IPC-4556 印制板化学镍钯浸金（ENEPIG）规范ENG - 第34页

(2) A second source of potential error lies with the electroless nickel layer . Proportional counter based XRF instruments cannot measure phosphorous content in the Ni layer directly . Therefore it is generally assumed i…

APPENDIX 4

Factors Affecting Measurement Accuracy

of ENEPIG Coatings by XRF

Frank Ferrandino, Technical Manager

Calmetrics Inc.

XRF has become ubiquitous in the plating industry for measuring plated layer thicknesses.

Among its many virtues, the XRF method tends to typically be robust for a wide range of plating applications. However,

the plating industry in response to increasing demand for reliability, continues to develop new coating combinations that can

make the analysis of thickness by XRF more complicated and therefore increasingly subject to error if not calibrated and

used properly. In particular, the use of ENEPIG coatings in the PCB industry, to improve shelf life and solderability, pres-

ent some challenges to the typical XRF instrument used to measure it.

Typical XRF Instruments Used for Plating Thickness Measurements A sizable majority of the XRF instruments used by

PCB manufacturers are good general purpose plating thickness testers equipped with either a tungsten (W) target or molyb-

denum (Mo) target x-ray source and gas ﬁlled proportional detectors. Such products are capable of achieving reasonably

good accuracy when measuring ENEPIG plating thicknesses. On the other hand, it is also easy for measurement accuracy to

be compromised when such XRF instruments do not have certain software functions or are not properly calibrated.

Below is a list of potential sources of errors when measuring ENEPIG coatings with proportional counter XRF systems using

W or Mo x-ray sources:

(1) The PCB epoxy laminate often contains ﬁre retardant compounds of bromine.

The default measurement condition for these types of XRF instruments is to count the number of Au x-rays detected for

the Au L-β line (energy ~ 11.4 keV).

This is because the Au L-α peak overlaps the Cu x-ray peaks which originate from the Cu layer. In XRF analysis, we

try to avoid interferences caused by such overlaps when possible, hence the usual choice of analyzing the Au L-β line.

Normally when plating on Cu or Cu alloy substrates there is no interference with the Au L-β peak. However, in the

case of PCB samples, there is a reasonably good chance that some Br x-ray emission will be detected from the epoxy

substrate. Given the relatively poor energy resolution of proportional counter x-ray detectors, Br, with a K-α energy of

~ 11.9 keV will produce a spectral peak that overlaps or interferes with the usually reliable Au L-β peak. Normally, Br

x-ray intensity will be low since the Br x-rays must pass through the Cu, Ni, Pd, and Au layers to reach the detector

and be counted. These layers provide signiﬁcant reduction in Br intensity due to shielding, resulting in only a small,

minor Br peak intensity.

In the case of immersion Au layers, which only produce a low intensity Au L-β peak, this small Br contribution to the

peak intensity of Au, if not corrected, can be signiﬁcant since the Br contribution to the Au peak intensity is roughly

the same level of magnitude or even greater compared to the Au peak intensity itself.

When measuring immersion Au layers in the 0.05 µm - 0.13 µm [2 - 5 microinch] range uncorrected Br interference can

add anywhere from a few sub-microns or microinches for the gold measurements to as much as tenths of microns or

tens of microinches. Therefore, a Au layer which is truly 0.1 micron [4 microinches] for example, may measure any-

where from 0.15 µin to 0.25 µm [6 µin to 10 µin ] or more, if the Br interference is not accounted for and corrected.

The amount of error will depend primarily on the thickness of the Cu layer (thinner copper results in larger errors in

an Au measurement), the amount of Br compound in the epoxy and the spatial resolution of the x-ray beam and its

position relative to the plated areas being measured.

Fortunately, most if not all the XRF instruments used for plating thickness measurements offer peak deconvolution soft-

ware. This software will allow the instrument to breakdown the composite Au + Br peak into its component parts.

Therefore, it is possible to extract Au L-β peak intensity information independent of the contributing interference from

Br. For accurate measurement of immersion Au on PCB’s, use of a peak deconvolution routine is prudent and in many

cases absolutely vital to maximizing the accuracy of the Au thickness measurement. It should be noted that signiﬁcant

errors in Au thickness measurement due to uncorrected Br interference will contribute to errors in the calculation of

underlying electroless Ni and Pd thicknesses, as well.

