Measurement of carrier lifetime: Monitoring epitaxy quality
06/01/2000
Philippe Renaud, Motorola Semiconductors, Toulouse, France
Ariel Walker, National Engineering School, Marseille, France
overview
The corona oxide semiconductor method can be used to determine and quantify different processes of generation-recombination in the epitaxy layers on silicon, also accounting for the effect of the substrate. Experimentation has optimized the various measurements for practical application. The end result is a new, in-line monitoring technique for epitaxy quality via lifetime measurement. This method enables the detection of contamination and defects that are not detected by conventional measurements such as resistivity and delineation etch.
 Figure 1. Contribution of the different recombination processes to recombination lifetime as a function of doping level (inspired by [3]) |
A wide range of characterization tools, including transmission x-ray fluorescence (TXRF), secondary ion mass spectroscopy (SIMS), scanning electron microscopy (SEM), and delineation etch, are used for semiconductor materials quality control. By contrast, semiconductor properties affected by defects or impurities are not commonly used for in-line monitoring in IC manufacturing. However, tools that enable fast and reliable lifetime measurement [1] can be used advantageously to detect impurity concentrations in the range of 1010 atoms/cm3. This method — measuring carrier recombination — is particularly suitable for monitoring the quality of epitaxial layers.
Recombination processes
The parameter "lifetime" is related to the time of recovery when carriers have been put in excess or paucity from equilibrium in a doped semiconductor. If excess carriers are generated, recombination lifetime is measured, whereas generation lifetime is measured in the case of paucity of carriers (in a reverse-biased junction). Recombination processes are composed of three mechanisms:
- the Shockley Read Hall (SRH) process with a lifetime inversely proportional to impurities concentration;
- the radiative process with a lifetime inversely proportional to doping; and
- the Auger mechanism, whose lifetime is inversely proportional to doping squared.
The effective recombination lifetime (tR) is given by:
 |
null
Figure 1 shows that for the usual epitaxial doping level of 4 x 1015 atoms/cm3, recombination lifetime in silicon totally reflects the SRH lifetime. Consequently, it is directly related to impurity concentrations.
For generation lifetime (tG), only generation by impurities is usually considered. Generation lifetime is not equal to recombination lifetime because generation is a thermally activated process.
 |
null
Corona oxide semiconductor
Corona oxide semiconductor (COS) measurement is a nondestructive, contactless, lifetime technique. Its principle is to form a junction in silicon by the deposition of two opposite sign charges on an oxide layer. These charges are deposited by a high voltage source that ionizes the air. The silicon is therefore maintained at equilibrium in an inversion mode. By applying a pulsed charge onto the silicon, or optical excitation, it is possible to measure generation and recombination lifetime. A Kelvin probe monitors the transient decay of surface voltage.
Generation lifetime is measured when the silicon is pushed from equilibrium into a deep depletion mode by a short pulse of a negative charge (dQ). tG is given by [2]:
 |
null
where Winv is the depletion depth in inversion equilibrium. Here dW = Wmax - Winv = dQ/q x doping. dQ can be set for optimum results. For instance, dQ can be determined to keep Wmax <Wepi. For a standard epitaxial process (i.e., 4 x 1015 atoms/cm3), depth of measurement ranges from 0.8-4mm. With generation lifetime, depth of measurement is under user control.
With recombination lifetime, excess carriers are photogenerated in a high injection mode by a xenon flash. The junction is forward biased and the transient decay of Vsurf(t) gives recombination lifetime (tRecombination) [3].
 |
null
For recombination lifetime, the measurement depth is linked to light absorption and photogenerated carriers diffusion length that could be >300mm. Thus, recombination will mostly occur within the substrate.
The conclusion here is that generation lifetime can be used to monitor epitaxial quality and recombination lifetime can detect substrate contamination.
Developing the methodology
In our work, we studied 10mm of n-doped epitaxy on an n-substrate, matching our current process setup. We set out to optimize the generation lifetime measurement by first determining the optimal deposited charge (i.e., generation lifetime). With COS, a user can choose the amount of charge (dQ) deposited on the silicon; dQ ranges from 10-100% of Qtotal = -1.99 x 10-8 C/cm2. dQ may lead to oxide layer breakdown.
 Figure 2. Transient decay of Vsurf(t) after pulse for generation lifetime measurement. |
Looking at measurement reliability, we found that tG is determined from Vsurf(t) in deep depletion mode shortly after the measurement pulse (Fig. 2). The slope used for the calculation (Eqn. 1) tends to flatten with high lifetime (i.e., for >8000 msec, dVsurf(t)/dt <5mV/sec, which is the equipment's limit of detection).
Eqn. 1 shows that for a given lifetime, the slope of Vsurf(t) can be increased with deposited charge. Then the best sensitivity will be obtained with dQ slightly below the oxide breakdown value.
