of this approach with nano X-ray fluorescence detection [11] will make it possible to non-destructively characterize samples in three-dimensions in a single experiment. Finally, specular and off-specular RCXR with nano-X-ray beams featuring spot sizes of less than 10 nm [54] will enable imaging of patterned dopant structures such as the gates, sources and drains needed for future classical and quantum electronics.

Methods

First-order resonant-reflectometry theory. For a single interface between two materials, the reflectance $R$ for $\sigma$σしぐま polarised light, is given by the Fresnel equation

$$R=\left|\frac{n_{1}\sin\theta_{i}-n_{2}\sin\theta_{t}}{n_{1}\sin\theta_{i}+n_{2}\sin\theta_{t}}\right|^{2},$$θしーた start_POSTSUBSCRIPT italic_i end_POSTSUBSCRIPT - italic_n start_POSTSUBSCRIPT 2 end_POSTSUBSCRIPT roman_sin italic_θしーた start_POSTSUBSCRIPT italic_t end_POSTSUBSCRIPT end_ARG start_ARG italic_n start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT roman_sin italic_θしーた start_POSTSUBSCRIPT italic_i end_POSTSUBSCRIPT + italic_n start_POSTSUBSCRIPT 2 end_POSTSUBSCRIPT roman_sin italic_θしーた start_POSTSUBSCRIPT italic_t end_POSTSUBSCRIPT end_ARG | start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT , (3)

where $\theta_{i}$θしーた start_POSTSUBSCRIPT italic_i end_POSTSUBSCRIPT and $\theta_{t}$θしーた start_POSTSUBSCRIPT italic_t end_POSTSUBSCRIPT are the beam incidence and transmission angles and are equal ($\theta_{i}=\theta_{t}=\theta$θしーた start_POSTSUBSCRIPT italic_i end_POSTSUBSCRIPT = italic_θしーた start_POSTSUBSCRIPT italic_t end_POSTSUBSCRIPT = italic_θしーた) for specular reflection as defined in Fig. 1. $n_{1}$ and $n_{2}$ are the refractive indices of the top and bottom layer, respectively. The reflectance $R$ is related to the material’s atomic scattering factors $f_{1}$ and $f_{2}$ through the refractive index [35, 38]

$$n=1-\frac{r_{0}}{2\pi}\lambda^{2}\sum_{q}N_{q}f_{q}=1-\frac{\lambda^{2}}{2\pi}\rho,$$πぱい end_ARG italic_λらむだ start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT ∑ start_POSTSUBSCRIPT italic_q end_POSTSUBSCRIPT italic_N start_POSTSUBSCRIPT italic_q end_POSTSUBSCRIPT italic_f start_POSTSUBSCRIPT italic_q end_POSTSUBSCRIPT = 1 - divide start_ARG italic_λらむだ start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG start_ARG 2 italic_πぱい end_ARG italic_ρろー , (4)

where $r_{0}$ is the classical electron radius, $\lambda$λらむだ the photon wavelength, $N_{q}$ and $f_{q}=f_{1,q}+if_{2,q}$ are the number of atoms per unit volume and the complex atomic scattering factor for an atom of element $q$, respectively. $\rho$ρろー is the scattering length density. $f_{1}$ as well as $f_{2}$, and hence also $\rho$ρろー, depend on the incident beam energy.

For multiple layers, the reflections at boundaries interfere. Depending on the incident beam angle $\theta$θしーた and the layer thicknesses, this interference can be destructive or constructive. For two interfaces, as shown in Fig. 1, the condition for constructive interference is given by Bragg’s law $2d\sin\theta=m\lambda$θしーた = italic_m italic_λらむだ, and the reflectance is periodic in the wave-vector change on reflection $Q=4\pi\sin\theta/\lambda$πぱい roman_sin italic_θしーた / italic_λらむだ.

