4 Measurement Methods

4.1 Introduction

Infrared, Raman, and photoluminescence spectroscopy share common elements. Each determines an optical intensity versus frequency or wavelength. Each needs a light source, a means to set or measure wavelength, a detector, and a device to record the spectrum. But the different wavelength ranges over which they operate, and the different intensity levels, dictate that the details are very different. This chapter describes the actual measuring arrangements. I first present infrared Fourier trans­form infrared ( F T - I R ) spectroscopy, because the Fourier method is also used in photoluminescence ( P L ) and Raman scattering. Photoluminescence and Raman tech­niques, which have similarities, are presented next, with the simpler P L method first. In the final section of the chapter, I discuss semiconductor sample geometries, surface conditions, and cooling procedures for optical analysis.

The chapter is constructed around what is commercially available, and suitable for semiconductor characterization. It includes a listing of suppliers of instruments for F T - I R , Raman, and P L spectroscopy. The listing is not meant to be complete, but only to provide a starting point. The discussion of each technique in this chapter should give the reader enough information to begin choosing among commercial products to meet his or her characterization needs. The listing of manufacturers includes addresses where further information can be obtained.

4.2 Fourier transform infrared spectroscopy

4.2.1 Theory and methods

A standard technique for visible light spectroscopy is to use a black-body source with a dispersive grating spectrometer as the wavelength selector. Infrared gratings are cheap and easy to make, especially at the longer wavelengths, so grating spectroscopy should be an obvious choice for the infrared as well. But the infrared power available from a black body is small, especially at the very long wavelengths useful to character­ize impurities, lattice modes, and free carriers. Grating spectroscopy does not make the most effective use of the limited infrared power, although it gives good results when used with care. Instead, the dominant method is F T - I R spectroscopy, whose more efficient use of the available power makes it preeminent.

Fourier spectroscopy also uses a black-body source, but replaces the dispersive grating spectrometer with a Michelson interferometer. This gives the Fourier trans­form of the desired spectrum, known as the 'interferogram'. The interferogram requires extensive computer manipulation to yield the desired intensity versus wave-number spectrum, but the Fourier method gives a much higher signal-to-noise ratio

46 Measurement methods

than does grating spectroscopy. One reason is the Fellgett advantage, which states that (under certain conditions) the signal-to-noise ratio is higher when many wave­lengths are measured simultaneously, as in Fourier spectroscopy, than when one wavelength is measured at a time, as in dispersive spectroscopy. The second reason is the Jacquinot or throughput advantage, which means that high resolution can be attained in F T - I R spectroscopy without using narrow slits, whereas in conventional spectroscopy there is always a trade-off between resolution and the amount of light reaching the sample.

The basic process in F T - I R spectroscopy is light wave interference. In the Michel-son interferometer shown in Fig. 4.1, radiation leaves the source and reaches the beam splitter ( B S ) . Part of the beam passes through the BS and is reflected from a fixed mirror M l , whereas another part is reflected from the BS and then from mirror M 2 , which is movable. A s M 2 moves, it changes the difference in length Δ between the paths the two light beams traverse. When the beams recombine they produce an intensity which depends on Δ, called the 'interferogram' / (Δ). Except for a constant multiplicative factor, the interferogram is given by:

1 / ( A ) = J S ( / ) [ l + c o s ( 2 ' n / A ) ] d / = ^ / ( 0 ) + | 5 ( / ) c o s ( 2 * / A ) d / (4.1)

where S(f) is the intensity spectrum of the source, / is the optical frequency in wavenumbers, and 1(0) is the intensity at zero path difference. The integral in equation (4.1) is the cosine Fourier transform of 5 ( / ) , which can be recovered by carrying out the inverse Fourier transform (also to within a constant factor):

oo 1

S(f) =_Jm [/(Δ) - \ 7(0)]cos(2-n/A)dA ( 4 2 )

This is implemented by measuring the interference signal / (Δ) with a detector, whose

Black-body

source

Computer

Detector

Fig. 4 .1 Michelson interferometer for transmissive FT-IR spectroscopy. The black body illumi­nates the beam splitter (BS). The two resulting beams recombine in the interferogram signal / ( Δ ) , where Δ is the difference in path length, which continues through the sample and to a detector. The computer performs a Fourier transform to recover intensity versus wavenumber. Reflectance can also be measured with a different geometry.

Fourier transform infrared spectroscopy 47

output is then processed by a computer. A l l commercial Fourier systems include a computer with software to carry out the Fourier integral.

T o measure sample transmission rather than the source spectrum, the sample is placed as shown in Fig. 4.1. Then equation (4.2) yields T(f)S(f) where T(f) is the wavenumber-dependent transmission coefficient. If T(f)S(f) is measured with the sample in place, and the source spectrum S(f) is measured with the sample removed, the ratio of the two quantities gives T(f) alone. A similar method, with radiation reflected from the sample instead of passing through it, yields the reflection coefficient / ? ( / ) , except that the reflected intensity is measured relative to a 100% reflector, which can be a piece of polished metal.

