Effects of oxygen-inserted layers and oxide capping layer on dopant activation for the formation of ultrashallow p-n junctions in silicon

Planar bulk-silicon (bulk-Si) metal-oxide-semiconductor field-effect transistors with sub-50 nm gate lengths must have ultrashallow (sub-25 nm) source and drain (S/D) extension regions to adequately suppress short-channel effects.^1,2 To achieve low parasitic S/D resistance, the active dopant concentration in the heavily doped S/D extension regions must be high. A recent study of ultrashallow junction (USJ) formation via ion implantation followed by rapid thermal annealing (RTA) showed that the incorporation of partial monolayers of oxygen into crystalline silicon is effective for impeding interstitial-assisted dopant diffusion, resulting in shallower junction depth $(XJ)$ and reduced dopant loss.³ Since that study involved only secondary ion mass spectroscopy (SIMS) analyses, however, the impact of the oxygen-inserted (OI) layers on dopant activation could not be ascertained. In this work, sheet resistance $(Rsh)$ measurements together with technology computer-aided design (TCAD) simulations are used to determine the effect of OI layers on dopant activation in USJ regions. In addition, the effect of a low-temperature-deposited oxide (LTO) capping layer on dopant activation in USJs is investigated.

Spreading resistance profiling (SRP) and four-point probe measurements are known to be inaccurate for electrical characterization of p-n junctions with sub-80 nm $XJ$⁠, due to probe penetration and carrier spilling effects that make the apparent electrical junction depth shallower than the metallurgical junction depth $(XJ)$⁠.⁴ Other state-of-the-art characterization techniques such as micro-four-point probe,⁵ sheet resistance and leakage probe,⁶ and elastic metal four-point probe⁷ have also been shown to be inaccurate for determining $Rsh$ of heavily doped regions with $XJ$ below 15 nm.⁴ In this work, we present a new method for accurately determining $Rsh$ from electrical measurements and TCAD simulations. This $Rsh$ extraction method is used to demonstrate that OI layers can provide for lower $Rsh$ with reduced $XJ$⁠, whereas an LTO capping layer has an opposite effect.

Ultrashallow p-n junctions were formed in p-type (001) Si wafer substrates by high-dose (1 × 10¹⁵ cm⁻²) ion implantation through a 2 nm-thick screening oxide layer, with 0° tilt and acceleration energies chosen such that the projected range $(Rp)$ is less than 20 nm: BF₂ ions were implanted with 90 keV energy, phosphorus (P) ions were implanted with 1 keV energy, and arsenic (As) ions were implanted with 1 keV energy. For the boron (B) doped samples, the substrate was also doped with a 90 keV, 2 × 10¹³ cm⁻² As ion implant to form a n-type well. The OI samples comprise several partial monolayers of oxygen located at a depth approximately10 nm below the Si wafer surface. After they were implanted, the wafers were diced into $2×2cm2$ chips, some of which were subsequently coated with a 10 nm-thick layer of silicon dioxide formed by chemical vapor deposition at a relatively low process temperature of 350 °C. To activate the implanted dopants, a 1050 °C “spike” annealed with a ramp-up rate of 116 °C/s was performed using an AccuThermo AW610 RTA tool.

To fabricate $Rsh$ test structures, annealed chips were first cleaned in a sulfuric peroxide mixture (H₂SO₄:H₂O₂) bath at 120 °C for 10 min, followed by a 1-min dip in the dilute hydrofluoric acid (DHF) solution (10:1 H₂O:49% HF) to remove the LTO and/or native oxide. Figure 1 illustrates the fabrication process steps used to form pairs of $100×100μm2$ metal contact pads on the surface of the chips, with spacing ranging from 400 to 1400 μm.

The $Rsh$ extraction method presented herein is reminiscent of the variable probe-spacing measurement method⁸ in which SRP probe tips of radius 1 μm are forced into contact with the sample surface, and values of spreading resistance $(Rsp)$ for various probe tip separations (e.g., 25, 50, 100, 200, 400, and 800 μm) are measured. A probe tip penetration depth of 5 nm has been measured by atomic force microscope for a loading force of 5 g,⁹ which affects the accuracy of $Rsp$ measurements, as demonstrated by the example presented below. Also in previous studies, $Rsh$ has been assumed to be directly proportional to $Rsp$⁠. Specifically, $Rsh$ is simply calculated as the product of $π$ and the slope of $Rsp$ versus the natural logarithm of the probe separation, i.e., $Rsh=π×dRsp/dln⁡(separation)$⁠. Figure 2 shows TCAD simulated $Rsp$ data points for various probe tip separation values for two USJ cases, each assuming an ideal step-function doping profile and 100% dopant activation. (The probe resistance and metal-semiconductor contact resistance are each assumed to be negligible compared to $Rsp$⁠, which is reasonable given the very high doping level at the metal-semiconductor junction and the relatively large tip probe separation.) Table I shows that $Rsh$ is not directly proportional to $Rsp$ and that the values of $Rsh$ extracted from the least squares fitted lines in Fig. 2 are inaccurate even when the probe penetration depth (5 nm) is much shallower than $XJ$ (100 nm).

In this work, the probe tip penetration problem is circumvented by depositing metal contact pads onto the Si surface; these pads can be probed without penetrating the Si. For the $Rsp$ measurements, the samples were maintained at 25 °C and the applied voltage was swept from −10 to 10 mV. Ohmic contact behavior was observed for all samples. $Rsh$ is extracted by fitting (using the least squares method) the measured $Rsp$ versus pad separation data to TCAD simulations as shown in Figs. 3–5 for boron-doped, phosphorus-doped, and arsenic-doped USJs, respectively. Each data point in these figures represents the average of measured $Rsp$ values for ten different test sites. The standard deviation of these measurements is too small to be shown.

