SRP/SIMS Junction Depths -- Summary
Chandra Mouli -- -- May 1, 1996
Spreading Resistance Profiling (SRP) and Secondary Ion Mass Spectrometry (SIMS are the two most common means of measuring doping profiles. Both have limitations on accuracy and it is important to consider them carefully while interpreting the results -- particularly while calibrating a simulation tool.
Below is a summary of the email discussions I had with several people in the Silvaco user group community regarding SRP and SIMS:
Chandra Mouli/Micron wrote on Thu Jun 22 11:32:50 1995
Why is the carrier junction (n=p) measured by SRP always shallower than the metallurgical junction (Na=Nd) measured by SIMS? For a n+/p type junction shouldn't the carrier junction be deeper than the metallurgical junction?
Trudo Clarysse/IMEC replied:
R&D Spreading Resistance
Hello, If you are interested in SRP analysis then it is indeed important to know the answers on the questions you pose. I will try to clarify the situation for you. For more details I refer you to some of the SRP papers which have been published in JAP, APL and JVST (see references at the end).
When discussing SRP carrier profiles one has always to discriminate clearly between the "on bevel" and "internal (vertical)" carrier profile: The "on bevel" carrier profile is the one which you get from a typical SRP measurement using bevel angles of less than a few degrees. The "internal" profile is the one which you "would" get from a hypothetical electrical SRP-like measurement along the cross-section of a sample (equivalent with considering a bevel angle of 90 degrees). Due to the physical size of the SRP probes (several micron radius) the latter is not physically possible. Note that it is the internal (and not the on bevel) carrier profile which determines the performance of the device in practice.
Now for your questions:
1) In equilibrium, the metallurgical junction can in principle be the same as the "internal" carrier junction. For instance, if you consider the case of a (sufficiently thick) constantly doped p-type epilayer 1e15/cm3 with an abrupt transition on an equally doped n-type substrate, then both junctions will coincide. This is due to the fact that the amount of carrier diffusion in both directions (p->n and n->p) is exactly the same. In almost all other cases, however, the metallurgical and internal carrier junctions will be different. For example, for a highly doped p-type epilayer of 1e19/cm3 with an abrupt transition (or for a source/drain implant) on a typical n-type substrate of 1e15/cm3 the electrical junction (n=p) of the "internal" carrier profile will always be DEEPER than the metallurgical junction. This is due to the fact that the p-type carriers diffusing from the highly doped epilayer into the substrate dominate the carrier impurity type up to a certain depth beyond the metallurgical junction. Note that the peak electric field in this example is indeed located at the metallurgical junction, but the internal carrier junction is DIFFERENT from the metallurgical junction. This is the kind of result which you will get from any one-dimensional Poisson solution. Note that in the depletion approximation the amount of carriers in the depleted region is considered to be neglectable relative to the charges due to the ionized dopants. However, this does not imply that there are no carriers present in the depletion region, neither does it imply that the internal carrier junction should be at the same depth as the metallurgical junction.
2) As indicated above for a n+.p structure the "internal" carrier junction will be deeper than the metallurgical one as you expect. However, in SRP one measures "an other kind" of electrical junction, namely the "on bevel" junction position which can be quite different from the "internal" electrical junction (which you get from a straightforward Poisson calculation). It is well known that the "on bevel" electrical junction is ALWAYS shallower than the metallurgical junction, opposite to what one might expect at first sight. The fundamental reason for this apparent strange behaviour of the on bevel carrier junction is the bevelling process which is neccessary to access the in depth information with SRP. Bevelling means removing material, i.e. removing dopant ions and their carriers. Therefore as one moves over the bevel away from the bevel edge, the thickness of the n+ layer (in the case of a n+.p structure) diminishes and so does the amount of n-type carriers available for diffusion into the p-type substrate. Therefore the initial internal (=vertical) carrier junction (when the SRP tips are at the bevel edge and which is not directly measurable) which is deeper than the metallurgical junction, becomes less deep as one moves the SRP tips away from the bevel edge up to a point where there are too little carriers left in the remaining n+ top layer to dominate over the substrate carrier level and actually the reverse occurs, carrier diffusion from the p-type substrate into the remaining top n+ layer starts to dominate, i.e. the remaining n+ layer (metallurgically) becomes p-type carrier. The transition point (depth) where reverse carrier spilling (diffusion) starts to dominate is the apparent "on bevel" carrier junction position. The depth of this transition point is always before the metallurgical junction depth, i.e. closer to the surface. This process can be simulated by performing a series of one-dimensional Poisson calculations at different SRP tip positions (depths) with corresponding different dopant profiles.