The application of this theory has complications in practice. One is the computer evaluation of equation (4 .2) , which can be time-consuming even on a fast machine. But the fast Fourier transform ( F F T ) algorithm allows real-time Fourier spectroscopy, and can be implemented on microcomputers (DuVarney and Perkowitz, 1981). A second factor is that the integral in equation (4.2) is not known over all values of Δ from - o o to oo. It is measured only over the range of mirror movement, say —L to L . This limitation in the evaluation of the Fourier integral turns out to determine the wavenumber resolution of a Fourier spectrometer, which (in c m - 1 ) is Δ / = IIL. Bell (1972) gives a complete discussion of this, and other aspects of Fourier spectroscopy. But for most users, it is sufficient to know that commercial F T - I R instruments come with prewritten software ready to deal with these complications. In actual F T - I R instruments, Δ/ranges from a few wavenumbers, adequate for far infrared reflectance characterization of free carriers and phonons, to < 0 . 0 1 c m _ 1 , necessary for some shallow impurity work.

In principle, Fourier spectroscopy can operate in any part of the spectrum. In practice, different sources, even broadly emitting black bodies, tend to favor certain spectral regions. The same is true for beam-splitter materials, and for most detectors. The mechanical tolerances of the Fourier spectrometer itself set the shortest wave­lengths where interference can be seen. Judicious choice of these elements determines the spectral range. Most commercial systems are designed so that different sources, beam splitters, and detectors can be introduced, to cover the range 10 to 4 x 1 0 4 c m _ 1 , the ultraviolet to the extreme far infrared.

Although there is a variety of commercial F T - I R instruments available, some users may need to add or fabricate additional optical elements. Appropriate materials and optical components are well known in the visible to the mid-infrared, but the less common far infrared components need discussion. Beam splitters are made from mylar, quartz, and other materials, depending on the wavelength range to be covered. A newer development, the wire grid polarizing beam splitter, is superior at very low wavenumbers. Far infrared lenses can be formed from polyethylene, with a nearly constant refractive index of 1.5, but mirror optics are often preferable. Ordinary metals, when polished, have far infrared reflectances of > 9 5 % , and this value is insensitive to wavelength. It is simple to machine high reflectivity mirrors, even complex paraboloidal designs, from aluminum or other metals.

Black polyethylene, which contains carbon, effectively blocks visible and ultraviolet light while passing far infrared radiation. Far infrared filters can also be made from wire grids, which form efficient polarizers as well. Reasonably transmissive windows for sample Dewars can be made from polyethylene, but may unreliably change shape when they separate vacuum from atmosphere. Crystalline quartz and sapphire are

48 Measurement methods

superior for such applications, offering high transmissivity at low temperatures. One component peculiar to the extreme far infrared is the light pipe, an internally polished metal pipe acting as an oversized waveguide. A typical diameter is 1cm, ten times larger than the longest wavelength of 1 mm. Such pipes can terminate in a semifocus-ing element, a gently tapering internally polished cone.

Even with the signal-to-noise advantages of F T - I R spectroscopy, sensitive detectors are important. In the mid-infrared, mercury cadmium telluride photovoltaic or photo-conductive detectors give good sensitivity when operated at liquid nitrogen tempera­tures. Devices cooled to liquid helium temperatures are the most sensitive. A n example is the doped germanium bolometer which gives a minimum detectable or noise-equivalent-power ( N E P ) of about 10~ 1 2 W H z ~ ° 5 . Such detectors are sophisti­cated devices, which are expensive to operate because they require liquid helium. Fortunately, some far infrared work can be carried out with pyroelectric detectors, which are based on a thermal principle. They operate at room temperature and are far less sensitive than the helium-cooled devices. The sensitivity, however, is often adequate, especially for reflectance work, where the signal tends to be larger than in transmission. The simplicity and low operating cost of the pyroelectric detector is to be preferred for far infrared characterization work whenever its sensitivity is adequate.

T w o other requirements must be met in an F T - I R unit which will be used in the mid- to far-infrared. Water vapor absorbs heavily in this spectral region. Hence the interferometer must be maintained at ^0.1 Τ pressure, although flowing dry nitrogen is sometimes used. Second, it is often necessary to cool the sample to cryogenic temperatures. I discuss methods for this in the section on sample techniques.

There is a variant of the F T - I R spectrometer, the F T - I R microscope, which can be useful for semiconductor applications. This combines an optical microscope with an F T - I R unit to allow examination of small samples, or micropositioning on a larger sample. Typical units allow either reflectance or transmittance measurements, and magnifications of several hundred.

