dEvans Texas, Round Rock TX 78681 e Sematech, Austin TX 78702. Abstract. Monte Carlo ion-implant models for germanium and indium implantation into ...
Monte Carlo Simulation of Heavy Species (Indium and Germanium) Ion Implantation into Silicon Y. Chen
a
B. Obradovic ", M. Morris b, G. Wang a, G. Balamurugan" ,D. Li a, A. F. Tasch a, D.Kamenitsa C, W. McCoyC, S. Baumann" R. Bleier d, D. Sieloff", D. Dyer b, P. Zeitzoff" a,
Microelectronic Research Center, The University of Texas at Austin, Austin TX, 78712 b Motorola, Austin TX 78721 C Eaton Corporation, Boston MA 78758 d Evans Texas, Round Rock TX 78681 e Sematech, Austin TX 78702
Abstract Monte Carlo ion-implant models for germanium and indium implantation into singlecrystal silicon have been developed and are described in this paper. The models have been incorporated in the UT-MARLOWE ion implantation simulator, and have been developed primarily for use on engineering workstations. These models provide the required as-implanted impurity profiles as well as damage profiles, which can be used as inputs for transient enhanced diffusion simulation and subsequent multiple implant simulation. A comparison of simulation results with experimental data shows that the models predict both the impurity profiles and the damage profiles very successfully for a wide range of implant conditions. The damage profiles from germanium implant simulations have been used for subsequent multiple implant simulations and excellent agreement with experimental results has been achieved.
Introduction As integrated circuit devices scale into the deep sub-micron regime, ion implantation will continue to be the primary means of introducing dopant atoms into silicon. Different types of impurity profiles such as ultra-shallow profiles or retro-grade profiles are necessary for deep submicron devices in order to realize the desired device performance. In order to fulfill these requirements, heavy species implants are used more and more frequently. For example, germanium implantation has two major applications in deep submicron device fabrication: one is for pre-arnorphization a method of suppressing channeling tails and achieving shallow impurity profiles; the other application is to use a high dose germanium implant to formhetero-structure SiGe MOSFET[I-3]. Also, indium and antimony implants have been used to provide retrograde profiles in MOS devices. Through implantation of large mass atoms such as indium, very abrupt orretro-grade doping profiles can be achieved. The retrograde doping profiles can serve the needs of channel engineering in deep submicron MOS devices for punch-through suppression and threshold voltage control. In addition to retrograde profile and low coefficient of diffusion at high temperatures, indium has been shown [4] to have several other advantages compared to boron : indium implanted channel doping results in a lower silicon surface roughness and a higher channel mobility, resulting in better turn-off behavior and drive current in ultra-short-channel MOS devices. 1
The development of ion implant models for germanium and indium not only will facilitate the applications of indium and germanium implants, but also will provide a basis for modeling other heavy species implants into silicon. Previously, models have been developed and reported for boron, BF2, arsenic, phosphorus and silicon self implants[5-9]. Arsenic is the heaviest ion species for which ion implant models have been developed. Because germanium is very close to arsenic in mass, it is expected that the model parameters for germanium should be close to those of arsenic. Therefore, germanium modeling provides a case where the strength of the physical components of the previously developed models can be tested. With the success of indium modeling, which extends the range of the ion mass covered by our previous models, models for other species such as antimony may be generated by taking the approach of interpolation or extrapolation with a relatively small amount of effort. The objectives of modeling heavy species ion implants into single-crystal silicon are as follows: first we expect the models to accurately predict the as-implanted impurity profiles over a wide range of ion implant conditions; second, we expect the models to predict the damage profiles accurately. The damage profile prediction is very important for the following reasons: First, the defect dechanneling effect needs to be simulated so that the impurity profiles as a function of dose can be accurately predicted. Second, accurate damage profiles are needed so that the damage profiles can be used as inputs for transient enhanced diffusion (TED) simulation. Third, damage profiles are needed in the simulation of multiple implants, in which the existing damage from the initial implant has a large effect on the impurity and damage profiles of subsequent implants. As ions enter silicon, they give up their energy to the lattice atoms and finally come to rest at certain depths in the silicon. The energy loss of an ion can be divided into two components: nuclear stopping and electronic stopping. For Monte Carlo simulation, the nuclear stopping is treated with the binary collision approximation (BCA). The energy is also lost to the electronic cloud in two ways: one is due to the inelastic loss during a collision with silicon lattice atom and the other is inelastic loss between collisions. In all of the important modeling components, which include nuclear stopping, electronic stopping, damage generation and interaction, it is expected that heavy ions may exhibit some different behavior from the lighter ones. Before describing the models, the details are given for the experiments that were performed in order to provide experimental data for comparison with the prediction of the models.
