UBM Measurement by Microbeam X-ray Fluorescence
BY RYAN NELSON, Thermo Electron Corp.
As chip manufacturers drive to smaller device size follow the International Technology Roadmap for Semiconductors (ITRS), material challenges arise in every part of the manufacturing process. Interconnecting signals from the chip to the outside world raises new issues with metal films used for interconnects and under-bump metallurgy, and in the metallurgy of leadframes used in chip packaging. To ensure high performance, the composition of metal layers is more complex and must be kept in tighter control. In addition, the thickness of each layer in the metal stack is moving from the micron down to the nanometer level. These changes are based on economic, material performance and environmental regulatory requirements.
Energy dispersive X-ray fluorescence (EDXRF) is a widely accepted, non-destructive technique for measuring the thickness and composition of metallic multilayer materials. In recent years, progress in EDXRF instrumentation development has accommodated increasing demands in terms of smaller spot size, higher throughput, improved accuracy, and better precision. By using collimated X-rays to achieve small spot sizes and energy intensities, combined with highly precise sample positioning, a new category of EDXRF, called microbeam XRF, enables the measurement of metal films found in today’s interconnects and under-bump metallurgies.
Measurement Method and Instrumentation
In microbeam XRF, an energy beam generated by an X-ray tube interacts with the electrons of the inner shell of the elements in the measured material (Figure 1). When the inner shell electron is ejected, it is replaced by an electron from the outer atomic shell. The energy difference of this jump from outer to inner shell is emitted as characteristic fluorescence radiation. The number of X-ray energies that an element can emit depends on the atomic number Z. Elements with a higher atomic number have more electron shells and therefore exhibit a greater variety of transitions, represented in a line series (Figure 2). Transitions from the N-, M-, L-shells down to the K-shell are denoted as the K-series; transitions from O-, N-, M-shells down to the L-shell are labeled as the L-series; finally, transitions from the O- or N-shell build the M-line series. Elements from beryllium to potassium (Z 4-19) emit only K-lines; elements from calcium to barium (Z 20-56) emit K- and L-lines, and elements from lanthanum to plutonium (Z 57-94) emit K-, L-, and M-lines.
Figure 1. Source sample interaction: X-ray fluorescence process.
The basic design of a microbeam X-ray fluorescence metrology system incorporates an X-ray tube that generates the primary X-ray beam to excite the sample; collimating optics that focus the size of the X-ray beam to as small as 20 µm; high-precision sample positioning; X-ray detection, processing, and analyzing electronics; and algorithmic tools for determining the thickness and composition of metal layers ranging from nanometers to 10s of microns in a single stack.
Figure 2. Intra-atom electron transitions responsible for X-ray spectral lines of analytical interest.
Any of these features’ configurations will have an impact on the overall analytical performance of the instrument. One critical point, for example, is the collimation of the X-ray beam. In comparison to conventional pin-hole aperture optics, polycapillary collimating optics achieve three orders of magnitude higher X-ray flux density. This translates into a much higher detection sensitivity and shorter measurement time.
Figure 3. Thickness range and sensitivity for Sn layer based on fluorescence line selection.
The reliable detection of the fluorescence lines of a complex multi-layer system, i.e. gold, palladium, nickel, and copper (Au/Pd/Ni/Cu) on silicon, requires the energy resolution that only electrically cooled solid state detectors such as Si(Li) or silicon drift detectors (SDD) provide.
The fluorescence line selected for the analysis can impact sensitivity performance. Tin (Sn), for example, allows selecting either the K- or L-line for data evaluation. Figure 3 shows a calculation of the sensitivity and maximum measurable thickness range, based on the selected line. The relative intensity change is defined as:
With I (Sn)d : measured intensity of a Sn layer with thickness d
I (Sn)d : measured intensity for “infinite” thick Sn layer.
If the thickness range of interest is between 100 nm and 4 µm, the L-line provides better sensitivity compared to the K-line. The K-line selection has an advantage if the Sn layer is thicker than 4 µm because it allows a broader measurable thickness range. In general, L- and M-lines provide a much better sensitivity for very thin coating in the nanometer range. The trade-off is that those lines are not as suitable for thicker layers.