Despite the dominance of F T - I R spectroscopy using black bodies, other sources expand possibilities for characterization. Where a limited range of wavelengths will do, a C 0 2 laser emits watts of power between 9 and 12 μπι. Solid state lasers operate in the near infrared. The optically pumped infrared laser can also be useful. Here a C 0 2 laser pumps a gaseous medium, such as C H 3 O H , in a cavity. The proper pump line excites the medium's molecular vibrational levels. This causes population inver­sion and hence lasing among rotational levels, which occurs in the mid- to far-infrared. A t a pump power of 20-30 W , output powers are typically 0 .1-10mW per line. This enormously exceeds the power from a black body and permits the measurement of very small transmitted intensities. The hundreds of such step-tunable infrared lines give sufficient spectral coverage to characterize semiconductors, as shown by Bean and Perkowitz (1977, 1979). However , no such sources are commercially available at this time, to my knowledge. T w o other sources, the free electron laser and the synchrotron, provide kilowatts of infrared power in microsecond to picosecond pulses. Several of these are in operation, but are not yet at the stage of routine use for characterization. The solid state terahertz source is an innovation now in the early stages of development. I comment further on these new sources in Chapter 8.

The classic books on Fourier transform spectroscopy, which cover theory and instrumentation, include those by Bell (1972) and Chamberlain (1979). Kimmitt

Fourier transform infrared spectroscopy 49

(1970) gives details of instrumentation, components, and materials for the range 10-2 5 0 c m - 1 . Chantry (1971), and Môller and Rothschild (1971), also focus on the mid-to far-infrared.

4.2.2 Commercial FT-IR spectrometers and microscopes

4.2.2.1 Bomem, Inc., 450 St. Jean-Baptiste, Quebec, Canada G2E 5S5

The D A 8 F T - I R spectrometer can operate over 10 to 5.5 x 1 0 4 c m - 1 , if appropriate sources, beam splitters, and detectors are selected. Spectral resolution varies with spectral range, with values to 0.003 c m - 1 available. The sources for the near-, mid-, and far-infrared, respectively, include quartz halogen lamps, globars, and mercury vapor arc lamps. Three different sources can be installed and selected by computer. Beam splitters for different portions of the infrared are made from glass, BaF 2 , KBr , mylar, or wire-grid assemblies. Available detectors for the near- and mid-infrared include silicon and germanium photodiodes, and cooled InSb and H g C d T e as­semblies; for the far infrared, cooled silicon bolometers, and room-temperature pyro-electric triglycine sulfate ( T G S ) detectors. The Fourier transform and other data analysis are performed by a V A X 3100 workstation with appropriate software. The unit includes a built-in vector processor for the fast Fourier transform, and an Ether­net interface. Optical accessories include reflectance attachments.

4.2.2.2 Bruker Instruments, Inc., Manning Park, Billerica, MA 01821, USA

The IFS48 covers 400-4800 c m - 1 at 2 c m - 1 resolution. The range can optionally be increased to 220-440 c m - 1 , and the resolution to 0.5 c m - 1 . The IFS66 also has a standard coverage of 400-4800 c m - 1 , but at 0.24 c m - 1 resolution. This can be extended to 0 . 1 c m - 1 , and the wavelength range to 20 to 3 x 1 0 4 c m - 1 . The IFS88 offers the same standard coverage and resolution, but can be extended to cover the range 20 to 4 x 1 0 4 c m - 1 . The IFS66 and IFS88 also offer different measurement configurations, and both operate as purged rather than vacuum systems. The IFS120HR operates under vacuum and offers the highest resolution, 0.008 c m - 1 stan­dard to 0.002 c m - 1 optional. The standard wavenumber coverage is 450-4800 c m - 1 , which can be extended over 10 to 4 x 1 0 4 c m - 1 , allowing work into the deep far infrared. A l l instruments include data acquisition and analysis systems based on micro- or mini-computers. Sources can be chosen from globar, mercury lamp, tung­sten lamp, and xenon arc lamp units. Ge /KBr , Mylar, and quartz beam splitters are available. Detectors include H g C d T e and InSb units operating at 77 K , silicon diodes, and a liquid-helium-cooled bolometer.

Bruker also supplies an F T - I R microscope which operates in transmission or reflec­tion, with a mercury cadmium telluride detector for the range 600-4800 c m - 1 . Cover­age can be extended to include the near-infrared region 1800-10 4 cm - 1 . The standard magnification factor is 300, which can be optionally increased to 1480. A manual stage for sample translation and rotation is standard, and a computer-controlled stage can be added.