Experimental Details For the development and verification of the models, a detailed experimental study was conducted to understand the germanium and indium profile dependence on key implant parameters such as energy, tilt/rotation angle, and dose. Over forty l50mm bare silicon wafers were implanted with germanium. An HF dip was performed before each implant in order to remove all native oxide and to suppress rapid re-growth of the native oxide prior to implantation. The energies used were lOkeY, 20keV, 40keV and 80keV, and the
2
doses ranged from 5 xI0 12cm-2 to 2xI0 15 cm-2 . The tilt/rotation angles ranged from 0%° to 6°/30°. The indium implants into bare silicon were performed over an energy range of 5-200keV, and included a similar range of doses and implant angles. The tilt angles were measured relative to the direction perpendicular to the [100] wafer surface. The rotation angle measures the rotation of the wafer about an axis perpendicular to the wafer surface. These two angles together determine the incident angle of the implant ion beam. The implants with tilt angles near {) are not only highly sensitive to the tilt/rotation angles, but are also affected by the crystal cut-error, which is the angular deviation of the [100] direction in the crystal from the normal direction of silicon wafer. In order to correctly account for the crystal cut-error, samples from the 8 wafers with on-axis(O° tilt angle) implants were measured using X-ray Scitag Spectrostroscopy. The measured crystal cut-errors were used to calculate the real tilt/rotation angles relative to the [100] crystal orientation. In this way, the crystal cut errors are properly taken into account. The implanted wafers were then diced into 5mm dumbbell configuration has the lowest energy for a neutral interstitial silicon atom. In this configuration, two silicon atoms displace along the (110) direction by about one bond distance, sharing one single lattice site. If a projectile encounters an interstitial, the interstitial is assumed to be in a dumbbell configuration and a normal scattering event takes place. The separation between the two ''split'' atoms is taken as 2.76 A, and for simplicity, the other atoms are assumed to remain in their lattice sites. A random configuration is used byHobler et al. [19], wherein the interstitial is placed randomly in the plane perpendicular to the ions direction of motion within a sphere with a specified radius centered at the lattice site. In practice, this approach can be used without significantly changing the simulation results. However, this introduces another parameter, i.e., the radius of the distribution sphere of the interstitials. The basic assumption in the modified Kinchin-Pease model is that the trajectories of the secondary ions are much shorter compared with that of the primary ion. When an ion is implanted into a silicon crystal at higher energies, a large amount of kinetic energy can be transferred to a target atom in a single collision. This occurs more frequently for heavy species implants. The damage models are implemented in such a way that the silicon crystal is divided into many cells with a fixed volume. The energy loss in each collision is converted into a defect number in a local regior( a single cell) according to Eq. (7). Therefore, if the deposit energy is too high, and the energy loss during a collision is deposited in a single cell, this local region can be totallyamorphized as the result of a single collision. Although a high-energy impact may cause significant damage in a local region, amorphization of a local region due to a single collision is not realistic. In fact, when the deposited energy is larger than a certain valutf22], the energy loss from
7