Microbeam XRF Applications
Under-bump metallization (UBM) layers build the critical interface between the metal pad of the chip and the solder bump used for the flip chip interconnect to the substrate. One popular UBM application used for lead-free solder bumps in combination with flip chip packaging is Au/NiV/Ti/Si.
Figure 4. Au/NiV/Ti/Si X-ray fluorescence spectrum. Measurement conditions: Mo target tube, 47kV, 1mA, polycapillary optics with 75-µm beam size, electically cooled silicon drift detector with 175eV resolution.
Figure 4 shows the spectrum of a sample collect by a microbeam XRF system* for about 100-nm Au on 450 nm NiV on 65-nm Ti. The V concentration is about 9%. Due to the strong overlap of the Ti K beta line with the V K alpha line, for the analysis of the NiV layer, the V K beta line is selected. This example shows how essential high detector resolution is to separate the information from the sample.
Based on ten repetitive measurements at 30-sec. measurement time, the following relative standard deviation ranges can be expected: Au between 1.5 and 3%, NiV 0.5-1%, and Ti 1-2%. An improvement in repeatability can be achieved if the measurement time is extended. At a measurement time of 60 sec., precision improves for all three layers: Au <1%, NiV <0.5%, and Ti between 1-1.3%.
Besides measuring layer thicknesses, the V-composition in the NiV layer also can be determined. At 30-sec. measurements the relative standard deviation will be <5%. For 60-sec. measurements this value can be improved to <2.5%.
Pre-plated frames (PPF) used in microchip packaging have many advantages over conventional solder-coated leadframes. The pre-plating process was introduced to eliminate the solder coating process in manufacturing to reduce semiconductor assembly costs. It also has been successfully proven for the lead-free soldering process. The typical layer system for PPF applications consists of a flash Au top layer, followed by a flash Pd layer and a Ni layer in the 0.5-1 µm thickness range. Recently introduced processes replace the Au top flash layer by a flash Au-Pd or Au-Ag alloy layer.
For a repeatability and reproducibility check, a gage repeatability and reproducibility (GR&R) test was performed to analyze operator and equipment variation. Three different operators measured a leadframe sample at ten different spots. This procedure was repeated by each operator three times over a time period of three days. The relative GR&R value for a specified tolerance value of 0.3 is 7.01%, which qualifies the measurement method and instrument as accurate and repeatable for the given tolerance range.
One PPF process** requires thickness and composition analysis of the Au top layer. As the L-lines are used for the analysis for Ag and Pd, a complex peak deconvolution algorithm has to be applied to separate the information. Figure 5 shows the acquired spectrum of a thickness standard with 30.5-nm AuAg, 42-nm Pd and 2225-nm Ni layer thickness. The Au concentration is certified at 46wt%.
Figure 5. AuAg/Pd/Ni/Cu X-ray fluorescence spectrum. Measurement conditions: Cr target tube, 40kV 0.4mA, polycapillary optics with 75-µm beam size, electrically cooled silicon drift detector with 175eV resolution.
Based on five measurements at 30 sec., the absolute standard deviation is in the range of 0.3 nm for Au, 0.15 nm for Pd, and 15 nm for Ni. The Au concentration uncertainty is in the 0.5 wt% absolute.
This industry trend toward thinner coating, more complex layer systems, and smaller features is challenging the traditional EDXRF technique. Microbeam X-ray fluorescence, in combination with latest developments in detector technology and improvements in calculation and correction algorithms, helps to meet these requirements.
Based on the complexity of the measurement tasks, instrument configuration has to be adapted to each particular application to achieve best performance. This includes tube type, detector type, and application setup. With an application-specific configuration, microbeam XRF can achieve the level of accuracy and repeatability that is necessary for the quality and process control requirements emerging today.
*MicroXR Microbeam XRF system
** The Samsung Techwin “upgrade μ-PPF”
Contact the author for a complete list of references.
RYAN NELSON, application scientist, may be contacted at Thermo Electron Corp. 5225 Verona Road, Madison, WI 5377-4495; 608/276-6100; E-mail: Ryan.firstname.lastname@example.org.