50 Measurement methods

4.2.2.3 Bio-Rad, Digilab Division, 237 Putnam Ave., Cambridge, MA 02139, USA

The FTS60A F T - I R spectrometer can operate over the near-, mid-, and far-infrared. The standard configuration uses a water-cooled ceramic element source, a KBr beam splitter, and a pyroelectric detector to cover the mid-infrared range (400-4400 c m - 1 ) at a resolution of 0.25 c m - 1 , which can be extended to 0.1 c m - 1 . The coverage is extended to the near infrared (4400 to 1.5 x 1 0 4 c m _ 1 ) by the addition of a tungsten halogen lamp, CaF 2 or quartz beam splitters, and PbSe, InSb, or silicon detectors. Far infrared coverage over 10-400 c m - 1 requires a mercury arc lamp, mylar or metal mesh beam splitters, and a pyroelectric detector. The Fourier transform function, and data analysis utilities, are supported by a UNIX-based workstation running at 33 M H z . Accessories are available for infrared microscopy and other spectroscopic tech­niques.

Another F T - I R spectrometer, the QS408, is designed specifically for semiconductor characterization. Its spectral range of 220-4000 c m - 1 , at a resolution of 0.5 c m - 1 , supports the measurement of epitaxial layer thickness over 0.25-200 μηι, and of the concentration of interstitial oxygen and substitutional carbon in silicon. Specialized software is available to calculate these quantities. The results can be plotted in con­tour form to give a map of a semiconductor wafer. The sampling spot size is 5 mm, and the largest wafer which can be accommodated is 200 mm. In one version, the QS408 is configured for automatic handling of wafers.

4.2.2.4 Spectra-Tech Applied Systems, 200 Harry S Truman Parkway, Annapolis, MD 21401, USA

The ΙΚμδ/SIRM scanning infrared microscope is a F T - I R unit with 0.5 μηι spatial resolution in the visible, and 5 μιτι resolution in the infrared. A transmission mode, two reflectance modes, and a grazing incidence mode for surface analysis are avail­able. The sample position can be set manually or by computer. Integrated software supports two-dimensional mapping of infrared data.

4.3 Photoluminescence spectroscopy

4.3.1 Methods

The P L arrangement is the most straightforward of our three types of optical system. Figure 4.2 shows its main elements. The source can be any laser whose photon energy exceeds the band gap of the material to be examined, and whose power is sufficient to excite an adequate signal. Many commercial types of laser including H e N e and A r +

units meet these criteria. A n A r + laser operating on its blue 488 nm line produces photons at 2.54eV, which exceeds the gaps of silicon (1 .12eV) and A ^ G a ^ ^ A s (1 .42-2 .16eV). These photons can be generated at continuous powers of watts, whereas tens of milliwatts are often adequate to give good signals. This depends, however, on the material (as well as on system throughput, and detector sensitivity). Indirect gap semiconductors like silicon have poor P L efficiency. But laser power

Photoluminescence spectroscopy 51

Laser (Ar or Kr)

S

Tunable laser

L1

! C2

Grating monochromator

Sample in cryostat

Fig. 4 .2 Photoluminescence arrangement, with laser, sample and cryostat, monochromator, and detector ( D ) . Lens L2 focuses the PL signal; filters Fl and F2 block unwanted laser light; chopper CI modulates the light for lock-in detection. (The same arrangement serves for Raman spectroscopy, with the single-grating monochromator replaced by a double unit.) The tunable laser and chopper C2 are used for photoluminescence excitation (PLE) spectroscopy. Lamp S and lens L I are used for absorption spectroscopy. (After Lightowlers (1990).)

cannot be increased indefinitely, since too high an intensity (watts per square centi­meter) at the focused spot can damage a sample. It is usually possible to obtain an adequate signal-to-noise ratio without damage, by defocusing the laser or reducing its output power. If a laser providing fixed-wavelength lines is replaced by a dye laser or other tunable unit, the arrangement shown also serves for the alternative technique of photoluminescence excitation ( P L E ) spectroscopy, where the exciting energy is varied and the monochromator is set at a fixed wavelength to track the resulting lumines­cence. This technique offers advantages in some applications, because it allows res­onant excitation at important P L features such as exciton peaks. Figure 4.2 shows another feature, a secondary lamp source which illuminates the sample for absorbance measurements. A s the discussion in Chapter 3 showed, absorption and P L are closely linked. It is valuable to be able to perform either on a given sample.

The P L signal passes through a single-grating monochromator which selects a wave­length to transmit to the detector. The resolution of the system, its ability to accu­rately measure energy, is determined by the focal length of the monochromator, whereas the grating spacing sets the wavelength coverage. Even a modest 0.22m monochromator with a 1200 groove/mm grating covers the visible to the near infrared with an energy resolution of approximately l m e V at mid-range. Much higher resol­utions are available. A standard photocathode tube with the common S-l response is often adequate for P L work below 1.1 μιη (above 1.1 e V ) . A t longer wavelengths a photomultiplier tube with a GaAs or other composite cathode is useful. Germanium photodiodes are good for the near-infrared range (1.1-1.8 μιη, 0.7-1.1 e V ) . Further into the infrared, PbS and In A s detectors are used, and InSb photodiodes are es­pecially sensitive. For the best signal-to-noise ratio these detectors should be operated cooled, at 0°C for a photomultiplier, and at 77 Κ for a solid-state photodiode. Lightowlers (1990) provides a brief but trenchant summary of P L characterization and its instrumentation.

