analyzer chamber and which contains equipment for secondary ion mass spectroscopy, electron-stimulated desorption, ion milling, metal deposition, and gas-solid reactions. A sample-preparation area is attached to the inlet chamber of the spectrometer consisting of a dry box with introduction lock. The dry box is under 5 psi pressure of nitrogen, which is dried over LN2. A high-temperature furnace is provided for controlled growth of oxides on silicon. The furnace has an introduction lock permitting direct transfer of samples to spectrometer without aqueous or carbonaceous contamination. A plasmacleaning facility is interlocked to the dry box to permit removal of trace organic contamination.

Sample Preparation

The

For reproducible results in XPS, considerable attention must be given to samplemounting procedures. In this specific case, where the SiO2/Si interface is the object of study, reproducible electrical contacts must be fabricated to the silicon substrate. problems of ohmic contact and that of physical mounting were solved by using lowmelting-point solders, such as InSn and GeAu. The sample and the mounting configuration are indicated in figure 11. A gold (Au) platen is prepared by washing with TCE, acetone, then several rinses with high resistivity (20 megohm) water. The platen is bonded to the rough underside of the oxidized silicon wafer by heating on a hot stage in the nitrogen atmosphere of the dry box. A gold window is placed over the top surface of the sample. The area illuminated by x-rays is indicated in the center of the window zone and is somewhat less than 1 × 5 mm (0.3 x 4 mm). Since the detector of the HP-5950A spectrometer is positionally sensitive, the long dimension can be reduced to approximately 1 mm, giving 0.3 mm2 as the minimum analytical area.

During the experiments described in the following sections, the spectrometer was operated at a pressure of 6 x 10-10 torr. The ambient gas consisted primarily of H2, He with lower levels of N2 and N. H2O and A were minor components, but some CO was present.

Applications

Silicon-Silicon Dioxide Interface

In CMOS technology, the fundamental device characteristics are determined by the oxidesemiconductor interface. Our initial experiments indicated that the escape depth of

In order to demonstrate the chemical shift between silicon in the elemental state and silicon in the +4 oxidation state (SiO2), the spectra of figure 13 show both cases. The upper spectrum arises from the 2p doublet of elemental silicon. This sample is identical to that of the lower spectrum except that, in this case, the SiO2 has been stripped with H20/HF and a monolayer of fluoride is present on the sample surface. The lower spectrum is due to a wet 40 A oxide grown on silicon 100. In this spectrum, the Si 2p line due to the substrate occurs at nearly the same energy as in the elemental case. The peak at higher binding energy (104.3 eV) is due to the oxidized silicon of Si02. This shift of about 4.0 eV is characteristic of the 2p binding energy differences between Si° and Si4+.

an electron hole pair. If the electron is emitted into the vacuum, a hole is left behind. Therefore, a positive charge can build up in the thin surface layer interrogated in the XPS experiment (~ 100 Å). The steady-state surface charge reflects the balance of the depletion current (photoemission) against the replenishment currents (vacuum secondaries, tunneling currents, etc.). Since the vacuum secondaries are minimized through x-ray collimation, and low x-ray flux densities, a significant positive surface charge can be developed in Si02. In very thin oxides this potential is ultimately limited by electrons tunneling from the Si interface.11

A corresponding negative potential can be induced by increasing the current density of low-energy electrons to the sample surface from the flood gun. Therefore, a bias of either polarity can be applied dynamically in the course of the XPS experiment.

Figure 14 illustrates the effect of bias on Si 2p spectra of a 44 Å Si02 layer on silicon (100). The positive bias spectrum is given as the lower curve. The negative bias spectrum (upper curve of figure 15) was recorded while bombarding the sample surface with electrons of an average energy of 8.6 eV above the vacuum level. A change in line shape of the elemental line is observed, as is a change in the relative position of the oxide peak.

Previous work of just two to three years ago identified the escape depth of photoelectrons in SiO2 at 10 15 Å ().2 Figures 15-a and 15-b give the Si 2p spectra of a 113 Å film of SiO2 over Si. In figure 15a, the large difference in relative intensities of the oxide and substrate peak are evident, while figure 15b shows the line shape of the substrate silicon line at higher gain (20 X). Its binding energy of 99.06 eV is in good agreement with that of the Sio peaks of figures 12 and 13.

The Si4+ state (SiO2) is clearly resolved from that of Si°. The determination of silicon to oxygen ratio can be directly calculated by comparing the oxidized silicon peak intensity with the intensity of the oxygen is peak. The magnitude of excess silicon, and reduced silicon oxides can now be directly measured, within the sensitivity of the experiment. For these thin oxides, no ion milling is required to see the interface and hence no ion-stimulated redistributions need be considered in interpreting the data.

Migratory-Gold-Resistive-Short

The technology of metallization in integratedcircuit devices is based on either aluminum or gold. The presence of moisture within the device package has long been known to leak to electrolyte corrosion of aluminum. This corrosion results in a well-documented fail

ure mode. The chemistry of gold suggested that metallization schemes using this noble metal would be free from susceptibility to such electrochemical attack.

