Shallow junction formation via lateral boron autodoping during rapid thermal process

Autodoping is a well-known phenomenon of unwanted dopant transfer in silicon epitaxy process. In this work, we discovered boron lateral autodoping in normal rapid thermal process (RTP) and used it to controllably form shallow junctions for device fabrication. The redeposition of boron from the gas phase to the solid surface was identified to be the limiting step of the boron incorporation into the undoped silicon area in the RTP process. At a given RTP temperature, boron autodoping could be increased by elevating the concentration of the boron source or enhancing the evaporation coefficient. Extending the annealing time can substantially improve the uniformity of the boron concentration in the gas phase, thus reducing the pattern dependence of the autodoping results. In addition, the autodoping process also avoids the traps and defects induced by ion implantation. Therefore, the described mechanism holds great promise for shallow junction formation in selectively patterned area with low cost.


Introduction
Autodoping is a well-known phenomenon of dopant transfer, from one wafer to another or from one region to another region in the same wafer, during silicon epitaxy process [1,2].It is generally classified in two types: vertical autodoping refers to the doping of the epitaxial layer right over the doped area in the * Author to whom any correspondence should be addressed.
Original content from this work may be used under the terms of the Creative Commons Attribution 4.0 licence.Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.starting substrate, while lateral autodoping refers to the doping of the epitaxial layer adjacent to the doped area in the starting substrate.Vertical autodoping is mainly caused by solid-state diffusion.Via redeposition from the gas phase, lateral autodoping normally involves the following sequence: (1) evaporation of dopants from the doped areas to the gas phase; (2) mass transport in the gas phase; (3) dopants incorporation into the growing epitaxial layer.Since autodoping normally generates unintentional doping in the epitaxial layer, which will degrade the performance and reliability of the fabricated devices, it should be removed in the silicon epitaxy process.
In this work, boron lateral autodoping from the adjacent area on a same chip is discovered in the rapid thermal process (RTP).Subsequently, it could be used to controllably form shallow junctions for device fabrication.If a selectively implanted sample is not covered by any cap oxide during the RTP annealing, boron lateral autodoping can form sub-30 nm shallow boron junction in the unimplanted area.This autodoping process can be well blocked with a thin cap oxide.Our numerical model shows the redeposition of the boron from the gas phase to the solid surface is the limiting factor in boron incorporation into the unimplanted area, so the autodoping result is controlled by the boron concentration in the gas phase.At a given annealing temperature, boron autodoping could be enhanced by increasing the gas phase boron concentration, via elevating the concentration of the boron source (the implanted area) or enhancing the evaporation coefficient (with a different gas ambient).Extending the annealing time can also substantially improve the uniformity of the boron concentration in the gas phase, thus reducing the pattern dependence of the autodoping results.In addition, the conventional ion implantation technique inevitably generates points defect and vacancies in the silicon lattice [3].This drawback can be overcome by diffusion-based doping methods such as monolayer doping [4,5] and spin-on doping [6,7].Our demonstrated autodoping process could also avoid the traps and defects induced by ion implantation, which is reflected by the ideal factor of the shallow p + -n diode fabricated using this process.It therefore holds great promise for shallow junction formation in selectively patterned area (for example, piezoresistive elements formation in nanoelectromechanical devices [8]), without the need of advanced doping facility and high process cost, especially for small batches production.

Experimental
The test samples were fabricated using silicon-on-insulator (SOI) substrates from SOITEC.The SOI substrate features a 55 nm-thick lightly p-type doped (∼1 × 10 15 cm −3 ) Si layer on top of a 145 nm-thick buried oxide.The process started with thermal oxidation to form a 5 nm thick cap oxide.Then 8 mm long UVN negative resist strips with varying width ranging from 100 µm to 500 µm were exposed in repeating 1 cm × 1 cm chips on the wafer by electron beam lithography (EBL) as implantation mask.BF 2 + ions with 25 keV energy and at a dose of 1 × 10 15 cm −2 were implanted into the masked wafer.After the UVN photoresist removal, an array of ten four-point probe test structures surrounded by 2 µm wide gaps at their boundaries were patterned by EBL in each of the unimplanted regions, as presented in figure 1(a).The gap was etched by reactive ion etching to completely separate the test structure from the implanted area.Right after the cap oxide stripping in 50:1 HF solution, the samples were loaded to a RTP furnace (RT600s system from Modular Process Technology Corporation) for annealing at 950 • C in 80 psi N 2 ambient.The ramp rate of the RTP is about 45 • C s −1 .The system was rapidly cooled to room temperature after the annealing.Time of flight secondary ion mass spectrometry (ToF-SIMS) measurements were performed using a ToF SIMS V Instrument from ION TOF GmBH equipped with a 30 kV bismuth-liquid metal ion gun (LMIG).The electrical characterizations of the test structures were done at room temperature in a probe station connected with a Keysight B1500A precision semiconductor analyzer.