52 Measurement methods

The P L sensitivity can be increased, or the measurement time decreased, if the dispersive grating monochromator in Fig. 4.2 is replaced by a Michelson interfer­ometer. Such Fourier transform photoluminescence spectroscopy ( F T - P L ) has the advantages already discussed for F T - I R spectroscopy. Thewalt et al. (1990) point out that F T - P L is advantageous both above 1 μπι, where highly sensitive photon-counting methods are not available, and below 900nm. They describe a system with A r + or titanium-sapphire lasers and a commercial interferometer used as built, except for the addition of photoluminescence-collecting optics. The detectors were InGaAs and InSb units working at nitrogen temperature, and samples were cooled in superfluid helium. Lightowlers (1990) also shows an experimental arrangement for near infrared F T - P L based on another commercial Fourier spectrometer.

The best P L spectra come from samples held below room temperature. Lower temperatures reduce the thermal broadening of the excited carrier energies, which at temperature Γ is roughly kBT, where kB is Boltzmann's constant. This gives a signifi­cant broadening of 25 m e V at room temperature, which reduces to 6 m e V at 77 Κ , and to < 1 m e V at liquid-helium temperatures, for the finest work. Cooling therefore produces sharper, more readily identified peaks. It also tends to reduce the role of competing nonradiative paths for recombination, giving a higher efficiency for the P L process which results in improved signal-to-noise ratio. Finally, cooling prevents impurity centers from undergoing thermal ionization. Methods for attaining the necessary temperatures are discussed in the section on sample techniques.

One special application of P L is in making two-dimensional maps of properties over a semiconductor wafer. The short wavelength of visible light gives spatial resolution comparable to device dimensions. This technique has proven valuable enough that at least one commercial system is available, as described by Moore and Miner (1990). In the design (Fig. 4.3) the laser beam and optics are fixed to eliminate variations in beam geometry or intensity. A n x-y stage moves the sample in a raster pattern relative to the beam. The system accommodates two lasers to allow a choice of exciting energy. Typical laser power is 80 μ\Υ. The optics simultaneously collect P L light, and allow measurements of the reflected light. In addition, a camera with a white light source gives television images for site selection on the sample. A 0.32m monochromator with a computer-controlled grating selects the wavelength. A t each physical location on the wafer, the grating is scanned to produce a full spectrum which is stored on a hard disk. The P L beam can be mechanically chopped to allow sensitive lock-in detection. Detectors are a silicon p-i-n model below 1.1 μπι, or an InGaAs unit for 1.1-1.8 μπι.

Moretti et al. (1989) describe a different system, which uses an optical multichannel analyzer ( O M A ) , a linear array of silicon diodes. The O M A presents about 700 photodiodes behind an image intensifier, which enhances the P L signal. With appro­priate optics, this detects within milliseconds the entire spectrum at once, spread out over the array. A s in the design by Moore and Miner (1990), the system also moves the sample on an x-y stage while the optical train is fixed; but rather than sweep a grating, the O M A quickly records a full spectrum at each point in the spatial scan. The source is a 25 m W H e N e laser whose light reaches the sample via fiber-optic cable. This maintains optical alignment, but decreases the incident power to 9 m W . Both P L and reflected light are collected with a beam-splitter arrangement. Part of the light enters a photodiode, where the signal is used to ratio out fluctuations in the laser intensity. The P L light is analyzed with a 0.27 m monochromator. For strong signals, a

Photoluminescence spectroscopy 53

Laser 1

Jf-

NDF Laser wheel select

Illuminator

CCD camera

Laser 2

RL detector

Computer: image display,

instrument i J control,

data acquisition

Mono­ PL chromator detector

Sample on X-Y stage

Fig. 4 .3 Commercial system for two-dimensional PL scanning. Either laser can be selected. The NDF wheel is a neutral density filter to set beam intensity. The illuminator and charge coupled (CCD) camera simplify the setting of the beam location. For a HeNe laser line at 633 nm, the spot size on the sample is 10 μπι. See text. (After Moore and Miner (1990).)

2in. wafer can be scanned on a 1 mm grid (a total of 2000 spectra) in about 2h, the limiting factor being the speed of the x-y stage. Data analysis is computerized, with the computer searching for P L peaks and determining their wavelength, peak inten­sity, and integrated intensity.

