Measuring Semiconductor Wafer Dopant
Density with the LEXS
David OHara
Parallax Research, Inc.
As semiconductor features get smaller in horizontal dimension, they must also get thinner in vertical dimension so that many features of such circuits are thinner than 50 Angstroms. For example, so-called diffusion barrier layers of TaN or TiN may be as thin as 50 A and semiconductor dopants may be deposited in very shallow layers or in ultra-thin layered structures. Dopants are the materials that are used to determine whether a silicon wafer will be either p type or n type and often consist of ultra-low concentrations of B, P, or As in an otherwise pure silicon (or Ge) crystal. For quality control purposes, the spatial density and density depth profile of the dopants needs to be determined. Previously, when the dopants were spread throughout the bulk of the semiconductor wafer, there were techniques that used electrical resistivity to determine the bulk dopant density but as the dopant depths have gotten very small, these methods are of limited utility. Currently, the best way to measure the dopant depth profile is a semi-destructive method called Secondary Ion Mass Spectroscopy which is both expensive and slow and cannot be used over the entire surface of a wafer without destroying it. However, we have found that the LEXS spectrometer can be used to measure the dopant density and produce dopant profiles in a non-destructive way.
To investigate the potential of using the LEXS spectrometer for measuring boron dopants in wafers, Parallax obtained the NIST boron dopant standard NIST SRM 2137 and used low beam voltages to look at the boron signal. This standard has boron implanted via ion beam with a 50 KV beam and the peak concentration depth is about .2 microns. This deep doping is a worst case scenario for LEXS as we specifically intend to use LEXS for very shallow dopants, however, it does allow us some basis for comparison. “Sensitivity” is defined as being
(B peak height-background)/sqrt(background).
Fig. 1. Sensitivity for boron in the NIST standard as a function of beam voltage.
Fig. 2. Boron peak for 1 KV electron beam.
Obviously, a lower voltage might be even better but our old JEOL 35 could go no lower than 1 KV.
To further investigate the use of LEXS for shallow dopant measurements Parallax used 5 wafers that had been shallow ion implanted in a 1 KV BF3 plasma with boron with doses that varied by nearly 10X. Energy scans were done on the various samples from 100 eV up to 250 eV covering the boron line at 182 eV to see how well the B peak stood out against the background. In all cases, even at the lowest B dose of 1E15 the B peak stood out very well. We found that the best combination of peak to background and count rate in terms of counts/sec/nano-amp (counts/sec for a given electron beam current) were obtained at 2 KV electron beam energy. There may be a slightly better energy but our SEM could only be varied from 1, 2, 5, and 10 KV. In all measurements we used ONLY 2 nano-amps because our old SEM ( a JEOL T300 in this case) could only produce this very low current at this low energy. Furthermore, it was difficult to keep the beam current stable and we have reservations about the voltage stability at this low voltage. Data is shown in Fig. 3. taken in 10 second count times. Uncertainty ranges from 2.5% to 1.7%.
Fig. 3. Data taken using LEXS spectrometer at 2KV beam voltage and 2 nano-amp beam current. Count times were 10 seconds. Curvature may be due to bad voltage calibration in this very old SEM.
While this simple experiment shows the feasibility of using the LEXS to detect the dopant, what we really want is a depth profile and for this we propose to use the variable electron beam energy. The depth of penetration of the electrons is a very strong function of the electron energy with most of the electrons stopping at a very specific depth. Fig. 4.shows the depth of penetration of various energy electron beams in Si showing that they tend to stop at a well defined depth. If the theoretical shape of the depth profile is known, we only need measurements done at two different beam energies to get the profile while if it is completely unknown we might have to sweep the electron
Fig. 4. Simulation of 500 eV electrons incident onto Si with 1 KeV ion implanted Boron with dose of 8E15/cm3. Blue shows electrons coming to rest in the Si with red showing backscattered electrons. Heavy black lines show calculated dopant density above line.
Fig. 5. Simulation of 1000 eV electrons incident onto Si with 1 KeV ion implanted Boron with dose of 8E15/cm3.
Fig. 6. Simulation of 1000 eV electrons incident onto Si with 1 KeV ion implanted Boron with dose of 8E15/cm3.
beam in energy while measuring Boron x-rays to determine a very complicated distribution. However, in every case, the depth profile could be completely determined within seconds due to the very good response of the LEXS to low energy x-rays.
Figs 4-6 show simulations of electrons incident onto Si which as been ion implanted with Boron with ion energy of 1 KeV and density of 8E15/cm3. The blue tracks represents the trajectories of electrons scattering and coming to rest in the Si while the red tracks show electrons that eventually leave the Si. Heavy black horizontal lines show calculated dopant density for orders of magnitude. It is immediately obvious that electron penetration depth is a strong function of electron energy and that electrons with energies above 2 KeV mostly pass through the thin dopant layer. Lower energy electrons mostly stop in the dopant layer causing emission of Boron x-rays. Conventional electron microanalysis with higher energy electrons is essentially useless for measuring shallow dopant profiles while very low energy electron energies can be used to do depth profiling. The electron beam voltage that is optimal may be somewhat higher than that which produces optimal stopping within the dopant layer because the production of x-rays increases with overvoltage, that is the difference between the energy necessary to produce x-ray excitation and the electron energy. Thus, in the example we show in Fig. 4, the optimal electron beam voltage was nearly 2 KV while you would expect it to be only about 1 KV from electron stopping calculations. However, at a higher voltage of 5KV, the Boron signal from the thin layer was very weak because almost all of the electrons simply passed through the layer.
PO Box 12212