Results and discussion
The sequence of the lateral boron autodoping process during the RTP is illustrated in figure 1(b): evaporation of boron from the implanted area to the gas phase (step 1); boron transport in the gas phase (step 2); boron redeposition to the surface of the unimplanted area (step 3); and boron diffusion from the surface to the bulk of the unimplanted area (step 4).The autodoping result is confirmed by quick two-terminal tests between V s (grounded) and V d on the four-point test structure.The I d vs V d curves for samples under different process conditions are shown in figure 1(c).For the sample without cap oxide, the measured I d increases linearly with increasing V d and reaches 100 nA range at 0.1 V V d .This measured conductivity indicates the unimplanted area is doped by boron via the ambient, since the solid-state diffusion to the test structure is blocked by the isolation gap.Furthermore, the ohmic contact behavior to the test structure also implies high concentrations of surface dopants.The autodoping can be well blocked by a 5 nm cap oxide, as evidenced by the I d -V d plot measured in an otherwise identical sample capped with 5 nm SiO 2 .The measured current is fluctuating at noise floor of the measurement (pA range).This observation is consistent with the published fact that the boron out-diffusion can be sufficiently suppressed by injecting small amount of O 2 into N 2 ambient during rapid thermal annealing (RTA) to form cap SiO 2 on surface [9].To further rule out possible dopant sources in the RTP furnace, a reference chip without any implanted area or cap oxide was annealed at the same annealing condition.The measured current is still in the pA range, which confirms that the implanted area is the source of dopants in the autodoping process.Boron evaporation from silicon to N 2 ambient during RTP process was indicated in many published work [10][11][12], although the boron transport coefficient at silicon to ambient interfaces in N 2 is much lower than H 2 ambient [13].
To further confirm boron autodoping in the unimplanted area, the implanted samples were annealed at 950 • C in RTP for different times after removal of the cap oxide.Additionally, ToF-SIMS measurements were carried out in the center of unimplanted area with 100 µm width.No detectable fluorine signal was found in the unimplanted area, confirming this area was very well blocked by the photoresist mask during the BF 2 implantation process.No boron signal was detected in the unimplanted area of a control sample before the RTP annealing, either.In direct contrast, in the annealed samples boron concentration reaches 10 19 cm −3 at the surface of the unimplanted area and declines to 5 × 10 17 cm −3 (boron detection limit in the ToF-SIMS analysis) with increasing depth, as shown in figure 2(a).The boron depth profiles also exhibit a clear dependence on the annealing time.These results confirm that boron dopants indeed were transported from the heavily implanted area (see the black line in figure 2(a)) to the unimplanted area via the gas phase, as illustrated in figure 1(b).
The boron lateral autodoping process can be numerically simulated by a simple model.In the first step, boron dopants are evaporated into pure nitrogen ambient under annealing.In the immediate vicinity of silicon surface, the boron gas phase concentration, C m , should be equal to C h /k e , where C h is surface boron concentration of the heavily implanted area and k e is the effective segregation coefficient between the gas and solid [1,2].Since the area of the unimplanted region (100 µm width) is significantly smaller than the surrounding heavily implanted area, the evaporated boron is assumed to be uniformly distributed over the unimplanted region.The diffusion process of the redeposited boron (from the gas ambient) in unimplanted silicon can be described with Fick's second law in one dimension: The diffusivity, D, is assumed to be independent of the concentration.C(x, t) represents the boron concentration at location of x from silicon surface, and t is the annealing time in RTP.The boundary conditions of the diffusion process are as follows: where h is the boron evaporation coefficient [1].The differential equation can be solved as [14]: As discussed before, k e C m is equal to the surface boron concentration of the heavily implanted area, 1.3 × 10 20 cm −3 according to the SIMS results.It is well accepted that the evaporation coefficient h of boron in N 2 is much lower than that in H 2 (8 × 10 −8 cm s −1 at 1188 • C) [2].The reported boron diffusivity in silicon is about 2 × 10 −14 cm 2 s −1 [15,16].Therefore, the exponential term, exp ) approximates to 1. Equation ( 4) is then simplified to: .
(5) As shown by the solid lines in figure 2(a), all the three boron diffusion profiles from 10 s to 30 s RTP annealing can be fitted with equation (5).Furthermore, the averaged boron diffusivity, D, extracted from the fitting is 2.2 ± 0.5 × 10 −14 cm 2 s −1 , which is close to reported values [15,16] The extracted evaporation coefficient, h, is 2.8 ± 0.3 × 10 −9 cm s −1 .
From equation ( 5), the redeposited surface boron concentration in the unimplanted area C(0, t) can be calculated as shown in equation ( 6) . (6) For a given RTP time, the surface boron concentration, C(0, t), is determined by h √ t/D.Please note that h √ t/D reflects the ratio of the boron transfer rate at the gas-solid interface (ht) to the diffusion rate in the bulk silicon ( √ Dt).Based on equation ( 6), the ratio of C(0, t) to the surface boron concentration in the heavily implanted area (C h = k e C m ) versus time is calculated and plotted in figure 2(b).The solid line in figure 2(b) represents the calculated C(0, t)/C h under our experimental condition with h = 2.8 × 10 −9 cm s −1 .It clearly evidences that extending annealing time is not an efficient way to increase surface boron concentration in the unimplanted area under this condition.For example, it takes about 100 s to double the surface concentration results of 10 s annealing.This indicates the boron transfer at the gas-solid interface is the rate limiting step in process of boron incorporation into the undoped silicon.To further illustrate the limiting effect of gas-solid interface transfer, we also calculate the normalized surface boron concentration versus annealing time with five and ten times higher h (see dashed lines in figure 2(b)).Indeed, the surface boron concentration can be significantly elevated by increasing the boron transfer rate at the gas-solid interface.Higher gas-to-solid boron transfer rate could be realized by using a different gas ambient (such as H 2 ) [13].Alternatively, using a stronger source, i.e., heavier doping of the implanted area (higher C h ), is another efficient way to enhance the autodoping effect.