January 2013 IPC-4556

(2) A second source of potential error lies with the electroless nickel layer. Proportional counter based XRF instruments

cannot measure phosphorous content in the Ni layer directly. Therefore it is generally assumed in the calibration of the

instrument that the Ni layer contains some constant amount of phosphorous composition, typically about 8%. However,

if the sample does not contain 8% phosphorous in the Ni layer, errors in electroless Ni thickness will occur. As a very

rough rule of thumb, one can expect about a 4% thickness error for each 1% difference in phosphorous content between

the sample under test and the standards used for calibration.

The electroless nickel thickness measurement will be erroneously high if the sample phosphorous content is less than

the calibration standard. The measurement of thickness will be erroneously low if the sample contains more phospho-

rous than the standards used for calibration. Since most platers maintain the phosphorous content at approximately 8%,

the electroless nickel thickness measurement error is normally minor.

Although electroless Ni thickness errors due to phosphorous content variations are normally less than 10%, it is pos-

sible to correct for changes in sample phosphorous content if they are known. Some XRF software does offer such cor-

rections which allow the user to enter the known % P content in the nickel layer and the thickness measurement will

automatically be corrected. Alternatively, suppliers of such XRF instrumentation should be able to provide correction

factors for changes in sample % P which can be entered as ‘‘density’’ factors that can correctly compensate for changes

in sample % P.

(3) Measurement of the immersion Pd layer is subject to two possible errors. The more signiﬁcant potential error is caused

by variations in the x-ray background level in the spectral region where the Pd K-α peak is detected. The Pd K-α peak

is added to this background level. Typically, XRF plating thickness measuring instruments integrate this area to obtain

Pd intensity and relate that intensity to Pd thickness. In the case of immersion Pd where thicknesses are in the order of

0.05 µm - 0.15 µm [2 µin - 6 µin], the spectrum background level is often as intense or more intense than the Pd peak

itself. If the background level is constant, this inﬂuence may be included in the calibration. However, the background

level may, in fact, vary. This potential background level variation is a function mainly of the Cu layer’s thickness below

the Pd.

The background scatter originates mainly from the epoxy substrate. The W or Mo target x-ray sources provide a broad

band of x-ray energies to the sample during the measurement. It is mainly the higher energy x-rays that will penetrate

the thick Cu-clad laminate, scatter back from the epoxy through the Cu layer to be detected and seen in the measured

spectrum. Lower energy x-rays originating from the source, do not have enough energy to penetrate the Cu layer twice

and be detected. Therefore background ‘‘noise’’ is only an issue in the higher end of the x-ray energy spectrum.

It is in this high end of the spectrum that the Pd K-α appears (~ 21.1 keV). The amount of background scatter that

reaches the detector is again a function of the Cu thickness since the scattered x-rays are shielded primarily by this rela-

tively thick layer. As with the Br interference with Au, the background scatter will be a function of the underlying Cu

thickness as well as the spatial resolution of the x-ray beam and the position of the beam relative to the edges of the

sample plated area.

Again, to be prudent and to optimize accuracy of Pd thickness measurements, the potential varying background level

should be compensated for to obtain net Pd intensity information. Most XRF instruments are equipped with background

correction software to deal with this issue and obtain reasonably accurate net Pd intensities. The user should be famil-

iar with how to access this feature if its use is optional, or know if automatic background correction is always used by

their XRF instrument to measure plating thickness. There are a wide variety of background correction schemes that are

available, some of which are more effective in this case than others.

Other approaches to mitigate Pd measurement errors due to varying background levels include the use of primary beam

ﬁlters to attempt to suppress the background level to a magnitude far below that of the Pd K-α peak intensity. This

method can work well when using, for example, a Mo ﬁlter with a larger collimator. Since primary ﬁlters reduce over-

all signal intensity the ﬁlter material and thickness must be selected with an attempt to balance the gains from reduc-

ing the background and the loss due to reduced Pd signal intensity. Typically, primary ﬁlters are effective only when

using larger x-ray beam collimators. Users should consult with the manufacturer of the instrument before attempting to

use a primary ﬁlter.

Less signiﬁcantly for the commonly used XRF instrument is the error bias that may be introduced by the fact that

immersion Pd layers are not pure Pd. Instead, it is common that such layers contain about 2% phosphorous. Like the

electroless nickel measurement, the level of phosphorous in the plating has some effect on the thickness measurement.