Recently, electrolytic corrosion has been observed and described phenomenologically as the migratory-gold-resistive-short (MGRS) mode. 12 The failure is seen as the abrupt decrease of resistance between two or more adjacent metal stripes. The shorting current are on the order of milliamperes. SEM micrographs of failed parts reveal that fernlike or dendritic gold deposits have formed a bridge between metal stripes.

The dendritic deposits appear to progress from the negatively-biased stripe (cathode) toward the positive one. The general form or morphology of the growth is quite specific and similar to that of deposits found in electrochemical silver migration.1 13 This well-known phenomenon results from the surface transport of silver which is driven by large electric fields. This surface transport is strongly dependent on the presence of moisture and on the nature of the substrate surface.

The MGRS failure is observed on the surface of integrated circuits using Ti/W/Au metallization which has been covered with a layer of e-beam-evaporated quartz. Failure analysis of affected parts showed significant levels of water in the purportedly hermetically sealed package.

The presence of water made possible a corrosion cell. This implied that the rate of attrition of parts is determined by the amount of time adequate potential is applied across adjacent metal stripes. The major difficulty, however, was that dendritic growth requires as few as 1010 atoms of gold to bridge a stripe.

Since gold cannot be simply dissolved in aqueous solutions, we studied the surface of failed and reference parts to determine the nature of the impurities residing atop the passivating glass layer. The presence of significant levels of halogens was observed by means of XPS.

This is an example of surface corrosion chemistry, and three points are relevant to this discussion: 1) positionally sensitive detection, 2) limiting sensitivity, and 3) decomposition of sample system. The first of these points is illustrated in figures 16 and 17. Figures 16 shows a 0 1000 eV binding energy scan of the top of the chip, together with the bonding pads, wire leads, and some parts of the package base. This is determined by the relative intensities of the gold 4f photoelectron line at about 85 eV and of the oxygen ls line at 530 eV. The gold line is approximately 25% more intense than the oxygen ls line. In figure 17, the detector is adjusted to scan the top of the chip, and the oxygen/gold intensity ratio has changed by a factor of 22.

The analysis of the surface of the glassivated substrate by XPS indicates that significant quantities of the halogens (c1 ̄, Br ̄, I ̄) are present together with hydroand fluorocarbons and several trace metals, such as Na+, K+ and Hg. Of the halogens,

Br was most abundant while Cl and I had relatively equal concentrations. The bromine 3d region of the photoelectron spectrum is given in figure 18. The structure arises from at least 2 different chemical states of the element. The iodine 3d region of the spectrum is plotted in figure 19. spin orbit doublet is clearly defined in this spectrum.

The

Over the range of failed and unfailed parts which have been examined by XPS, the concentration of I has varied over four orders of magnitude. Levels of iodine as low as 1 ppm can be readily detected. This represents a sensitivity limit about three orders of magnitude lower than was previously observed. 2 This is related to the relatively high photoelectric cross-section of the iodine 3d electrons for 1.5 keV x rays, and to the magnitude of the spin-orbit splitting. This separation of two lines with an appropriate intensity relationship facilitates the use of sophisticated noise-removal techniques for data manipulation.14

Interestingly, many of these impurity atoms were not observed in electron-stimulated Auger and x-ray spectra. This suggests that for many problems, electron-stimulated desorption is a serious experimental difficulty. Monochromatized x-ray excitation offers a minimal source of secondary electrons for such desorptions.

Mobile Ion Instability

The role of mobile ions in causing instabilities in MOS devices is well known and sev15-19 eral reviews have appeared. The pre

vious work has demonstrated that sodium is the most common source of such instabilities. Sodium contamination in thermally grown Si02 is often introduced by the sodium in the oxidizing or annealing ambient.20 Most of the sodium present in the ambient can be introduced through the walls of the furnace tube since this tube is usually fabricated of silica, which permits extremely high diffusion rates of sodium at the required operating temperatures. Additional furnace tube liners are often used to reduce this source of contamination. The processing steps following high-temperature oxidation are also potential sources of sodium contamination, since most solvents organic and aqueous-show significant sodium levels if not controlled. The effective concentration of sodium in the bulk arising from surface contamination of the oxide is directly determined by the maximum temperature to which the structure is exposed. Finally, sodium contamination can be introduced during metallization unless extreme precautions are taken.

It has been established that the rate of removal of sodium from the Si02/Si interface is much faster than that from the Si02/metal or Si02/vacuum interface. The mobility of Na ions in the "bulk" of the Si02 is quite high (at least 4 × 10-13 cm2/V sec at room temperature). The rate-limiting step during forward drift appears to be the "emission" of ions from "traps" at the Si02/metal interface. The trapping at the Si02/Si interface is much weaker and is strongly dependent on the conditions of oxidation and annealing.

