Ultrasonic Multiplayer Metal film metrology
06/01/1997
Ultrasonic multilayer metal film metrology
C.J. Morath, G.J. Collins, R.G. Wolf, Rudolph Technologies Inc., Flanders, New Jersey
R.J. Stoner, Brown University, Providence, Rhode Island
Picosecond ultrasonic laser sonar (PULSE) is the first metrology technique capable of measuring multilayer metal films directly on production wafers. PULSE uses ultrafast light pulses to launch ultrasonic waves into such films, then analyzes the resulting echoes to measure the thickness of all layers simultaneously with a precision of better than 1%. This method can measure stacks of six or more layers with individual layer thicknesses in the range from <20 ? to over 5 ?m. It can also characterize film density, stoichiometry and roughness, and interface properties such as adhesion and contamination.
Process-control film-thickness metrologies can be roughly categorized according to whether they measure transparent or opaque (i.e., metal) films. Optical techniques for transparent films, like ellipsometry, have steadily improved since the development of the first integrated circuits. Modern multi-angle, multiwavelength ellipsometers can measure films ranging from gate oxides <30 ? thick to photoresist layers thicker than 10 ?m. These systems can also make multiparameter measurements on complex multilayers such as oxide-polysilicon stacks and advanced interlayer dielectrics (ILDs). Tool features such as small measurement spots, pattern recognition software, and high-precision wafer positioning have made it possible to perform many transparent film process-control measurements directly on product wafers. The reduction in blank monitor wafer usage has substantially reduced process-control costs and increased efficiency.
Meanwhile, metrologies for metal films have remained very limited compared with the increasingly sophisticated deposition processes they monitor. Although four- or five-layer multilayer metals (MLMs) deposited in cluster tools are now common, they are most often monitored by sheet-resistance measurements on blanket films corresponding to each layer. As much as 10 percent of the production capacity of a multichambered deposition system is consumed by blanket wafer monitoring. The associated costs of capital equipment, test wafers, reclaim, labor, and lost production are substantial (see table). For 300-mm wafers, costs will increase, making an even stronger case for eliminating monitor measurements in favor of measurements on completed stacks.
Metal-film metrology on completed stacks also opens up new possibilities for process control. For example, conventional single-layer metrologies cannot detect a costly MLM layer omission. Characterizing MLM layers as they are deposited in the stack allows more accurate process control, since film properties such as growth rate, roughness, and interlayer reaction can all depend significantly on neighboring layers.
Opaque film metrology requirements
There is a clear need for a measurement technique that can be performed on complex metallization stacks as deposited on product wafers. Requirements include:
a high degree of precision and repeatability over a wide range of film thicknesses;
a noncontact and nondestructive means of measurement withthe ability to measure six or more layers simultaneously;
a measurement spot that can fit within existing test structures (typically 20 - 40 ?m in dia.);
a measurement time no longer than the process deposition time;
robust pattern recognition; and
a high level of automation.
PULSE, a novel measurement technology that meets these requirements, was pioneered by Humphrey Maris and Jan Tauc and their students [1-7] at Brown University beginning in 1984. Over the past two years, in collaboration with researchers at Brown, Rudolph Technologies Inc. has developed the MetaPULSE system, based on the PULSE technique, for metal film process-control applications.
How PULSE works
One of the greatest virtues of the PULSE technique is its conceptual simplicity: the system generates and detects sound in opaque films with pulses of light. An ultrafast laser produces a brief (10-13 sec) light pulse focused onto a region of the film surface approximately 10 ?m in dia. (Fig. 1a). The absorption of this "pump" light pulse causes a small (5-10?C) rise in the film`s temperature near the surface. The resulting rapid thermal expansion generates a sound wave that propagates away from the surface at the speed of sound (Fig. 1b). The heat dissipates into the sample in a few nanoseconds, playing no further role in the measurement. When the sound wave encounters the interface with a lower layer, some of it reflects back toward the surface as an echo; the remainder continues down into the lower layers of the sample (Fig. 1c).
When the echo from the first interface reaches the surface, it changes the sample`s reflectivity (Fig. 1d). A second light pulse, diverted from the pump beam by a beam splitter, detects the change. This detection, or "probe," pulse is delayed relative to the pump pulse by guiding it to the sample over a longer path (Fig. 2).
 |
Figure 1. Steps in a PULSE measurement: a) instantaneous heating produces a sound wave that b) propagates down through the first layer; c) a partial echo returning from the first interface is d) detected by a second laser pulse.
 |
Figure 2. Simplified schematic of the MetaPULSE measurement system.
 |
Figure 3. Time-dependent change in reflectivity for TiN/Si. The time difference of 42 psec between successive echoes gives a thickness of 2000 ? for the TiN, from Eqn. 1.