electronic excitation would exceed the energy loss from all other sources, and the energy lost in this way will not cause as much damage as calculated byEq. (7). For this reason, the cutoff energy was introduced into the damage models in order to account for this effect. Whenever the deposited energy E is more than Ecut-ojJ, only Ecut-ojf is deposited and the number of defects is thus calculated. WithEcut-ojf being roughly the energy needed to amorphize the local region (125 unit cells) for boron implants, this local region cannot be amorphized due to one single collision since ~ calculated from Eq. (8) is always less than N a. Because of the significant influence of damage on the impurity profiles at high energies, Ecut-ojf is adjusted for each species in order to obtain the best overall agreement between simulation results and experimental data. Its value is approximately 3keV for the current local region size(125 cells). It should be noted, however, that this approach still effectively ignores the large amount of damage scattered in a deeper region, which is caused by the high kinetic energies of secondarySi atoms. A more physical approach would be to follow the secondary projectiles that receive high kinetic energies until their energies fall below a certain predefined value, and then replace them with the KinchinPease damage model. On the other hand, this increases the CPU time considerably. In order to keep the computational time at a reasonable minimum the deposit energy cut-off approach is chosen for the energy range to which the K-P model is applied. 3. Damage Modeling 2- -KADM Model Description
The Kinetic Accumulative Damage Model (KADM) was first developed byTian [5]. In this model, the physical processes during ion implantation are divided into two phases due to different time scales. The first phase is defect production, while the second phase is defect diffusion and interaction during the time between incident ions. The first phase (displacement cascade) takes place in less than 1pico-second, while the second phase happens shortly after the displacement cascade and lasts until the next ion comes to the same damage region. This phase lasts for approximately 0.1 millisecond or longer for typical dose rates. The algorithm of this damage model is as follows: For the first ion, the primary damage state is generated by the binary collision approximation (BCA) code. This damage information is then fed into a module, which simulates defect diffusion and reactions for an amount of time determined by the dose rate and the ion implanter scanning pattern. Subsequent ions are allowed to collide with the previous damage, and the damage thus generated is accumulated over the previous damage with near neighbor defect reactions being taken into account. The resulting (mobile) defects then diffuse and react for a certain amount of time, which again is determined by the dose rate and the scanning pattern. This cycle is repeated until all of the ions are implanted (simulated) for a given dose. The defect diffusion is simulated by a Monte Carloalgorithm which is based on defect hopping. The defect interactions are governed by "interaction rules", which include interstitial-vacancy recombination, clustering of same type of defects, defect-impurity complex formation, emission of mobile defects from clusters, surface effects, etc. Similar to the Kinchin-Pease model, a 10% critical defect density is assumed to be the ion beam induced amorphization threshold.
8
Ion implantation simulators have traditionally used a super-ion approach, in which each simulated ion represents a large number of actual implanted ions. The model developed by Tian et al.[5] proposed that each simulated ion represent a real ion. The basic idea is to treat the ion implantation as implanting into an ensemble of implant windows. Assuming that the ions are uniformly distributed across the wafer, the number of ions implanted into each window can be calculated for a given dose. In principle, this is the number of ions required to do the simulation. More ions are used only to generate statistically significant profiles. This approach is termed "ion splitting" because of its splitting of the simulation ions into many different implant windows. Due to the fact that each ion cascade generates hundreds or even thousands of point defects, the number of ions required to generate the statistically significant damage profiles is much less than the number requiredto generate the impurity profiles. We can actually utilize this fact to advantage to reduce the CPU time considerably. The basic idea is to divide the simulation ions into 'teal" ions and "shadow" ions. The 'teal" ions generate full cascades, while the "shadow" ions are followed only when they travel through the damaged (by the 'teal" ions) crystal. The additional shadow ions provide the desired statistical significance in the computed impurity profile. The displaced silicon ions are not followed when the "shadow" ion is simulated. This approach allows us to obtain both the impurity profiles and damage profiles with comparable statistical significance at the same time. This technique of reducing the computational time is referred to as "shadowing"