4.3.2 Commercial photoluminescence systems

4.3.2.1 Bio-Rad, Digilab Division, 237 Putnam Ave., Cambridge, MA 02139, USA

The PL6100 is a F T - P L system, with an A r + laser operating at 488 nm, and a liquid nitrogen-cooled germanium photodiode detector. With these components, the system can examine semiconductors with gaps up to 2.5 e V , and can detect luminescence over the range 0 .7-1 .8eV (0.67-1.8 μπι). Samples may be cooled with a continuous flow liquid helium cryostat, with an option for immersion in liquid helium to permit operation to 2 K . A variant, the PL6120, replaces the flow system with a closed-cycle refrigerator which cools samples to 2 0 Κ without liquid helium. A built-in computer controls data analysis, display, and storage.

54 Measurement methods

4.3.2.2 SPEX Industries, 3880 Park Ave., Edison, Ν J 08820, USA

The SPEX line of monochromators, gratings, data-acquisition computers, sample holders and other accessories is suitable for P L spectroscopy. The modular design of the components makes it possible to configure varied characterization systems.

4.3.2.3 Waterloo Scientific, Inc., 419 Phillip St., Unit 9, Waterloo, Ontario, Canada N2L 3X2

The Waterloo SPM-200 is designed for P L mapping over semiconductor wafers, as discussed in detail above. It obtains a full P L spectrum at each point in the spatial scan. This makes it possible to determine at each point on the wafer the wavelength, intensity, and full width at half maximum of the peak, each of which gives different information about the material. With an optional accessory, the unit can also deter­mine epitaxial layer thickness by analysis of interference effects in reflected light. This provides a two-dimensional map of layer thickness over the wafer.

4.4 Raman scattering spectroscopy

4.4.1 Methods

T w o main factors influence the design of Raman spectrometers. One is the weakness of the signal, as shown in comparison with the much larger P L signal in Fig. 4.4, which also shows that Raman and P L data can be obtained with the same optical arrange-

488 nm, 200 mW,

80 Κ / PL \

Raman /

• I I I ι 2.50

Energy (eV) 2.00

Fig. 4 .4 Unpublished data (Perkowitz, 1991), showing the PL peak at the gap, and Raman phonon modes, from a Cdo.72Mno.2sTe film on a GaAs substrate. The diagram illustrates the small Raman intensity compared with the signal from a strong PL emitter. It also shows that both PL and Raman data can be obtained with the same experimental arrangement.

Raman scattering spectroscopy 55

ment. The second factor is that this weak signal is spectrally very near the orders-of-magnitude larger exciting laser signal. The typical Raman shift < * > p h o n o n / ( 0 i a s e r is only about 1%. This means that the feeble Raman peaks must be measured against a background of intense Rayleigh scattering. Raman measurements require the strongest available sources (short of damaging the sample), well-designed optics to filter out the undesired Rayleigh peak, and excellent detectors to record the few Raman-shifted photons.

Any powerful laser operating in the visible region, such as a H e N e unit, can be a Raman source. But the ability to choose among several lines gives flexibility in penetration depth, and a limited ability to excite resonance Raman scattering. Hence A r + and K r + lasers are good choices. Each delivers watts of power in any of several lines, the A r + from the ultraviolet to the green, the K r + toward the red end of the spectrum. For the exact coincidence in energy needed for true resonance Raman work, and for better control over the sampling depth, a tunable dye laser is one good choice. Available dyes make it possible to cover most of the desired energy ranges. The continuous wave power from a dye laser is usually milliwatts, much lower than that from A r + and K r + lasers; but the resonant enhancement more than compensates for this. Other tunable lasers are also available.

A typical Raman arrangement is similar to that for P L (Fig. 4.2) with one major difference: the single monochromator used for P L work does not discriminate suf­ficiently to separate the Raman signal from the strong Rayleigh light that accompanies it. Increased discrimination comes from the use of a double monochromator, consist­ing of two ganged gratings turning together and sequentially selecting the light. In some cases, for instance to examine Raman peaks within a few wavenumbers of the exciting line, triple monochromators must be used. It is usual as well to provide prefiltering before the monochromators to eliminate other unwanted light. A resol­ution Δ / = 0.1 c m - 1 is typical for a good double monochromator. The Raman system may also include means to polarize the exciting light and analyze the polarization of the scattered light. These polarization states are intimately connected to the orien­tation of a crystalline sample.

Since the Raman shift is small compared with the exciting frequency, the scattered light remains in the visible region. Standard visible light photomultiplier tubes ( P M T ) work well as detectors, if chosen and operated for maximum sensitivity and broad spectral coverage. They should be selected for a low dark count, and should be operated cooled to further reduce dark count as much as possible. Another approach is to use a solid-state O M A detector, such as that described in Section 4.3. This either decreases the time required to acquire the spectrum, or provides a better signal-to-noise ratio than a P M T gives in the same time. Its one drawback is that the resolution is determined by how many individual silicon diodes can be placed side by side over the width of the displayed spectrum. With present O M A technology, the resolution is generally lower than is achievable by a good double monochromator with a P M T . Nevertheless, the resolution with an O M A is adequate for many applications.