To validate the assumption that the evaporated boron uniformly distributed in the gas phase over the unimplanted region, we tested the array of test structures (figure 3(a)) in 8 mm long and 100 µm wide unimplanted areas located in the middle of the 1 cm × 1 cm chip, after 10 s RTP annealing.The sheet resistance extracted from each test structure is plotted versus their position on x axis, as shown figure 3(a).If the RTP annealing was done on a single 1 cm × 1 cm chip, the sheet resistance reaches minimum at the chip center and almost doubles when the test structure is close to chip edges.Since the width of the unimplanted area is significantly smaller than the dimension of the chip (1 cm), the chip center can be regarded as being fully surrounded by heavily implanted regions.Therefore, boron is readily supplied from the gas phase of the whole surrounded area.However, test structures close to chip edge can only receive dopants from one side, requiring longer time for the gas phase above to form balance with that on the heavily implanted area (C m ).This leads to less boron redeposition in the chip edge.This was further confirmed by annealing two adjacent chips (1 cm × 2 cm chip) together, as shown in figure 3(a).As expected, the sheet resistance measured in the center of the new chip is lower, despite that the test structure is at the edge of the unimplanted area.
Similar sheet resistance distribution also appeared in the unimplanted areas with different width after 10 s RTP annealing.As shown in the solid lines in figure 3(b), sheet resistance increases when the test structure is closer to the chip edge.Sheet resistance results almost overlap in 100 µm and in 300 µm wide unimplanted areas.However, it is higher in the 500 µm wide unimplanted area.This behavior could also be caused by the difference of the gas phase boron concentration in the initial stage of the RTP, due to a variation in geometry.Since the gas phase will be much more uniform with maximum boron concentration after the gas phase is balanced with C m on the heavily doped area, slight extending the annealing time under this condition with uniform and higher boron redeposition rate could potentially overwhelm nonuniformity generated at the initial stage of the RTP.Indeed, after the annealing time was extended to 30 s, sheet resistance measured in different locations of the unimplanted areas with different width became much more uniform (see the dashed lines in figure 3(b)).With measured sheet resistance, the boron activation rate after rapid thermal processing could be roughly estimated by comparing it with the calculated value from the integrated ToF-SIMS boron depth profile in figure 2(a) (assuming 100% activation).The estimated activation rate after the 30 s process is roughly 60%.These results evidence that the use of the boron lateral self-doping process holds great promise to form uniform shallow junctions without the need for advanced doping facility and high process cost.
Using the demonstrated autodoping process, we fabricated shallow p + -n diode with 100 µm diameter on an n type substrate (525 µm phosphorus-doped wafer with 1.32-1.98Ω•cm resistivity).As illustrated in figure 4, the unimplanted region was first doped using the boron autodoping process with 30 s RTP annealing.Then the diodes were patterned in the unimplanted area and isolated by a 250 nm trench using reactive ion etching.Finally, the diodes were metallized using 100 nmthick Al on substrate with 1 cm 2 contact area.The large area in the substrate ensures that the junction current is not limited by the back contact.A reference diode was also processed without any heavily implanted area on the chip for comparison.The current (I)-voltage (V) characteristics of the diodes are shown in figure (4).The one without doping source during the RTP shows typical Schottky junction behavior with close to 10 −5 A saturation current under −2 V reverse bias.Since no boron source was available during the RTP annealing, no p type junction could be formed.The tungsten probe tips were therefore directly contacting the lightly doped n type silicon substrate during the electrical characterization, thus measuring Schottky junction behavior.As a direct contract, the diode with boron autodoping shows an ideal p + -n diode behavior, with an ideal factor of 1.1.The saturation current was reduced to 10 −11 A at −2 V revers bias.The ideal factor indicates negligible Shockley-Read-Hall recombination in the diode.This shows additional advantage of the demonstrated boron autodoping technique, which could avoid implantation induced traps and damages in the formed junction.Despite the fact that ion implantation can provide better control on the effective doping profile of the incorporated boron, the demonstrated autodoping technique still holds great promise for shallow junction formation without the need of advanced doping facility and high process cost, especially for small batches production.In addition, since boron autodoping can be fully blocked by a thin layer of cap oxide, the scalability of the process could be enabled by well-established patterning techniques, which could precisely open autodoping windows through cap oxide with size down to sub-100 nm range.