Simulation Results and Discussion The models for indium and germanium produce the impurity profiles and damage profiles as functions of depth. During each simulation, the interstitial/vacancy concentration profile is transformed into an amorphization percentage versus depth profile, in order to be able to compare with the RBS data. For each implant, the impurity profile is compared with the SIMS profile, and the damage profile is compared with the RBS profiles if the damage is significant enough to be detected by RBS measurements. The models have been tested over a wide range of implant conditions. Table 1 shows the range in which the impurity profiles have been validated, and Table 2 shows the range in which the damage profiles have been validated. Species
T a bIe 1. R ange 0 Energy Range (keV)
Indium
5-100
0
0
0
0
2xl0 12 - 2xl0 14
0
0
0
0
lxl0 12 - lxl0 13
Germanium
110-200 through oxide 10-80
a I a Ion or mpurny ro I es Tilt Angle Rotation Angle Dose Range(em")
o -7 0
9
0
o -30 0
0
5xl0 12
-
2xl0 15
Species Indium Germanium
Ta bl e 2. R ange 0 Energy Range (keV) 5-100 10-80
a I a Ion or amage P ro til I es Tilt Angle Rotation Angle 0 0 o -7 o -30 0
0
0
0
0
0
Dose Range(em") 2x10 14 2x10 13 - 2x10 15
1. Germanium Implants Figure 1 illustrates the ability of the new germanium model to successfully predict the impurity profiles as a function of dose. As can be clearly seen, the germanium implant model can very successfully simulate the implant-induced dechanneling effect. As shown in the figures, both the Kinchin-Pease and the KADM damage models provide accurate as-implanted impurity profiles. Compared to the computationally efficient KP model, the KADM model is somewhat more accurate and provides more detailed damage information, as shown in Figure 2. The predicted and experimental tilt/rotation angle dependence of both the impurity and damage profiles are compared in Figure 3 at a dose of 1x1014 cm' and an energy of 80keV. It is can be seen that the predicted impurity profiles and the SIMS profiles are in excellent agreement, and that the predicted damage profiles and the RBS profiles are in very good agreement. It should also be noted that the cross-over of the damage profile at approximately 0.07 urn is correctly predicted. As indicated in the figure, the channeling effects and damage build up are closely related. Since the on-axis implant has more channeling, it not only exhibits a deeper impurity profile, but a deeper and wideidamage profile as well. This is more pronounced for relatively higher energy implants. In addition, since the channeled ion suffers less scattering, the damage level of the on-axis implant is relatively lower than that of the off-axis implant, with all other implant conditions identical. For on-axis, relatively high energy implants, if the dose is appropriate, the formation of a buried amorphous layer can be observed. Because the amorphization transformation is not easy to capture, only the lx10 14cm-2-dose implant exhibits a buried layer for germanium 80keV implants, as shown in the figure. Figure 4 depicts the simulation results for both damage and impurity profiles as functions of energy compared with experimental data. Again, it is can be seen that the impurity profiles and the SIMS profiles are in excellent agreement, and that the predicted damage profiles and the RBS profiles are in very good agreement. As indicated by the damage profiles, the new germanium model can successfully simulate the effect that as the energy increases, the amorphization layer thickness increases accordingly. The predicted impurity and damage profiles are compared with SIMS and RBS data in Figure 5, for 80keV off-axis Ge implants at various doses. It is can be seen that the new germanium model can very effectively simulate the process of damage accumulation as the implant dose increase. The damage profiles also indicate the amorphization process takes place between the doses of 2x1 013cm-2 and 1x10 14cm-2 . Generally, at all energies between 10keV and 80keV, the amorphization threshold is approximately 3 - 5Si energy=80keV dose=1x1014cm-2