The time t for the sound wave to propagate through the film at the speed of sound* vs and return to the surface is related to the film thickness d by
Figure 3 shows a typical series of echoes for a single layer (2000 ?) of TiN on Si. The echoes appear as changes in the sample reflectivity vs. the time delay between the pump and probe pulses. Successive echoes are separated by an interval equal to the time it takes for the sound wave to travel to the bottom of the film and return to the surface. In TiN, sound travels at vs = 95 ?/psec so that about 42 psec elapses between echoes in Fig. 3.+
The echoes also decrease in amplitude with increasing time. Assuming an ideal interface between film and substrate (and in the absence of bulk attenuation), each successive echo becomes smaller than the previous one by a factor equal to the acoustic
reflection coefficient R between the film and substrate, namely
where ZFILM and ZSUB are the acoustic impedances (defined as the product of the density and sound velocity) of the film and substrate, respectively. Nonideal properties of an interface between two layers in a film stack, such as poor adhesion, roughness, interdiffusion, or contamination, can cause the measured acoustic reflection coefficient at an interface to differ substantially from the value predicted by Eqn. 2. Thus, the amplitudes of the returning echoes can indicate subtle variations at the interfaces between layers. PULSE uses only the echo times to determine film thicknesses, so its metrological precision is insensitive to echo amplitudes. This robust behavior allows the technique to monitor individual metallization layers and their interactions within a stack.
PULSE technology applications
The MetaPULSE system has been tested successfully on virtually all opaque films used in semiconductor manufacturing, including: Ti, Co, and Ti and Co silicides; TiN, WN; AlCu, Cu, Pt, and W; complete M1-M6 stacks (TiN/Ti/AlCu/TiN/Ti/SiO2); W adhesion stacks; PVD and CVD Ti/Ti adhesion stacks; and buried and top-layer ILDs.
Applications range from monitoring the first Ti deposition and silicide formation, through measuring Ti/TiN adhesion stacks, tungsten, completely clad metallization stacks, and ILDs.
The basic physics underlying PULSE`s generation and detection mechanism are robust and do not require frequent calibration, hardware adjustments, or reference samples. Calibration is simply based on the distance from the beam splitter to the probe-beam reflector mirror. An encoder-calibrated translation stage positions the mirror to an accuracy of better than 5 ?m over its full range of travel, corresponding to an uncertainty in film thickness of less than 1 part in 104. As a result, system-to-system matching is simple and thickness measurements are inherently accurate, highly repeatable, and precise.
To make a measurement, an engineer first constructs a model film stack from a stored list of materials. After the system performs a measurement, software adjusts the parameters of the model film stack iteratively to produce a best fit to the experimental data. The simulation software uses rigorous physical models [2] for both the optical and acoustical properties of the sample.
 |
Figure 4. Time-dependent change in reflectivity for a sample of 2000-? TiN/200 ?-Ti/Si.
 |
Figure 5. a) Measured time-dependent reflectivity change (blue) and best model fit (red) for a sample of TiN/Ti/Al-Cu/TiN/Ti/SiO2/ Si, and b) calculated layer thicknesses
To produce simulated data for comparison with the measurement, the modeling software calculates:
1. the absorption of the pump light pulse;
2. the generated strain distribution near the surface;
3. the time-dependent propagation of this strain distribution away from the sample surface; and
4. the time-dependent change in the sample reflectivity arising from the arrival of the strain wave near the surface.
To create a model film stack for a TiN/Ti adhesion layer, the engineer selects the substrate material, then adds the Ti and TiN films. Thickness values are entered to reflect the nominal process values. Limits on film values and goodness-of-fit can help detect process variations. Once created, the trial film stack can be used to test the application and can be easily incorporated into production recipes for routine thickness measurements.
For accurate measurements in multiple film stacks, changes in the film thickness of one layer must not affect the measured thickness of the others. Figure 4 shows data for a sample with 200 ? of Ti buried beneath a 2000-? TiN film. The echo pattern is more complex than in Fig. 3 because sound is reflected at both the TiN-Ti and Ti-Si interfaces. The return time of 42 psec for the echo reflected off the TiN-Ti (first) interface is unchanged since the TiN film has the same thickness in both structures. The 6.5 psec/interval between the first and second peaks in Fig. 4 corresponds to sound making a round trip through the Ti layer. The system is able to provide independent thicknesses for the two films because the thickness of the TiN film is completely determined by the first peak in Fig. 4, while the thickness of the Ti film is completely determined by the interval between the first and second peaks.
