Understanding, Measuring & Tracking Phosphorous in Electroless Nickel Bath

Estimated reading time: 16 minutes

Editor's Note: This column originally appeared in the December 2011 issue of The PCB Magazine.

Electroless nickel immersion gold (ENIG) has been used as a final finish for decades now, and is one of the most widely used surface finishes today, with approximately 54% of PCBs employing it. Recently, IPC reported that 2010 sales for final finishes was $285 million, with $155 million of that being ENIG, or electroless gold [1]. The ENIG coating has several outstanding attributes, including excellent solderability and extended shelf life before assembly, and is well known by PCB manufacturers and assemblers alike.

An out-of-control ENIG process can wreak havoc on yields, cause re-work and down time and lead to costly field failures. Understanding, measuring and tracking phosphorous in the electroless nickel deposit is one potential way of keeping a finger on the pulse of the process.

For phosphorous content the IPC-4552 ENIG Specification references the ASTM International standard for electroless nickel-phosphorous coating on metal, which is a great place to start. The ASTM International specification states the following [2]:

Paragraph 1.5: The coatings have multifunctional properties, such as hardness, heat hardenability, abrasion, wear and corrosion resistance, magnetics, electrical conductivity provided diffusion barrier and solderability. They are also used for the salvage of worn or mismachined parts.

Paragraph 1.6: The low phosphorus (2 to 4% P) coatings are microcrystalline and possess high as-plated hardness (620 to 750 HK 100).These coatings are used in applications requiring abrasion and wear resistance.

Paragraph 1.7: Lower phosphorus deposits in the range of 1 to 3% phosphorus are also microcrystalline. These coatings are used in electronic applications providing solderability, bondability, increased electrical conductivity and resistance to strong alkali solutions.

Paragraph 1.8: The medium phosphorous coatings (5 to 9% P) are most widely used to meet the general purpose requirements of wear and corrosion resistance.

Paragraph 1.9: The high phosphorous (more than 10% P) coatings have superior salt-spray and acid resistance in a wide range of applications. They are used on beryllium and titanium parts for low stress properties. Coatings with phosphorus contents greater than 11.2% P are not considered to be ferromagnetic.

As the phosphorous content increases, wear and corrosion resistance increase. This begs the question, where should the phosphorous content be for optimal corrosion resistance and solder-joint strength? A quick review of the literature reveals:

Dong-Jun Lee et al. found increased solder joint strength with 10% bulk phosphorous as opposed to 6% bulk phosphorous in the electroless nickel deposit [3].

F.D. Bruce Houghton concluded that high phosphorous is not a problem, as thought when starting Round 1 testing. Low phosphorous can be a problem and it can also affect solderability [4].

Mei, Zequin et al. concluded that the root cause for the brittle interfacial fracture is not the high phosphorus content, as least in the concentration range of 6 to 12 wt % [5].

There is enough empirical evidence to state that the percent phosphorous should be somewhere between 6 to 12%. Staying within the 6 to 10% range is in line with the ASTM definition of a medium phosphorous electroless nickel deposit (mid-phos).

But, what influences the phosphorous content in the electroless nickel bath?  Let’s first look at some empirical test data [6].

Figure 1: As the pH increases in the electroless nickel bath, the phosphorous content decreases.

Figure 2: As the nickel ions increase in the electroless nickel bath, the phosphorous content decreases.

Figure 3: As the hypophosphite ions increase in the electroless nickel bath, the phosphorous content increases.

These graphs are not absolutes, simply because there are many other influences in the phosphorous deposition reaction. For example, Class I stabilizers (S, Se, Te) and Class II stabilizers (compounds containing oxygen such as AsO2-1, IO3-1, etc.) function by adsorbing on the catalytic surface where they prevent spontaneous decomposition of the bath. When too many of these stabilizers are adsorbed on the catalytic nickel surface they inhibit reaction 3, allowing reaction 1 to proceed while stifling reaction 2, resulting in a lower phosphorus content in the deposit. We’ll discuss stoichiometry a bit later.

As solution agitation increases, the rate of diffusion of Class I and Class II stabilizers to the surface increases. Each adsorbed stabilizer molecule or ion reduces the number of catalytic sites available for the dehydrogenation of hypophosphite resulting in a lower phosphorous in the deposit. This phenomenon is most likely to occur at sharp edges or corners. When taken to an extreme condition, the nickel deposit becomes thin; this is known as edge effect.