 Figure 3. Lifetime repeatability. |
Figure 3 shows the repeatability level for our epitaxy. For lifetimes >6000 msec, there is no repeatability. This corresponds to a slope of 20mV/sec for Vsurf(t). Figure 4 represents the detection limit for various doping levels.
 Figure 4. Limit of generation lifetime measurement reliability for various doping levels. |
We also considered the choice of oxide and the effects of interface states. To determine if a rapid oxidation process is sufficient for our monitoring, we compared two oxide qualities: 1. a simple rapid thermal oxide (RTO), and 2. a more precise gate oxide (GOX). We clearly determined that RTO gives lower lifetime values than GOX. To ensure that the differences were due to the oxide only and not to epitaxy quality, we did a process short loop experiment to demonstrate the influence of oxide-epitaxy interface on lifetime measurement. Sequentially processing the same wafer through RTO followed by measurement, then oxide strip, and then through GOX, we found that the effective lifetime increases with a GOX (i.e., 1100 msec after RTO and 4300 msec after GOX). Also, the reverse operation (i.e., GOX, strip, RTO) has been done and the effective lifetime decreased with RTO. Interface defects or impurities causing higher generation rate can explain this. The effective generation lifetime can be expressed as:
 |
null
where tBULK is the bulk lifetime in silicon and tS is the surface lifetime that depends on the density of traps of the epitaxy-SiO2 interface (Dit). As the same wafer has the two oxidization processes, tBULK is constant for the two lifetime measurements. Thus, the differences in effective lifetime must come from tS. Dits were measured and found to be 6 x 1010 cm-2/eV for GOX and 4 x 1011 cm-2/eV for RTO.
 Figure 5. Process results. |
We also looked at substrate choice. For cost savings, reclaim wafers are often used as test or dummy wafers, so we compared the two. We obtained generation lifetimes on reclaim and new substrates processed through four different epitaxial reactors. We did not see a good correlation in lifetime between new and reclaim substrates. For example, in one reactor, generation lifetime dropped with the reclaim substrate. However, in the same reactor, the other wafer gave high lifetimes. The contamination is due to initial contamination in the reclaim substrate diffusing into the epitaxial layer during the process. Because the use of reclaim substrates can lead to wrong interpretation, we recommend using new substrates.
Possible scenarios
Figure 5 presents a summary of our most significant results:
- Figure 5a "good epi, good substrate": Generation lifetime, which relates to epitaxy quality (3mm measurement depth), shows a high value of more than 2000 msec. The bulk recombination lifetime (more than 300 msec) is also high. These results show no contamination in the epitaxy or substrate.
- Figure 5b "good epi, bad substrate": This is a case where epitaxy generation lifetime is high with a recombination lifetime lower than 50 msec in the substrate. Here, the contamination comes from the substrate, but does not diffuse to the epitaxy layer. This case shows the possibility of detecting problems on substrate quality.
- Figure 5c "bad epi, good substrate": Generation lifetime is low in the epitaxy (<500 msec), while the substrate shows a recombination lifetime as high as 400 msec. This shows an interesting case where the contamination is due to the epitaxy process itself. The contamination was caused by a very small air leak in the reactor, which could not have been detected by resistivity measurement and surface inspection.
- Figure 5d "bad epi, bad substrate": This presents a case where both generation and recombination lifetimes are low. It is difficult to determine the source of contamination. It can be caused by contamination in the reactor chamber from reactive gases that contaminated the epitaxy layer and diffused into the substrate. Or, the contamination could come from the substrate itself diffusing in the other direction. The pollution here was due to dirty process chamber quartz.
Conclusion
Our work details a new method to monitor the quality of epitaxial layers. By using COS measurement, we can determine epitaxial layer quality using generation lifetime and recombination lifetime to determine substrate quality. The impact of oxide quality on lifetime through interface impurities or defects (RTO-GOX) has been demonstrated. We also showed a procedure to optimize the measurement by recipe adjustment and taking equipment limitations into account.
Acknowledgments
We thank Z. Chalupa and K. Marble for helpful discussions.
References
- KLA-Tencor, Quantox oxide charge-monitoring system.
- D.K. Schroder, "Corona-Oxide-Semiconductor Device Characterization," Solid-State Electronics, Vol. 42, No. 4, pp. 505-512, 1998.
- S.K. Pang, A. Rohatgi, "Record High Recombination Lifetime in Oxidized Magnetic Czochralski Silicon," Appl. Phys. Lett., 59, pp. 195-197, July 1991.
Philippe Renaud received his PhD in physics from the University of Nice, France, and has worked in the semiconductor industry for more than 10 years. He specializes in epitaxy and wet etch processes at Motorola, Le mirail, B.P. 1029, 31023 Toulouse cedex-France; e-mail [email protected].
Ariel Walker is currently in his last year of microelectronics study at the National Engineering School of Marseille.