In most cases, rather than consisting of perfectly homogeneous and sharp layers, samples have a continuously varying scattering length density profile $\rho(z)$ρろー ( italic_z ). If scattering is weak, multiple scattering events can be neglected, and the reflected signal is simply the sum of the partial waves emanating from the different scatterers driven by the unperturbed incident field. The resulting far-field amplitude is given by the kinematic Born approximation [32, 33, 27],

$$R(Q)=\frac{(4\pi)^{2}}{Q^{2}}\left|\int\rho(z)e^{-iQz}\ \mathrm{d}z\right|^{2}=r_{F}(Q)^{2}|F_{Q}(\rho)|^{2},$$πぱい ) start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG start_ARG italic_Q start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG | ∫ italic_ρろー ( italic_z ) italic_e start_POSTSUPERSCRIPT - italic_i italic_Q italic_z end_POSTSUPERSCRIPT roman_d italic_z | start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT = italic_r start_POSTSUBSCRIPT italic_F end_POSTSUBSCRIPT ( italic_Q ) start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT | italic_F start_POSTSUBSCRIPT italic_Q end_POSTSUBSCRIPT ( italic_ρろー ) | start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT , (5)

which is essentially the squared amplitude of the Fourier transform $F_{Q}(x)$ of $\rho(z)$ρろー ( italic_z ) for a given $Q$-vector times the squared Fresnel reflectivity $r_{F}(Q)^{2}=\frac{(4\pi)^{2}}{Q^{2}}$πぱい ) start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG start_ARG italic_Q start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG. For a continuously varying density profile, the spatial resolution of reflectometry is of the order of the inverse of the greatest $Q$-vector, which is $\approx 0.2$ nm for the 1300 eV X-rays used in this work.

Data collected with X-ray reflectometry are commonly analysed using software that solves Maxwell’s equations throughout the material [56, 39]. To obtain good fits it is necessary to consider each layer’s atomic-species, density, thickness and roughness, resulting in a high number of fitting parameters. However, often one is faced with the problem of characterizing a layer with low contrast hosted by an otherwise known heterostructure, such as nano-electronic stacks involving III-V or group IV semiconductors, where dopant structures are typically buried, such as the silicon and oxide surface layers above our dopant $\delta$δでるた-layers.

We introduce a new method to find and isolate the contribution to X-ray reflection from such a thin layer. First, we consider the general problem of a heterostructure with a scattering length density profile $\rho(z)$ρろー ( italic_z ), to which we add a second scattering length density profile $\delta\rho(z)$δでるた italic_ρろー ( italic_z ), accounting for replacement of atoms (e.g. As for silicon or aluminum for arsenic in III-V devices) from the original heterostructure. The Born approximation, expanding the square at the right of Eq. (5), yields three terms in the reflectivity of the perturbed heterostructure:

$$\frac{R(Q)}{r_{F}(Q)^{2}}=|F_{Q}(\rho)|^{2}+\underbrace{2\Re\left[F_{Q}(\rho)\bar{F_{Q}}(\delta\rho)\right]}_{I(Q)}+|F_{Q}(\delta\rho)|^{2}.$$ρろー ) | start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT + under⏟ start_ARG 2 roman_ℜ [ italic_F start_POSTSUBSCRIPT italic_Q end_POSTSUBSCRIPT ( italic_ρろー ) over¯ start_ARG italic_F start_POSTSUBSCRIPT italic_Q end_POSTSUBSCRIPT end_ARG ( italic_δでるた italic_ρろー ) ] end_ARG start_POSTSUBSCRIPT italic_I ( italic_Q ) end_POSTSUBSCRIPT + | italic_F start_POSTSUBSCRIPT italic_Q end_POSTSUBSCRIPT ( italic_δでるた italic_ρろー ) | start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT . (6)

Eq. (6) contains a background from the unperturbed heterostructure, an interference term $I(Q)$ which is first order in $\delta\rho(z)$δでるた italic_ρろー ( italic_z ), and a signal, of second order in $\delta\rho(z)$δでるた italic_ρろー ( italic_z ), from the perturbation. For a weak perturbation from a small number of atoms, we can ignore the second-order term, but the zeroth-order background remains troublesome as the major contributor to the reflectivity. However, if $\delta\rho(z)$δでるた italic_ρろー ( italic_z ) could be modulated by a known prefactor without affecting $\rho(z)$ρろー ( italic_z ), differences of reflectivities for different prefactors would remove the unperturbed signal, leaving simply the interference term to analyze. The tunable photon energy of a synchrotron source provides a natural modulator for scattering lengths, especially near resonances which are unique fingerprints for particular elements. If the perturbation entails simply replacing atoms of species A with species B, then