The existence of two forms of traps for Na+ at the SiO2/Si interface has been suggested where one of these becomes saturated at high ionic contamination levels. This patch model has been developed to explain variations at the metal/oxide interface as well.21

Recently, methods have been developed to getter sodium ions in Si02 or to neutralize Na at the Si/SiO2 interface. Phosphosilicate glass (PSG) is widely used to trap or getter Na* moving from the metal/Si02 boundary to the Si02 interface. Recent experiments 22 have indicated that the addition of HC1 in the high-temperature oxdiation step modifies the silicon interface and permits charge neutralization of

In a previous section, it was observed that a positive or negative bias could be applied to the surface of an insulating sample in the course of the XPS experiment. The combination of applied bias and elevated temperatures was employed to pull the Na+ in the oxide out to the oxide/vacuum interface, or alternately push the Na+ into the oxide silicon interface. The effects on the silicon, oxygen and sodium spectra have been catalogued and interpreted. 23 Figure 20 gives the Na 1s spectrum of an as-processed device. Electronic measurements of this sample indicated a mobile-ion concentration of about 1011 atoms/cm2. This spectrum is the result of 51 hours of accumulation time. In figure 21, the Na ls spectrum of a different area of the silicon wafer is given. In this case, however, the Si02 film has been etched in H20/HF. The accumulation time for this spectrum is identical to that of figure 20. A composite of the peak positions observed in this experiment is reproduced as figure 22.

Unfortunately, thin overlayers of carbon are difficult to analyze. Electron- and ionstimulated surface analyses cause rapid desorption of the layer in vacuum, and can induce chemical changes in its structure. Since XPS involves the lowest magnitude of electron flux to the sample surface, we have

begun a study of organic contamination in MIS device processing.

Figure 23 illustrates the presence of such films. This 01000 eV binding energy scan is that of a kovar package interior. Gold, copper, chromium, iron, cobalt, oxygen, chlorine, fluorine and a number of other elements are readily observed. The carbon line at about 285 eV represents a major component of this surface.

Simple element detection is not an adequate use of XPS. The importance of this experiment is the ability to identify chemical parameters. This is illustrated in figures 24 and 25. These represent scans of the surface of a GaAs sample on which a thin oxide has been grown. Experiments in a solar-cell research program at JPL have indicated that oxides grown in different conditions of carbon contamination had markedly superior operating parameters. The sample giving rise to figure 24 was significantly superior to that of figure 25. In both spectra, a number of aliphatic carbon states can be assigned. Similarly, carbonyl or carbon-oxygen chemistry is found in both samples. In the superior device an amide carbon signal is observed. The presence of amines in an epoxy used to attach the wafer to a metal substrate had resulted in the incorporation of some of these compounds on the surface of the GaAs. This observation permits the design of a process to inject this species at the surface. If this had not been observed directly, the process would have to be developed through random optimization of conditions.

Figure 25 gives a final illustration of the tenaciousness of surface adsorbed carbon. This silicon sample had been etched in hot H20, HF under dry N2 and directly introduced to the spectrometer. This 0-1000 eV spectrum is now dominated by the silicon 2s and 2p lines. The fluorine ls line is clearly observed as is the fluorine KLL Auger structure. Note, however, the intensity of the C ls is still quite significant. In these experiments, silicones, freons, and simple halocarbons can be readily detected. Proteins and lipids resulting from human handling can also be observed. In our work, only cleaning in an 02 plasma resulted in removal of the carbon overlayer.

Etchant Residues

Yet another of the fundamental chemical steps in device processing involves the etching of metals and oxides in conjunction with photoresist masking. Although the chemistry of

metal and oxide dissolution is well understood for bulk materials, little work has been done to characterize the effect of these solution techniques on the device surface.

One of the basic difficulties with liquid/ solid reactions concerns the possibility of etchant residues deposited on the wafer surface. Often the presence of the surface alters the solubility of the end products of the etch and consequently such films are exceedingly difficult to remove. Proper removal requires the ability to directly monitor their presence. In figure 26, several peaks are observed at high binding energy which can be attributed to the presence of two or more forms of cerium oxide.

This sample was prepared from a wafer which had a chromium/gold metallization over the gate oxide. The metal contact was etched with KI3 solution to remove the gold, followed by ceric ammonium sulfate solution to dissolve the chromium. This sample was then treated with hot H2O/HF to remove the Si02 film. After these processing steps, little tri-iodide is observed on the surface, but the cerium remains.

Such residues, if undetected, can then be driven into the device by subsequent processing.

Figure 27 shows a narrow energy scan of the cerium region showing the splitting of states in the individual components of the cerium 3d doublet. Exposure of the wafer to chemical solutions of appropriate pH can virtually eliminate this oxide residue.

Heavy Metal Impurities

In the region of 920 and 950 eV in the spectrum of figure 28, components of the copper 2p doublet can be detected. Most of the samples studied in our laboratory are fabricated on single-crystal silicon substrates which have been polished by the copper-ion polishing method. This polishing system involves exposure of the wafer to cupric ion solutions. After polishing, these wafers are pre-oxidized to a depth of several thousand Angstroms. This oxide is then stripped to remove any heavy metal contamination as well as crystallographic damage introduced by the polishing mechanism.

Our studies indicate that copper persists at the Si/SiO2 interface in spite of these precautions. Figure 28 illustrates this situation with the copper 2p region of a freshly etched wafer. The copper 2p doublet is