Fluids can flow steadily, or be turbulent. In steady flow, the fluid passing a given point maintains a steady velocity. For turbulent flow, the speed and/or the direction of the flow varies, which is what we have in an electroless nickel bath due to stroke agitation and vibration.

Bath loading can influence the phosphorous too. Bath loading is referred to as the ratio of total exposed surface area being plated to the volume of solution in the tank. If the workload-to-solution volume ratio is ≤ 0.2 sq ft/gal, the potential for stabilizer adsorption is greater, especially on sharp edges or corners.

Additionally, as the electroless nickel bath ages, the phosphorous in the deposit increases. This is due to the alkali metal cations (Na+, K+), or the ammonium ion (NH4+) build-up in the bath.  The buffers in the electroless nickel solution associate with these cations preferentially over the H+ ions, rendering more H+ ions available at the diffusion layer. This results in increased phosphorous in the deposit and a retarded nickel plating rate. The age of the electroless nickel bath is tracked by MTO (metal turn over), e.g., if we run the electroless nickel bath at 5 g/l nickel, once we replenish the bath with 5 g/l nickel we count that as 1 MTO. We can also titrate the reaction by-product, orthophosphite, and determine the electroless nickel bath MTO.   

For some applications within the integrated circuits (IC) and micro-processing units (MPU) industries it’s desirable to control the phosphorous within a tight range. This can be done by manipulating the above items discussed, and monitoring and controlling the Class I and/or Class II stabilizers [7].

Let’s now focus our attention on the primary stoichiometry in the electroless nickel bath. Note that there are several secondary and tertiary reactions which are not discussed here.

Reaction 1:

Hydrolyzed nickel species adsorbed at the catalytic surface reacts with hypophosphite depositing metallic nickel and producing dihydrogen phosphate and a proton.

Reaction 2:

Catalytic nickel surface reacts with hypophosphite depositing phosphorus and producing adsorbed nickel hydroxide and a hydroxide ion.

Reaction 3:

Hydrolyzed nickel species reacts with water producing nickel hydroxide and a proton.

Reaction 1 and 3 compete:

Reaction 4:

Hypophosphite reacts with a proton, depositing phosphorus and producing dihydrogen phosphate, water and hydrogen. This is due to the cation build up in the electroless nickel bath as it ages from 0-5 MTO.

As long as there is a constant supply of adsorbed NiOH species on the catalytic surface and reaction 1 takes place, the deposition of phosphorus by reaction 2 cannot occur.

The metallic Nicat surface must be available for a direct interaction with hypophosphite to deposit phosphorus. When reaction 3occurs, the metallic catalytic nickel surface that was previously covered by adsorbed NiOH species is now free to interact with hypophosphite.

Any periodicity between reactions 1 and 3 will lead to a distinct layered structure within the electroless nickel itself. 

Figure 4: The lamellar structure within the electroless nickel deposit (the electroless nickel is sandwiched between the upper x-section potting compound and the lower electroplated copper). Note that the coupon was etched with a modified Lepito’s etch solution. 

Now that we have an understanding of how the phosphorous varies in the bulk electroless nickel deposit, how can we measure it accurately?

Assume we have two measuring instruments; one is quite accurate but is both expensive to use and slow (scanning electron microscope/energy dispersive X-ray, or SEM/EDS), and the other is fast and less expensive to use, but is also less accurate (x-ray fluorescence, or XRF). Theoretically, the accuracy of the EDS is ±5%, whereas the XRF may be as high at ±20% [8]. As a side note, strict protocol must be used with XRF to measure phosphorous in the electroless nickel [9].

If the measurements obtained from the two devices are highly correlated, then the measurement that would have been made using the expensive measuring device should be predicted fairly well from the measurement that is actually obtained using the less expensive device. This is called calibration or inverse regression [10]. Why not just use linear regression? Because with linear regression it is assumed that the X variable is able to be held constant (no variance) and with measurement equipment we have variance in all measurements; calibration regression takes this into account.

Table 1: The percent phosphorous in several different samples of an electroless nickel deposit as measured by SEM/EDS and XRF.