$$\mathrm{<ruby><rb>\delta </rb><rt>でるた</rt></ruby>}\mathrm{<ruby><rb>\rho </rb><rt>ろー</rt></ruby>}(z)=r_0\mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}f{N}_{\text{2D}}\mathrm{<ruby><rb>\delta </rb><rt>でるた</rt></ruby>}\widetilde{\mathrm{<ruby><rb>\rho </rb><rt>ろー</rt></ruby>}}(z),$$Δでるた𝑓subscript𝑁2D𝛿~𝜌𝑧\delta\rho(z)=r_{0}\Delta fN_{\text{2D}}\delta\tilde{\rho}(z),italic_δでるた italic_ρろー ( italic_z ) = italic_r start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT roman_Δでるた italic_f italic_N start_POSTSUBSCRIPT 2D end_POSTSUBSCRIPT italic_δでるた over~ start_ARG italic_ρろー end_ARG ( italic_z ) , (7)

where $\mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}f=f_B-f_A$Δでるた𝑓subscript𝑓𝐵subscript𝑓𝐴\Delta f=f_{B}-f_{A}roman_Δでるた italic_f = italic_f start_POSTSUBSCRIPT italic_B end_POSTSUBSCRIPT - italic_f start_POSTSUBSCRIPT italic_A end_POSTSUBSCRIPT, $N_{\text{2D}}\delta\tilde{\rho}(z)$δでるた over~ start_ARG italic_ρろー end_ARG ( italic_z ) is the 3D density of A atoms at depth $z$ replaced by B atoms, $N_{\text{2D}}$ is the 2D density of the layer, and the integral over $z$ of $\delta\tilde{\rho}(z)$δでるた over~ start_ARG italic_ρろー end_ARG ( italic_z ) is unity. This integral is also the long wavelength ($Q\to 0$) limit of the Fourier transform of $\delta\tilde{\rho}(z)$δでるた over~ start_ARG italic_ρろー end_ARG ( italic_z ), so that $N_{\text{2D}}$ sets the scale of the interference term in Eq. (6), which we can rewrite as

$$I(Q)=2r_0{N}_{\text{2D}}\mathrm{\Re }\left[\overline{\mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}f}{F}_Q(\mathrm{<ruby><rb>\rho </rb><rt>ろー</rt></ruby>})\overline{F_Q}(\mathrm{<ruby><rb>\delta </rb><rt>でるた</rt></ruby>}\widetilde{\mathrm{<ruby><rb>\rho </rb><rt>ろー</rt></ruby>}})\right].$$Δでるた𝑓subscript𝐹𝑄𝜌¯subscript𝐹𝑄𝛿~𝜌I(Q)=2r_{0}N_{\text{2D}}\Re\left[\bar{\Delta f}F_{Q}(\rho)\bar{F_{Q}}(\delta% \tilde{\rho})\right].italic_I ( italic_Q ) = 2 italic_r start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT italic_N start_POSTSUBSCRIPT 2D end_POSTSUBSCRIPT roman_ℜ [ over¯ start_ARG roman_Δでるた italic_f end_ARG italic_F start_POSTSUBSCRIPT italic_Q end_POSTSUBSCRIPT ( italic_ρろー ) over¯ start_ARG italic_F start_POSTSUBSCRIPT italic_Q end_POSTSUBSCRIPT end_ARG ( italic_δでるた over~ start_ARG italic_ρろー end_ARG ) ] . (8)

To progress, we assume that for a relatively deep $\delta$δでるた-layer at depth $d$, $\delta\tilde{\rho}(z)=h(z-d)$δでるた over~ start_ARG italic_ρろー end_ARG ( italic_z ) = italic_h ( italic_z - italic_d ), where $h(z)$ is an even function. Hence $F_{Q}(\delta\tilde{\rho})=\exp(-iQd)H(Q)$δでるた over~ start_ARG italic_ρろー end_ARG ) = roman_exp ( - italic_i italic_Q italic_d ) italic_H ( italic_Q ) where $H(Q)$ is real, giving

$$I(Q)=2r_0{N}_{\text{2D}}H(Q)\mathrm{\Re }\left[\overline{\mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}f}\exp (iQd)F_Q(\mathrm{<ruby><rb>\rho </rb><rt>ろー</rt></ruby>})\right].$$Δでるた𝑓𝑖𝑄𝑑subscript𝐹𝑄𝜌I(Q)=2r_{0}N_{\text{2D}}H(Q)\Re\left[\bar{\Delta f}\exp(iQd)F_{Q}(\rho)\right].italic_I ( italic_Q ) = 2 italic_r start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT italic_N start_POSTSUBSCRIPT 2D end_POSTSUBSCRIPT italic_H ( italic_Q ) roman_ℜ [ over¯ start_ARG roman_Δでるた italic_f end_ARG roman_exp ( italic_i italic_Q italic_d ) italic_F start_POSTSUBSCRIPT italic_Q end_POSTSUBSCRIPT ( italic_ρろー ) ] . (9)