To summarize, it is the removal of bevel material which causes SRP junctions always to be shallower than SIMS junctions. Next comes the question how much difference can there be? On a well structure with a peak level below 1e17/cm3 going several microns deep, one can have junction shifts (difference SRP vs SIMS) of 2 microns! Simulations indicate that only 25 percent of this observed shift can be related directly to material removal. So there is an additional physical mechanism which causes on bevel carrier junctions to be 75 percent shallower than expected just from material removal arguments. I call this extra shift the stress-induced carrier spilling. A simple calculation (but also more detailed analysis) indicates that SRP is working closely to a pressure of 10 GPa. At such high pressures changes in the dielectric constant (factor 100) and bandgap narrowing (50 percent) become feasible. Such drastic changes in the silicon material properties can cause additional carrier junction shifts. The details on how this stress-enhanced carrier spilling exactly works is still a point of discussion between SRP specialists. Also surface states which can be related to surface roughness play a role in the latter discussion.
As all SRP measurements will give you a shallower "on bevel" carrier junction than metallurgical junction, one would expect conventional SRP software such as SSM version 2.30 software to correct for this. However, this is not the case. This is basically due to the fact that it is not so easy to correct for carrier spilling in a fast and stable way (numerical problem) and there is not yet agreement on the correct stress model which should be used and which determines the boundary conditions (accuracy problem). At IMEC we have developed over the last few years an experimental package (which can be purchased as IMECPROF version PS.7.1) which runs under OS/2 v2.x or higher (on IBM compatible systems) and can correct single junction isolated SRP profiles automatically with different contact (stress) models for carrier spilling, i.e. the program takes as input the measured resistances and generates the electrically active DOPANT (not carrier) profile. However it should be noted that the application of this Poisson based package is more tedious than the routinely used Laplace software. Also Solid State Measurements (SSM) has a package which can help you. However, this package only performs simulations, i.e. you give your dopant profile as input through the selection of predefined functions, and it calculates the expected spreading resistance profile taking into account only the material removal component of the carrier spilling phenomenon (stress effects are neglected in this package named SRP2).
Basics about carrier spilling:
S.M. Hu, J.Appl.Phys. 53, 1499 (1982)
A. Casel and H. Jorke, Appl. Phys. Lett. 50, 989 (1987)
SRP vs SIMS:
T. Clarysse, W. Vandervorst and A. Casel,
Appl. Phys. Lett. 57, 2856 (1990)
Carrier spilling modelling:
R.G. Mazur, J. Vac.Sci.Technol. B 10, 397 (1992)
T. Clarysse and W.Vandervorst, J. Vac.Sci.Technol. B 12, 290 (1994)
R&D Spreading Resistance
> I would first suspect that much of the n+ is not activated.
> Would that be consistent with your measurements?
> Is it Arsenic? At high concentrations, a large percentage of
> it can deactivate during low T diffusions.
Thanks for your reply -- no, I dont think this is related to activation because the n+ concentration is not that high. The reason I mentioned that the junction is n+/p is to make sure that it is clear that 'n' side of the junction is more heavily doped than the 'p' side. It is (phos+arsenic) and boron.
> Have you considered the carrier spilling effect? This always causes
> junctions to look shallower than they really are, but is particularly
> noticeable for deep junctions like wells.
Yes, I did -- in fact, that's what came to my mind initially, but I am having a hard time convincing myself that this effect alone could lead to such a huge discrepancy in the measured junction. Maybe it is true -- in such case, I would like to know from the experts if there are analytical corrections that can be done to account for the carrier spilling?
If I give you a doping profile and ask for an SRP type of plot, you can provide that. But the inverse problem is not so easy. I think that is why they don't (or why they didn't, if they do now) provide corrections. TCAD simulators can help here: If you simulate your process and follow that with a simulation of an SRP measurement, you can compare that result to SRP plots. If they don't agree, you would need to adjust the process simulation or improve the simulation models or investigate the measurement conditions and iterate.
The SRP junction should be shallower. The best way to understand this is through simulation!
Both TMA and Silvaco have macros in SUPREM3 for looking at spreading resistance profiles. Basically, they etch off some silicon, do a Poisson solution, extract the surface carrier concentration, and call that the SRP profile. When you etch off part of your surface dopant (as you effectively do in SRP beveling), you substantially change the carrier profiles. Imagine the extreme case where you etch off most of the surface dopant. Well, when you look at the carrier concentrations, you will see that the junction depletion layer has spread completely to the surface, i.e., the surface carrier conc is lower than the chemical conc. SRP vendors usually claim to have software to compensate for this, but it is actually a tough problem to work backwards (calculating the chemical conc from the carrier conc) and I wouldn't trust it that much.
Also see Intro to Calibration Figure 1.
This page last updated November 10, 1997 by