Dopant activation is complicated by phenomena such as clustering, precipitation, and segregation which can limit the dopant solid solubility, $Nmax,active$⁠.^10–12 In this work, the value of $Nmax,active$ is determined by matching TCAD simulations to measured data, using dopant depth profiles obtained from SIMS analyses and assuming 100% dopant activation at concentrations below $Nmax,active$⁠.

Figure 6 shows the simulated dependence of $Rsh$ on $Nmax,active$ for each of the boron-doped USJs fabricated in this work. The symbols indicate the fitted values of $Rsh$ (cf. Fig. 3) and the corresponding values of $Nmax,active$⁠. Table II provides a summary comparison of $XJ$ (defined at $5×1018cm−3$⁠), retained boron dose, and extracted $Rsh$ and $Nmax,active$ values. The $Nmax,active$ values are higher than the plateau level of the diffusion tail, consistent with previous work by Jain et al.¹³ It can be seen that the OI layers provide for lower $Rsh$ even if $Nmax,active$ is somewhat lower, due to greater retention of the implanted B atoms. The OI layers impede the diffusion of interstitial Si atoms (I) into the substrate and thereby retard B diffusion away from the surface.³ Because the B–I clustering mechanism is responsible for B deactivation,¹⁴ lower B activation is expected with the OI layers, which can explain the lower $Nmax,active$ for the uncapped OI sample versus the uncapped control sample. $Nmax,active$ is lower for the LTO capped samples versus the uncapped samples, but much less so for the OI samples. B deactivation has been found to correlate with uphill diffusion, caused by the diffusion of Si interstitials toward the surface.¹⁵ As Si interstitials diffuse toward the surface, they can kick out substitutional B (dose loss) or form B–I clusters (deactivation). In addition, they can recombine with vacancies at the free surface for uncapped samples. This recombination process can help to reduce the contributions of Si interstitials to B dose loss and deactivation. It can be expected that B dose loss and B deactivation are worse for LTO capped samples, consistent with the observed steeper B concentration gradient near the surface and lower $Nmax,active$⁠. Since it is energetically favorable for Si interstitials to locate around OI layers, B deactivation caused by an LTO cap is much less for OI samples than for control samples because OI layers impede diffusion of Si interstitials to the surface.

Figure 7 shows the simulated dependence of $Rsh$ on $Nmax,active$ for each of the phosphorus-doped USJs fabricated in this work. The symbols indicate the fitted values of $Rsh$ (cf. Fig. 4) and the corresponding values of $Nmax,active$⁠. Table III provides a summary comparison of $XJ$⁠, retained phosphorus dose, and extracted $Rsh$ and $Nmax,active$ values. From a comparison of the retained dopant doses in uncapped samples in Tables II–IV, it can be seen that P is most susceptible to dose loss caused by uphill diffusion toward the surface, which serves as a sink for interstitials during RTA.^16,17 Since OI layers impede interstitial diffusion into the substrate, uphill diffusion and hence P dose loss are enhanced for the uncapped OI sample. Due to lower P concentration, $Nmax,active$ is lower with OI layers in uncapped samples. P dose loss is significantly reduced with an LTO capping layer because P prefers to remain in Si.¹⁸ The presence of OI layers results in higher $Nmax,active$ for LTO capped samples due to retarded diffusion of P away from the surface.

Figure 8 shows the simulated dependence of $Rsh$ on $Nmax,active$ for each of the arsenic-doped USJs fabricated in this work. The symbols indicate the fitted values of $Rsh$ (cf. Fig. 5) and the corresponding values of $Nmax,active$⁠. Table IV provides a summary comparison of $XJ$⁠, retained arsenic dose, and extracted $Rsh$ and $Nmax,active$ values. It can be seen that $Nmax,active$ levels are consistently higher in the OI samples than in the control samples, in correlation with the As concentration. The impacts of an LTO capping layer on $XJ$ and $Nmax,active$ are less for As than for B and P, because As diffusion is only 40% interstitial-driven at 1050 °C (Ref. 19) so that the free surface acting as a sink of interstitials has lesser impact on As diffusion. Because of higher $Nmax,active$ and higher retained As dose, the presence of OI layers is beneficial for achieving low $Rsh$⁠.

Figures 9–11 show the effects of OI layers and an LTO capping layer on $Rsh$ and $XJ$ for B, P, and As doped USJs, respectively. It can be seen that OI layers are beneficial for reducing $XJ$ and also can provide for lower $Rsh$ for B and As doped junctions, thanks to higher retention of dopants. For P doped junctions, the increase in $Rsh$ is smaller than that which ordinarily would be expected for the reduction in $XJ$⁠. This can be seen from Table V, which tabulates the product of $Rsh$ and $XJ$⁠, associated with the resistivity of the heavily doped region. $Rsh×XJ$ is consistently lower for the OI samples versus the control samples. In contrast, $Rsh×XJ$ is consistently higher for LTO capped samples due to the reduced peak dopant concentration.

The effects of subsurface OI layers and an LTO capping layer on dopant activation in USJs formed by high-dose ion implantation followed by spike annealing at 1050 °C are studied in this work. The OI layers are beneficial for achieving shallower junctions with low sheet resistance $(Rsh)$ because they reduce dopant diffusion away from the surface and can provide for higher levels of active dopant concentration. An LTO capping layer is found to result in larger $Rsh$ due to the lower peak active dopant concentration. The presence of OI layers is found to mitigate this detrimental effect.