This type of analysis can be extended to a film stack with several layers. As the number of layers increases, the echo patterns become more complex and more is gained by comparing measurements with simulations. Figure 5a shows the pattern of echoes for a typical five-layer interconnect metallization, along with the best fit of the model to the data. The sample consists of TiN/Ti/Al-Cu/TiN/Ti/SiO2/Si. At small elapsed times, "ringing" of the thin TiN and Ti surface layers on top of the thick Al-Cu causes oscillations. After about 120 psec, echoes from the Al-Cu and underlying layers begin returning to the surface. The layer thicknesses are shown in Fig. 5b, with fit errors smaller than 1% for all six deposited layers. Film roughness creates a broader echo than expected for an ideal Al-Cu film; the calculated fit is based on a model in which the RMS roughness for the film was 62 ?. Generally, the system can automatically determine roughness along with thickness for films such as Al-Cu and W.
Figure 6 demonstrates the system`s ability to detect missing layers in complex MLMs. A second sample consisted of TiN/Al-Cu/TiN/Ti/SiO2/Si. This sample, which is nearly identical to the one measured in Fig. 5, except for the absence of the Ti layer beneath the top TiN layer, illustrates the system`s ability to detect a misprocessed metallization stack. The absence of the Ti layer causes a dramatic change in the echo pattern (Fig. 6a). The system easily detected the absence of the upper Ti layer (Fig. 6b).
As discussed previously, the timing of the echoes depends on the layer thicknesses, while the relative amplitudes of the echoes depend on the acoustic reflection and transmission coefficients of the interfaces. The echo amplitudes would be affected if, for example, the top TiN layer had been produced with an out-of-specification density (thereby affecting its acoustic impedance) or if an etch process prior to the MLM deposition had left a contaminant layer at the lower Ti-SiO2 interface. Indeed, experiments carried out at Brown in collaboration with IBM have detected the presence of even a single monolayer of etch polymer [7]. The ability to detect such subtle deviations as the absence of an intended layer, the presence of a buried contaminant layer, lack of adhesion between layers within a stack, or interfacial roughness, adds considerably to the tool`s value.
 |
Figure 6. a) Measured time-dependent reflectivity change (blue) and best model fit (red) for a sample of TiN/Al-Cu/TiN/Ti/SiO2/Si, and b) calculated layer thicknesses. This stack is identical to the one measured in Fig. 5, except for the absence of the 319-? Ti layer beneath the top TiN layer. This missing layer dramatically changes the echo pattern.
Conclusion
This technique provides new capabilities for metal process monitoring. All types of metal films ranging in thickness from <20 ? to several microns can be measured with a typical repeatability of better than 1%. The thicknesses of six or more films in a multilayer interconnect can be quickly determined from a single measurement on a completed stack. All measurements may be made on product wafers within sites as small as 20 ?m. Material and structural properties that can be determined include film density, interfacial roughness, adhesion strength, and silicide phase. The technique enables a transition away from expensive blanket wafer monitoring to product-based monitoring, while at the same time introducing an unprecedented range of capabilities for metal film and interface quality monitoring.n
References
1. C. Thomsen et al., "Coherent Phonon Generation and Detection by Picosecond Light Pulses," Phys. Rev. Lett., Vol. 53, p. 989, 1984.
2. C. Thomsen et al., "Surface Generation and Detection of Phonons by Picosecond Light Pulses," Phys. Rev., Vol. B 34, p. 4129, 1986.
3. H.T. Grahn, et al., "Picosec Ultrasonics," IEEE J. Quant. Elec., V. 25, p. 2562, 1989.
4. H-N. Lin, R.J. Stoner, H.J. Maris, "Nondestructive Testing of Microstructures by Picosecond Ultrasonics," J. Nondestr. Eval., Vol. 9, p. 239, 1990.
5. H-N. Lin et al., Amer. Soc. of Mech. Eng., Vol. 140, "Picosecond Optics Studies of Vibrational and Mechanical Properties of Nanostructures," Acousto-Optics and Acoustic Microscopy, ASME, New York, p. 134,1992.
6. C.J. Morath et al., "Picosecond Optical Studies of Amorphous Diamond and Diamondlike Carbon: Thermal Conductivity and Longitudinal Sound Velocity." J. Appl. Phys., Vol. 76, p. 2636, 1994.
7. G. Tas et al., "Noninvasive Picosecond Ultrasonic Detection of Ultrathin Interfacial Layers: CFx at the Al/Si Interface," Appl. Phys. Lett., Vol. 61, p. 1787, 1992.
For more information, contact: George Collins, director of marketing, Rudolph Technologies Inc., One Rudolph Rd., Flanders, NJ 07836; ph 973/691-1300 ext. 242, fax 973/691-5610, e-mail [email protected].
*More precisely, the compressional sound waves generated by PULSE travel at the longitudinal velocity of sound.
+The top speed of the Concorde is around 7 ?/psec. Most metals have a longitudinal sound velocity in the range from 40-100 ?/psec.