If $d$ is the largest characteristic length in the problem, i.e. $d$ is much larger than the thickness of a probable surface oxide layer which contributes to $F_{Q}(\rho)$ρろー ), Eq. (9) gives several results:

1. There are rapid oscillations with period $2\pi/d$πぱい / italic_d and a phase fixed largely by the complex scattering amplitude $\mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}f$Δでるた𝑓\Delta froman_Δでるた italic_f, implying that changes in photon energy will change the phase of the oscillations. On the other hand, changes in $N_{\text{2D}}$ or $H(Q)$ (as long as $h(z)$ remains even) will not change this phase. Furthermore, the third term in Eq. (6) scales as ${|\mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}f|}^2$Δでるた𝑓2|\Delta f|^{2}| roman_Δでるた italic_f | start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT and, therefore, does not undergo a phase shift on changing photon energy. This means that the importance of the interference term in Eq. (6) and the validity of first-order perturbation theory is readily checked by examining the phase shift.

2. Oscillation amplitudes scale by $N_{\text{2D}}\left|\mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}f{F}_Q(\mathrm{<ruby><rb>\rho </rb><rt>ろー</rt></ruby>})\right|H(Q)$Δでるた𝑓subscript𝐹𝑄𝜌𝐻𝑄N_{\text{2D}}\left|\Delta fF_{Q}(\rho)\right|H(Q)italic_N start_POSTSUBSCRIPT 2D end_POSTSUBSCRIPT | roman_Δでるた italic_f italic_F start_POSTSUBSCRIPT italic_Q end_POSTSUBSCRIPT ( italic_ρろー ) | italic_H ( italic_Q ), meaning that if we operate near a resonance of the substitute atom B but far from resonances of the unperturbed heterostructure, then there will be an anomaly at the resonance energy for B.

3. For an infinitesimally thin $\delta$δでるた-layer, $h(z)=\delta(z)$δでるた ( italic_z ) and $H(Q)=1$, the maximum amplitude of the anomaly is $2r_0{N}_{\text{2D}}\left|\mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}f{F}_Q(\mathrm{<ruby><rb>\rho </rb><rt>ろー</rt></ruby>})\right|$Δでるた𝑓subscript𝐹𝑄𝜌2r_{0}N_{\text{2D}}\left|\Delta fF_{Q}(\rho)\right|2 italic_r start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT italic_N start_POSTSUBSCRIPT 2D end_POSTSUBSCRIPT | roman_Δでるた italic_f italic_F start_POSTSUBSCRIPT italic_Q end_POSTSUBSCRIPT ( italic_ρろー ) | independent of the $Q$ sampled. For a Gaussian, $h(z)=1/\left(\sqrt{2\pi}\sigma\right)\exp(-[z/\sigma]^{2}/2)$πぱい end_ARG italic_σしぐま ) roman_exp ( - [ italic_z / italic_σしぐま ] start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT / 2 ), so that $H(Q)=\exp(-[Q\sigma]^{2}/2)$σしぐま ] start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT / 2 ), implying that if $N_{\text{2D}}$ and $|F_{Q}(\rho)|$ρろー ) | are known by other means, the amplitude of the anomaly at any non-zero $Q$ will be reduced by the factor $H(Q)$, from which the $\delta$δでるた-layer thickness parameter $\delta=2\sqrt{2}\pi\sigma$δでるた = 2 square-root start_ARG 2 end_ARG italic_πぱい italic_σしぐま can be extracted.