One reason that Raman scattering is excited in the visible or ultraviolet region is that the scattered intensity is proportional to ω 4 (equation (3.18)) , so efficiency is lost at the longer wavelengths of the infrared. But there are advantages to operating in the infrared. The strong process of fluorescence competes with and sometimes over­whelms the Raman signal. Fluorescence is excited by resonance with electronic states, which are less prevalent in the near infrared than in the ultraviolet-visible region. A

56 Measurement methods

T O V I D E O S Y S T E M

A N D M O N O C H R O M A T O R

z \

ο W H I T E

L I G H T

S O U R C E

L A S E R

L I G H T

L A S E R S P O T -~ 1 μηΐ

M I C R O S C O P E

O B J E C T I V E

S A M P L E

X - Y S T A G E

Fig. 4 .5 Schematic of a Raman microprobe. The microscope objective lens focuses the laser beam to a spot size of approximately 1 μπι. Scattered light is transmitted through the beam splitter to the double monochromator, giving the Raman spectrum of the selected region. Visible light from the illuminator is reflected from the sample to a video camera, giving a real­time image to allow positioning of the probe beam. The sample can be moved on an x-y stage.

Raman system incorporating Fourier methods offers enough sensitivity to compensate for the loss in scattering intensity, so that operation in the near infrared becomes possible and fluorescence problems are avoided. A typical system uses as source a Nd: Y A G laser operating at 1.064 μπι. Instead of the double or triple monochromator, a Fourier spectrometer collects the radiation, with a filter rejecting the strong Ray-leigh signal before it enters the interferometer. The detector is a cooled InGaAs unit.

In the variation of Raman characterization called microprobe Raman scattering, the exciting laser light is coupled to the sample by means of a microscope, which focuses the light to a spot as small as approximately 1 μιη across. A s shown in Fig. 4.5, the optics can be arranged so that the Raman scattered light is coupled to the mono­chromator through the microscope, and auxiliary illuminating light is sent to a video camera to give real-time images of the sample surface. This makes it possible to locate and probe areas of the sample as small as the beam size. The microprobe technique is becoming widely used to minutely examine materials and devices.

One of the strengths of Raman scattering, when polarization analysis is included, is its sensitivity to sample orientation. This means, however, that to obtain and analyze the spectra, the relation between the incident light and the sample surface must be carefully established. Perhaps the most commonly used geometry for semiconductor samples is the backscattering arrangement shown in Fig. 4.6. Most reports in the literature, and most examples I give in this book, use this configuration.

Further details about Raman instrumentation can be found in the five-volume series edited by Cardona (1975b), and Cardona and Guntherodt (1982a,b, 1984, 1989).

Raman scattering spectroscopy 57

SAMPLE

Fig. 4 .6 The backscattering geometry commonly used for Raman measurements from semicon­ductor samples. The exciting laser light passes through a hole in a mirror on its way to the sample. Light scattered from the sample surface is reflected by the mirror and collected by the lens, for transmission to a double monochromator.

4.4.2 Commercial Raman systems

4.4.2.1 Bomem, Inc., 450 St. Jean-Baptiste, Quebec, Canada G2E 5S5

The D A 8 F T - I R spectrometer described in Section 4.2.2.1 can be fitted with an attachment for FT-Raman excited by a N d : Y A G laser operating at 1.056 μπι.

4.4.2.2 Bio-Rad, Digilab Division, 237 Putnam Ave., Cambridge, MA 02139, USA

The FTS60A F T - I R instrument described in Section 4.2.2.3 supports an accessory for FT-Raman work.

4.4.2.3 Bruker Instruments, Inc., Manning Park, Billerica, MA 01821, USA

The IFS48 and IFS66 units discussed in Section 4.2.2.2 can be fitted with the FRA106 unit which adds a N d : Y A G laser and associated optics. Bruker also provides the RFS100 stand-alone FT-Raman system. This incorporates a permanently aligned interferometer, and allows for selection among different exciting lasers.

4.4.2.4 Jobin-Yvon Optical Systems, Instruments SA, Inc., 6 Olsen Ave., Edison, NJ 08820, USA

The U1000 Raman system is a 1 m double monochromator which supports changeable gratings to vary wavelength coverage and resolution. It can accommodate a Raman microprobe accessory and either single channel photomultiplier or multichannel

58 Measurement methods

detectors. The M O L E S3000 Raman system is a triple spectrometer which also accommodates a microprobe, single or multiple channel detectors, and an optional turret housing two selectable gratings. Its supporting software also allows use for transmission spectroscopy and other techniques. The T64000 system includes a 0.64 m triple monochromator which can be used in varied configurations to optimize wave­length selectivity or resolution. The system incorporates a Raman microprobe. Inte­grated software supports data acquisition and detector control functions.