1019
Model Prediction SIMS
o o tilt angle/rotation angle=0 /0 o tilt angle/rotation angle=6 /30o
1018
1017
1016 0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
depth (μm) (a) 120
Ge->Si energy=80keV 14 -2 dose=1x10 cm
110
Percentage amorphization
100
RBS Model Prediction
90 80 70 60 o
o
tilt/rotation angle=0/0 tilt/rotation angle=6o/30o
50 40 30 20 10 0 0.00
0.05
0.10
0.15
0.20
0.25
0.30
Depth (μm)
(b) Figure 3. Simulation of the tilt/rotation angle dependence for both Ge impurity and damage profiles at an energy of 80keV and a dose of IxI0 14cm-2 : (a) comparison with SIMS impurity profile and (b) comparison with RBS damage profiles. Note that the flat tails in the experimental RBS profiles are due to RBS background noise, and hence should be ignored. 15
germanium concentration (cm-3)
1021
Ge->Si dose=1x1014cm-2 tilt=6 o rotation=30 o
1020
Model Prediction SIMS
energy=10keV energy=20keV energy=40keV energy=80keV
1019
1018
1017
1016 0.0
0.1
0.2
0.3
depth (μm) (a)
120
Ge->Si
110 100
Percentage amorphization
RBS data Model Prediction
dose=1x1014 cm-2 tilt=6o rotation=30o
90 80
energy=80keV energy=40keV energy=20keV energy=10keV
70 60 50 40 30 20 10 0 0.00
0.05
0.10
0.15
0.20
Depth (μm) (b)
Figure 4. Simulation of the energy dependence for both the impurity and the damage profiles for germanium implants at a dose of Ix10 14cm-2 : (a) comparison with SIMS impurity profiles and (b) comparison with RBS damage profiles. 16
-3
germanium concentration (cm )
1021
Ge->Si Energy=80keV o o tilt=6 rotation=30
1020
dose: 2x1015cm-2 5x1014cm-2 1x1014cm-2 2x1013cm-2 5x1012cm-2
1019
Model Prediction SIMS
1018
1017
1016 0.0
0.1
0.2
0.3
0.4
depth (μm)
(a) 120
Ge->Si
110
energy=80keV tilt=6o rotation=30o
Percentage amorphization
100
RBS data
Model Prediction
90 80 70 dose=2x1015cm-2 dose=5x1014cm-2 dose=1x1014cm-2 dose=2x1013cm-2
60 50 40 30 20 10 0 0.00
0.05
0.10
0.15
0.20
0.25
0.30
Depth (μm)
(b) Figure 5. The dependence of both the impurity and the damage profiles on dose is shown for germanium implants at a energy of 80keV and tilt/rotation angles=6 °/30° : (a) comparison with SIMS impurity profiles and (b) comparison with RBS damage profiles.
17
1020
B->Si
Energy=35keV 14 -2 dose=1x10 cm o o tilt=0 rotation=0
concentration (cm-3)
1019
SIMS Model Prediction 1018
1017
1016
1015 0.0
Pre-amorphization w/o preamorphization 13 -2 with Ge 25kev 2x10 cm 14 with Ge 25kev 5x10 cm-2
0.1
0.2
0.3
0.4
0.5
0.6
depth (μm) Figure 6. Illustration of the ability of the new germanium model to accurately simulate the dependence of boron implant profiles on pre-amorphization doses: the energy of the boron implant is 35keV. The tiltlrotation angles are 0°10°.
18
Indium concentration (cm-3)
1020
In->Si Energy=50keV tilt=0o rotation=0o
1019
SIMS Model Prediction 1018
dose: 2x1014cm-2 2x1012cm-2
1017
1016
1015 0.0
0.1
0.2
0.3
0.4
0.5
0.6
depth (μm)
(a)
Indium Energy=100keV o o tilt=0 rotation=0
Indium concentration (cm 3)
1020
Model Prediction SIMS data
1019 dose=2x1014cm-2 dose=2x1012cm-2
1018
1017
1016
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
depth (μm) (b) Figure 7. Simulation of the dose dependence for the impurity profiles of indium on-axis implants. (a) Energy = 50keV. (b) Energy=lOOkeV.
19
Indium Energy=140keV o o tilt=0 rotation=0 Implant through oxide of 9nm
Indium concentration (cm-3)
1019
1018
Model Prediction SIMS
dose=1x10 13cm-2 12 -2 dose=1x10 cm
1017
1016
1015
0.0
0.1
0.2
0.3
0.4
0.5
depth (μm)
Figure 8. Illustration of the ability of the new indium model to accurately predict indium implants through oxide. The energy of the indium implants is 140keV, and the implants are through 9nm of oxide. The oxide was removed before the SIMS measurements. The depth =0 position corresponds to the Si and Si0 2 interface. UT-MARLOWE simulation of damage changing with energy 120 110
Percentage amorphization
100
In->Si dose=2x1014 cm-2 o tilt=0 o rotation=0
90 80 70
energy=100keV energy=50keV energy=20keV
60 50 40
RBS data Model Prediction
30 20 10 0 0.00
0.05
0.10
0.15
0.20
Depth (μm)
Figure 9. The dependence of damage profiles on energy is shown for indium on-axis implants at a dose of 2xl0 14cm-2 . 20