4. Assuming that $F_{Q}(\rho)$ρろー ) and $H(Q)$ vary slowly with $Q$ on the scale of the oscillations from the $\delta$δでるた-layer, a moving average or low-pass filter with window of width $2\pi/d$πぱい / italic_d will remove the interference term in Eq. (6), leaving an estimate $|F|^{2}_{\mathrm{LF}}$ of the strong zeroth-order term $|F_{Q}(\rho)|^{2}$ρろー ) | start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT. The interference term is then estimated as

   $$I(Q)\approx\frac{R(Q)}{r_{F}(Q)^{2}}-|F|^{2}_{\mathrm{LF}}.$$ (10)

   Thus we can approximate $\delta{\rho}$δでるた italic_ρろー as the inverse Fourier transform of the measured oscillations divided by $\sqrt{|F|^{2}_{\mathrm{LF}}}\approx|F_{Q}(\rho)|^{2}$ρろー ) | start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT, assuming that $F_{Q}(\rho)$ρろー ) is imaginary for non-zero $Q$, which will be the case if $\rho(z)$ρろー ( italic_z ) is a step function at $z=0$, a reasonable approximation for samples of the type considered here:

   $$\bar{\delta{\rho}}=\mathrm{FT}^{-1}\left(\frac{I(Q)}{2\sqrt{|F|^{2}_{\mathrm{LF}}}}\right).$$δでるた italic_ρろー end_ARG = roman_FT start_POSTSUPERSCRIPT - 1 end_POSTSUPERSCRIPT ( divide start_ARG italic_I ( italic_Q ) end_ARG start_ARG 2 square-root start_ARG | italic_F | start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT start_POSTSUBSCRIPT roman_LF end_POSTSUBSCRIPT end_ARG end_ARG ) . (11)

   Finally, for the Gaussian $\delta$δでるた-layer profile discussed under point 3., dividing Eq. (9) by $\sqrt{|F|^{2}_{\mathrm{LF}}}$ gives:

   $$\frac{I(Q)}{\sqrt{{|F|}_{\mathrm{LF}}^2}}=2r_0{N}_{\text{2D}}|\mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}f|\exp \left(-\frac{{(Q\mathrm{<ruby><rb>\sigma </rb><rt>しぐま</rt></ruby>})}^2}{2}\right)\cos (Qd-\varphi ),$$Δでるた𝑓superscript𝑄𝜎22𝑄𝑑italic-ϕ\frac{I(Q)}{\sqrt{|F|^{2}_{\mathrm{LF}}}}=2r_{0}N_{\text{2D}}|\Delta f|\exp% \left(-\frac{(Q\sigma)^{2}}{2}\right)\cos(Qd-\phi),divide start_ARG italic_I ( italic_Q ) end_ARG start_ARG square-root start_ARG | italic_F | start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT start_POSTSUBSCRIPT roman_LF end_POSTSUBSCRIPT end_ARG end_ARG = 2 italic_r start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT italic_N start_POSTSUBSCRIPT 2D end_POSTSUBSCRIPT | roman_Δでるた italic_f | roman_exp ( - divide start_ARG ( italic_Q italic_σしぐま ) start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG start_ARG 2 end_ARG ) roman_cos ( italic_Q italic_d - italic_ϕ ) , (12)

   where $\varphi =\text{arg}(\mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}f)$Δでるた𝑓\phi=\text{arg}(\Delta f)italic_ϕ = arg ( roman_Δでるた italic_f ).

Instead of directly analyzing data as described above or via modelling software, it is also useful to consider differences $\mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}R(Q)=R(E^{\prime },Q)-R(E,Q)$Δでるた𝑅𝑄𝑅superscript𝐸′𝑄𝑅𝐸𝑄\Delta R(Q)=R(E^{\prime},Q)-R(E,Q)roman_Δでるた italic_R ( italic_Q ) = italic_R ( italic_E start_POSTSUPERSCRIPT ′ end_POSTSUPERSCRIPT , italic_Q ) - italic_R ( italic_E , italic_Q ) for two different X-ray energies $E$ and $E^{\prime}$ close to a type B atom resonance, such that changes in terms not containing B atom contributions can be omitted. In terms of the Born approximation this leads to:

${\displaystyle \frac{\mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}R(Q)}{r_F{(Q)}^2}}=2\mathrm{\Re }\left[F_Q(\mathrm{<ruby><rb>\rho </rb><rt>ろー</rt></ruby>})\left(\overline{F_Q^{E^{\prime }}}(\mathrm{<ruby><rb>\delta </rb><rt>でるた</rt></ruby>}\widetilde{\mathrm{<ruby><rb>\rho </rb><rt>ろー</rt></ruby>}})-\overline{F_Q^E}(\mathrm{<ruby><rb>\delta </rb><rt>でるた</rt></ruby>}\widetilde{\mathrm{<ruby><rb>\rho </rb><rt>ろー</rt></ruby>}})\right)\right],$Δでるた𝑅𝑄subscript𝑟𝐹superscript𝑄22subscript𝐹𝑄𝜌¯superscriptsubscript𝐹𝑄superscript𝐸′𝛿~𝜌¯superscriptsubscript𝐹𝑄𝐸𝛿~𝜌\displaystyle\frac{\Delta R(Q)}{r_{F}(Q)^{2}}=2\Re\left[F_{Q}(\rho)\left(\bar{% F_{Q}^{E^{\prime}}}(\delta\tilde{\rho})-\bar{F_{Q}^{E}}(\delta\tilde{\rho})% \right)\right],divide start_ARG roman_Δでるた italic_R ( italic_Q ) end_ARG start_ARG italic_r start_POSTSUBSCRIPT italic_F end_POSTSUBSCRIPT ( italic_Q ) start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG = 2 roman_ℜ [ italic_F start_POSTSUBSCRIPT italic_Q end_POSTSUBSCRIPT ( italic_ρろー ) ( over¯ start_ARG italic_F start_POSTSUBSCRIPT italic_Q end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_E start_POSTSUPERSCRIPT ′ end_POSTSUPERSCRIPT end_POSTSUPERSCRIPT end_ARG ( italic_δでるた over~ start_ARG italic_ρろー end_ARG ) - over¯ start_ARG italic_F start_POSTSUBSCRIPT italic_Q end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_E end_POSTSUPERSCRIPT end_ARG ( italic_δでるた over~ start_ARG italic_ρろー end_ARG ) ) ] , (13)

where $F_{Q}^{E^{\prime}}(\rho)=F_{Q}^{E}(\rho)=F_{Q}(\rho)$ρろー ) = italic_F start_POSTSUBSCRIPT italic_Q end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_E end_POSTSUPERSCRIPT ( italic_ρろー ) = italic_F start_POSTSUBSCRIPT italic_Q end_POSTSUBSCRIPT ( italic_ρろー ) as it does not contain type B atoms. This shows that measuring the difference in reflectometry at two energies near a $\delta$δでるた-layer’s atomic absorption edge isolates the dopant’s signal, dependent to first order only on the change in the dopant’s atomic scattering factors at the absorption edge.

Applying the assumptions that led to Eq. (9), we can recast Eq. (13) as

$$\frac{\mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}R(Q)}{r_F{(Q)}^2}=2r_0{N}_{\text{2D}}H(Q)\mathrm{\Re }\left[\left(\overline{\mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}{f}_{E^{\prime }}}-\overline{\mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}{f}_E}\right)\exp (iQd)F_Q(\mathrm{<ruby><rb>\rho </rb><rt>ろー</rt></ruby>})\right].$$Δでるた𝑅𝑄subscript𝑟𝐹superscript𝑄22subscript𝑟0subscript𝑁2D𝐻𝑄¯Δでるたsubscript𝑓superscript𝐸′¯Δでるたsubscript𝑓𝐸𝑖𝑄𝑑subscript𝐹𝑄𝜌\frac{\Delta R(Q)}{r_{F}(Q)^{2}}=2r_{0}N_{\text{2D}}H(Q)\Re\left[\left(\bar{% \Delta f_{E^{\prime}}}-\bar{\Delta f_{E}}\right)\exp(iQd)F_{Q}(\rho)\right].divide start_ARG roman_Δでるた italic_R ( italic_Q ) end_ARG start_ARG italic_r start_POSTSUBSCRIPT italic_F end_POSTSUBSCRIPT ( italic_Q ) start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG = 2 italic_r start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT italic_N start_POSTSUBSCRIPT 2D end_POSTSUBSCRIPT italic_H ( italic_Q ) roman_ℜ [ ( over¯ start_ARG roman_Δでるた italic_f start_POSTSUBSCRIPT italic_E start_POSTSUPERSCRIPT ′ end_POSTSUPERSCRIPT end_POSTSUBSCRIPT end_ARG - over¯ start_ARG roman_Δでるた italic_f start_POSTSUBSCRIPT italic_E end_POSTSUBSCRIPT end_ARG ) roman_exp ( italic_i italic_Q italic_d ) italic_F start_POSTSUBSCRIPT italic_Q end_POSTSUBSCRIPT ( italic_ρろー ) ] . (14)