Conclusion
We highlighted boron lateral autodoping in normal RTP and used it to form shallow junctions with good uniformity for device fabrication.The redeposition of the boron from the gas phase to the solid surface was identified to be the limiting step of the boron incorporation into the undoped silicon area in the RTP process.The autodoping could therefore be engineered by modulating the boron concentration in the gas phase or enhancing the boron transfer rate the gas-solid interface.Moreover, the autodoping process also avoids the traps and defects induced by implantation process.It therefore holds great promise for shallow junction formation in selectively patterned area, without the need of advanced doping facility and high process cost, especially for small batches production.

Data availability statement
All data that support the findings of this study are included within the article (and any supplementary files).

Figure 1 .
Figure 1.(a) Left: optical image of a four-point test structure surrounded by a 2 µm wide isolation gap in the unimplanted area; right: zoom-in view of the isolation gap.(b) Schematic of boron autodoping process including boron evaporation from heavily implanted area (step 1), transport in gas phase (step 2), redeposition to unimplanted surface (step 3) and solid phase diffusion (step 4).(c) Two terminal I d vs V d test results on the test structure processed under different RTP conditions.

Figure 2 .
Figure 2. (a) Boron ToF-SIMS depth profiles and their fitting curves of the center of the 100 µm wide unimplanted area on samples with different RTP annealing time (10 s, 20 s and 30 s).(b) Simulated plots of normalized surface boron concentration vs. annealing time for boron autodoping in 950 • C RTP with D = 2.2 × 10 −14 cm 2 s −1 and different h values.The solid line represents our experimental condition, and the red stars are experimental results taken from the ToF-SIMS plots.The dashed lines represent assumed conditions with higher h values.

Figure 3 .
Figure 3. Sheet resistance mapping unimplanted region with W window = 100 µm after 10 s RTP annealing at 950 • C, on a single chip and two adjacent chips respectively (a), and with different W window and annealing time at 950 • C on single chip (b).

Figure 4 .
Figure 4. I-V characteristics of diodes with 100 µm diameter fabricated with different annealing time and without dopants source.