4.4.2.5 SPEX Industries, 3880 Park Ave., Edison, NJ 08820, USA

The Model 1488 analytical Raman system is an integrated instrument with air-cooled A r + laser, 0.22 m double monochromator, sample compartment, photomultiplier detector which covers 200-850 nm, and a PC-compatible menu-driven data-acqui­sition system. SPEX also provides a variety of double and triple monochromators, gratings, sample holders, and other accessories for Raman spectroscopy, whose modular design allows the design of custom systems. Among these is the Model 1877 triple spectrograph for multichannel detectors. Wide-range, low dark count photo-multipliers and a charge-coupled multichannel detector are available. The Micramate accessory is an addition for microprobe Raman work.

4.5 Sample considerations

4.5.1 Geometry and surfaces

One major advantage of optical characterization is its freedom from the sample manipulation needed for other methods, such as the application of electrical contacts. But some comments and cautions are in order. Surface smoothness and flatness are important for accurate optical measurements. Most as-grown semiconductor struc­tures evince shiny surfaces which show that they are flat and smooth on the scale of visible wavelengths. When this is so, they are also more than acceptable at the much longer infrared wavelengths. When surface geometry is problematic, however, measured reflectances and transmissivities may be erroneous, and light scattered from a rough surface can interfere with Raman and P L spectra. Standard mechanical and electrochemical polishing methods can achieve good surface finishes for bulk samples of most semiconductors. However , these produce surface damage which is not rep­resentative of the bulk sample, and influences the spectroscopic results if the pen­etration depth of the light is not great.

Occasionally, excessive flatness can be a problem. Infrared interference fringes result if a transmitting bulk sample or film has highly parallel front and back surfaces. Fringes are useful to find sample thickness, but may complicate the analysis of spectra when other parameters are of interest. If the analysis cannot take them into account, or if they directly interfere with important spectral features, the only recourse may be to mechanically vary the thickness of a bulk sample or roughen its back surface, or to grow a film on a wedged substrate.

Surface films, due to contamination, oxidation, or other causes, may be an issue at shorter wavelengths, where the penetration depth is less than in the infrared; but it is

Sample considerations 59

precisely this sensitivity which can be valuable, for instance, in the use of the Raman microprobe to find contaminated areas. Surface films are less likely to be a factor in the infrared, where the thicknesses are rarely great enough to seriously affect spectra unless there is a very high carrier density or other source of absorption.

The lateral dimensions of the sample can be a concern in the extreme far infrared. Diffraction effects which distort a measured spectrum come into play at wavelengths comparable to these dimensions. This could be a factor in the extreme far infrared for samples smaller than a few millimeters across, but is not an issue when visible to mid-infrared light is used for typical bulk samples. It is important to keep in mind that diffraction is always the limiting factor in examining substructure in a sample, such as micrometer-size strips in a device configuration. Wavelengths longer than the sub­structure dimensions cannot return meaningful information.

4.5.2 Cooling

Much useful optical work can be done at room temperature; however, as I have pointed out, cooling of both detectors and samples can be helpful or even necessary. Detectors are usually sold along with cooling facilities to set and maintain their best operating temperatures. For samples, various alternatives are available. The first consideration should be whether sample cooling is needed at all. This complication is to be avoided or simplified as much as possible for routine characterization. But for exploratory work, or for research to establish the accuracy of a characterization method—for any application which requires high resolution or an exceptional signal-to-noise ratio—cooling may be essential. If so, the next decision is whether liquid nitrogen temperatures are adequate. Liquid nitrogen is inexpensive, and requires little in the way of specialized training or techniques. A sample is conveniently cooled by mounting it on a cold finger connected to a liquid nitrogen Dewar. If the assembly is held under a vacuum of 0.1 T , samples can be maintained to within a few degrees of 77 K .

There are also alternatives for attaining sample temperatures below 77 K . Since liquid helium is far more expensive than liquid nitrogen, and requires trained people for efficient use, a careful choice among these can minimize cost and complication for characterization work. When temperatures of 5-10 Κ are sufficiently low, a commer­cial continuous-flow helium system is convenient. The sample is mounted on a cold finger, which is cooled by a steady flow of liquid helium from a Dewar at a typical rate of 1-31/h - 1 . Proper shielding of the cold finger is important. Mechanical refrigerators which can hold samples at 10-20 Κ are also convenient, but vibration problems are sometimes reported.

If temperatures of 4 Κ or less are essential, immersion in liquid helium is the only choice. This is less convenient than a continuous flow system, and bubbling of the helium can interfere with the optical signal. If the liquid helium is pumped, bubbling is eliminated, and the temperature is reduced to 1.8 K , which gives a minute thermal broadening of only 0.15meV. Several commercial suppliers provide Dewars for optical work with a choice of windows and configurations for different types of optical access. Some designs also include an electrical heater and control equipment, to reach and hold any temperature from 1.8 to 300 Κ .