This matches Eq. (9), with $I(Q)\to \mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}R(Q)/r_F{(Q)}^2$Δでるた𝑅𝑄subscript𝑟𝐹superscript𝑄2I(Q)\to\Delta R(Q)/r_{F}(Q)^{2}italic_I ( italic_Q ) → roman_Δでるた italic_R ( italic_Q ) / italic_r start_POSTSUBSCRIPT italic_F end_POSTSUBSCRIPT ( italic_Q ) start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT and $\mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}f\to \mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}{f}_{E^{\prime }}-\mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}{f}_E$Δでるた𝑓Δでるたsubscript𝑓superscript𝐸′Δでるたsubscript𝑓𝐸\Delta f\to\Delta f_{E^{\prime}}-\Delta f_{E}roman_Δでるた italic_f → roman_Δでるた italic_f start_POSTSUBSCRIPT italic_E start_POSTSUPERSCRIPT ′ end_POSTSUPERSCRIPT end_POSTSUBSCRIPT - roman_Δでるた italic_f start_POSTSUBSCRIPT italic_E end_POSTSUBSCRIPT, so Eq. (11) & (12) apply with these substitutions. This resonant contrast analysis has the advantage of isolating the interference term [Eq. (13)] without needing to separate signals at different frequencies, as required when analysing reflectometry at a single energy [Eq. (10)].

Sample preparation. Four $2\times 9$ mm² Si(001) samples were diced from Czochralski-grown wafers (bulk doping $<5\times 10^{14}$ cm^-3), and cleaned ultrasonically in acetone, followed by isopropyl alcohol. Each sample was thermally outgassed in vacuum (base pressure $<5\times 10^{-10}$ mbar) for > 8 h at 600 ^∘C, then flash annealed multiple times at 1200 ^∘C using direct current resistive sample heating. This temperature was monitored using an infrared pyrometer (IMPAC IGA50-LO plus) with an uncertainty of $\pm 30~{}^{\circ}$C. Each sample was dosed with AsH₃, to the saturation As density of $(1.6\pm 0.3)\times 10^{14}$ cm^-2 [5], and heated to 350 ^∘C for $1-2$ min to incorporate the arsenic into the silicon lattice [57]. Afterwards, a $1-4$ nm silicon “locking layer” was deposited, without resistive sample heating, to confine the arsenic [58, 5]. Three of the samples were then heated to 500 ^∘C for 15 s, to improve electrical activation, whereas sample #1 was not heated. More silicon ($14-71$ nm depending on sample) was deposited, with the sample at 250 ^∘C. Silicon was deposited at a rate of $0.1-0.4$ nm/min using a silicon solid sublimation source (SUSI-40, MBE Komponenten GmbH). During deposition, the sample temperature was indirectly monitored by the sample resistance, while heating using a direct current resistive sample heater. A control sample with a buried oxide layer was also grown. The process was as above, except without the flash anneal or AsH₃ dose, and the silicon deposition (15 nm) was done without sample heating. All samples and parameters are shown in Tab.  1. The sample in Fig. 2 and 3 was etched with HF prior to the RCXR measurement to make the surface oxide layer thinner, which suppresses an additional slow oscillation to the reflectometry.

#	Dopant	Locking layer	500 ∘C anneal	$d_{\text{SIMS}}$ (nm)	$d_{\text{XRR}}$ (nm)	$\delta_{\text{XRR}}$δでるた start_POSTSUBSCRIPT XRR end_POSTSUBSCRIPT (nm)
1	As	Yes	No	16.0$\pm$1.7	16.8$\pm$1.7	0.8 (0.6–1.2)
2	As	Yes	Yes	15.3$\pm$1.7	17.9$\pm$1.7	0.6 (0.4–1.3)
3	As	Yes	Yes	28.7$\pm$2.3	32$\pm$2	1.4 (1.0–2.0)
4	As	Yes	Yes	70.5$\pm$3.0	77$\pm$3	1.2 (1.0–1.4)
5	SiO2	No	No	13.4$\pm$1.8	15.3$\pm$0.8	–

Table 1: Samples details, depths $d$ and thicknesses $\delta$δでるた measured by SIMS and XRR. SIMS errors are from the uncertainty in the sputter rate. XRR depth errors are from the FWHM of the main peak in the signal’s Fourier transform. The uncertainty on the thicknesses is the 95% confidence interval of fits to Eq. (2).