IPC-4556 January 2013

In the case of immersion Pd however, we can normally assume this effect is insigniﬁcant for two reasons:

a) The Pd thickness is very low where the thickness measurement is not very dependent on the layer purity.

b) The Pd x-ray energy is almost 3X higher than Ni so that the absorption of Pd x-rays within the layer by phosphorous

is much smaller than it is for nickel. This combined with (a) above will result in errors of about 5% assuming the cali-

bration standards are pure Pd (which is typically true).

Much of this error originates with the density difference between immersion Pd and pure Pd used for calibration standards.

By requesting standards with the Pd certiﬁed to a density of 11.6 g/cc instead of the conventional 12.0 g/cc, the bias error

is reduced to less than 2%.

Long term stability of XRF calibrations using proportional counter detectors for ENEPIG applications is often compromised

primarily by the use of the peak deconvolution method for the Au. This method places higher demands on the detector elec-

tronics stability, making the measurement more sensitive to instrument drift. To avoid additional error due to this effect, one

should check calibration standards frequently, at least twice a day, to assure that the instrument is still within calibration

tolerances. Use supplier provided drift correction features to correct for drift if standard measurements are outside their tol-

erances. If drift correction fails to restore the calibration to within tolerance, the instrument should be recalibrated.

Summary for Common Plating Thickness Measuring XRF Instruments The predominantly used XRF type in the PCB

industry to measure ENEPIG coatings should be calibrated with the following considerations in mind:

1. Use of peak deconvolution for the Au layer measurement.

2. Use of an appropriate background subtraction model for the Pd layer.

3. Use of appropriate corrections for differences in phosphorous content between calibration standards and test samples for

the nickel layer and, less importantly, the Pd layer.

4. Use of calibration standards with similar thicknesses to the speciﬁed ENEPIG thicknesses which are to be measured. If

possible, request Pd thickness certiﬁcation to the level of % P expected in the Pd layer and request electroless nickel lay-

er(s) with % P content within 1% of the expected % P content for the samples which are to be measured.

5. It is also helpful if the standards are plated over a Cu laminated epoxy substrate containing Br if the samples to be tested

will also contain Br.

By requesting standards with these parameters, one can most closely approximate the inter- element matrix effects and spec-

tral effects discussed above, resulting in an optimized calibration which minimizes thickness measurement error. Use cali-

brations standards to check and if necessary correct for long term stability problems. Calibrations standards are used to check

that the calibration is still within tolerance from day-to-day, to assure that long term stability limitations of the instrument

do not compromise results.

Alternative XRF Instruments & Configurations: Advantages and Disadvantages In recent years, more capable XRF

instruments have been introduced for plating thickness measurement. The most signiﬁcant advance involves the use of solid

state detectors, typically silicon PIN diodes and most recently silicon drift detectors (SDD). These detectors offer a major

advantage over the common proportional counter used in XRF’s employed in the plating industry in terms of energy reso-

lution. It is possible to achieve energy resolutions of about 5X better than the proportional counter with these detectors. The

advantage instruments with these detectors offer is primarily in the Au layer measurement. In the case of Au, the Au L-β

peak and the Br K-α peaks can be better resolved, reducing overlap interference.

Furthermore, the Au L-α peak is fully resolved from any overlap with the Cu K peaks.

As a result, the Au L-α peak may be used without resorting to deconvolution techniques. While peak deconvolution routines

needed for proportional counters minimize error due to peak interference from other elements like Br, they include their own

inherent potential errors, smaller but nevertheless present. Peak deconvolution is especially difficult to employ well when

plating thickness becomes very thin. For Au layers less than 3 microinches, peak deconvolution methods can struggle to

achieve good accuracy when Br is present.

In addition, the nature of the peak deconvolution method requires excellent electronics stability to maintain measurement

accuracy without frequent recalibration. In some cases, instrument design does not offer the needed stability to avoid fre-

quent drift correction and recalibration.

Use of solid state detector based XRF systems eliminates the need for peak deconvolution methods with respect to Au thick-

ness measurements and permits Au measurements of layers less than 1 microinch. Therefore, solid state detector XRF sys-

tems tend to offer much better long term measurement stability and are much less reliant on operator know-how and vigi-

lance to achieve accurate Au results.

January 2013 IPC-4556