Figure 5: Arsenic $L_{3}$-edge fits. Black dots correspond to the reflectometry data shown in Fig. 3a. The lines are fits to the data with colours corresponding to a Si:As layer thickness $\delta$δでるた as indicated in the legend.

X-ray scattering. RCXR was performed at the RESOXS endstation of the SIM beamline at the Swiss Light Source synchrotron of the Paul Scherrer Institute [59, 60]. Samples were kept in high vacuum at 10^-8 mbar, and at room temperature. They were mounted on a rotatable holder such that the incidence angle $\theta$θしーた could be swept from $0^{\circ}$ to $90^{\circ}$. The beam energy was set between 1200 and 1400 eV with linear polarization and a spot size set between 500 and 120 µm in the horizontal and 25 to 50 µm in the vertical direction.

Arsenic $L_{3}$-edge reflectometry simulations. To assess the expected change in reflection $\mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}R/R$Δでるた𝑅𝑅\Delta R/Rroman_Δでるた italic_R / italic_R at the arsenic $L_{3}$-edge, we use the DYNA program [39] to obtain the best fit to the resonant spectrum for different Si:As layer thicknesses. The resulting fits are shown in Fig. 5, each taking into account the oxide, silicon, and Si:As layer thicknesses, densities, and roughness. The silicon and oxide density is known, and the density of the Si:As layer is calculated from the 2D density $N_{\text{2D}}$ measured by X-ray fluorescence and the layer’s thickness $\delta$δでるた ($N_{\mathrm{3D}}=N_{\mathrm{2D}}/\delta$δでるた), such that the simulated $\mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}R/R$Δでるた𝑅𝑅\Delta R/Rroman_Δでるた italic_R / italic_R captures the larger arsenic density in thinner Si:As $\delta$δでるた-layers. The surface roughness obtained from the fits varies in the range $0.1-2~{}\text{\AA}$, in agreement with STM measurements where roughness was found to be in the range of 1 Å. The experimental values for $R(1320~{}\text{eV})$ and $R(1330~{}\text{eV})$ are taken as the mean of the data from 1312 to 1322 eV and 1330 to 1340 eV, respectively. The standard deviation is determined in the same intervals, yielding $\mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}R/R=0.09\pm 0.01$Δでるた𝑅𝑅plus-or-minus0.090.01\Delta R/R=0.09\pm 0.01roman_Δでるた italic_R / italic_R = 0.09 ± 0.01. $\mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}R$Δでるた𝑅\Delta Rroman_Δでるた italic_R of the simulation is taken to be $[\text{max}(R)-\text{min}(R)]$ in the interval 1320 to 1340 eV. The values for $\mathrm{<ruby><rb>\Delta </rb><rt>でるた</rt></ruby>}R/R$Δでるた𝑅𝑅\Delta R/Rroman_Δでるた italic_R / italic_R from the fits are shown as black dots on Fig. 3b with a “shape preserving interpolant” line as a guide to the eye.

Secondary ion mass spectrometry. Time-of-flight SIMS (IONTOF ToF-SIMS5) was conducted on all samples with a 25 keV, 1 pA Bi⁺ primary ion beam in high current bunch mode, and a 500 eV, 35–50 nA Cs⁺ sputter beam. Depth profiles were made with a 300 $\times$ 300 Î¼m² sputter crater, within which the analytical region was the central 50 $\times$ 50 Î¼m² or 100 $\times$ 100 Î¼m². The sputter rate was determined by measuring the crater depth with an interference microscope (Zygo NewView NX2). The crater depth was measured from line profiles of the topography in different directions, where the uncertainty was estimated from the standard deviation of these measurements. The results are shown in Fig. 6.

Figure 6: SIMS depth profiles. Arsenic depth profiles of the four Si:As $\delta$δでるた-layers of varying depth, along with the oxide profile of the buried oxide layer (see Tab. 1).

Scanning probe microscopy. STM was conducted using an Omicron variable temperature system. AFM was preformed with a Bruker Dimension Icon with a ScanAsyst-Air cantilever, using peak force tapping mode.