When we run simple correlation analysis we can see that the correlation between the XRF and the EDS = 0.985. With such high correlation we can employ calibration regression. As a general rule of thumb, for linear regression, we should have a minimum of five degrees of freedom (five more rows of data than model terms) to measure the amount of variation in the response left unexplained by the model. Others have suggested ~10 responses for each parameter in the model to avoid over-fitting [11]. For this example, seven samples are used.

When we regress the XRF on the EDS we derive a simple linear regression (equation 1). Note: coefficients are rounded off.

Equation 1:

XRF = 1.44 + 0.875*EDS

Through simple algebra we derive the calibration (inverse) regression (equation 2). Note: coefficients are rounded off.

Equation 2:

EDS = -1.65 + 1.14*XRF

With equation 2, we can develop 95% intervals for both a single point estimate, the prediction interval (PI), and a mean estimate--the confidence interval (CI)--for our EDS readings. Simply stated, the intervals give us a range within which a single value (PI), or the mean (CI) may be expected to fall, with 95% confidence.

Now we can simply insert this equation into an Excel file and achieve a predicted EDS reading from the XRF reading instantaneously.

Table 2: The equation inserted into an Excel spreadsheet with both the prediction and confidence intervals.

How do we track the phosphorus content once we have measured it? The first step is to make sure we understand the physical system of how the phosphorous is deposited in the electroless nickel. We know that bath components, loading, agitation and vibration influence the phosphorous content. And as the electroless nickel bath ages, the phosphorous content in the deposit increases (reaction 4). 

The second step is to decide the primary purpose of the control chart, and how we will collect the data. Here are some options:

1. If our primary purpose is to make a decision about the acceptance, or all boards produced since the last sample, then we could sample at specified MTO intervals of the electroless nickel bath life, say 0, 1, 2, 3, 4, 5, or 0, 2½, 5, which would give us subgroups of either 5 or 3 respectively. This type of subgrouping is a random sample of all process output over the sampling interval (the electroless nickel bath life).
2. If our primary purpose is to detect process shifts, then we could sample consecutive boards produced at specified MTO intervals of the electroless nickel bath life, say three samples at each MTO of 0, 2½, 5, which would give us subgroups of 3. This type of subgrouping gives us a snapshot of the process at each point in time where a sample is collected.
3. If our primary purpose is to plot and track the phosphorous on a SPC chart, then we could sample randomly at different MTOs. This type of subgrouping (n = 1) is useful if we are taking multiple measurements on the same part.

Note: Regardless of which method is used, there will always be some trade-offs.

The third step is to understand how the physical system and sampling scheme will influence the SPC charts. As the phosphorous content increases as the electroless nickel bath ages, we have potential trend data, and we must be careful. It is well known that trend data is deleterious to SPC charts [12 & 13]. 

If we collect data only at the beginning or near the end-of-bath life of the electroless nickel, we are likely to have non-normal data. It is well known that non-normal data can significantly alter the tail probabilities of the control chart, generating significantly more false alarm rates on either the lower control limit (LCL) or the upper control limit (UCL), while simultaneously significantly lowering the ability of the opposite control limit to even detect an out-of-control condition [12, 13 & 14]. The debate rages on over the validity and consequences of putting non-normal data on a control chart that is based on normality. Suffice it to say I am in the camp that believes non-normal data does not belong on a normal control chart. 

Sampling with subgroups with n>1 takes time and the process might have drifted out of control before we have even obtained our first subgroup!  SPC charts based on subgroups of n=1 are not as sensitive at detecting mean shifts, and the normality assumption is far more important since there is no central-limit theorem-type effect with individual observations; this may require a data transformation.

The fourth step is to choose the appropriate SPC chart. Let’s explore option 1 from above. We know that it will take time to gather enough subgroups to construct the SPC chart and there is a chance that the process could drift in and out of control before we have our SPC chart constructed and have begun monitoring the process. How can we work around this shortcoming?

A useful statistical tool for this is a tolerance interval. Tolerance intervals are intervals that cover a proportion (p) of the overall population with a given confidence level (1-α).

Tolerance intervals differ in an important way from confidence intervals (CI); tolerance intervals are constructed to contain a specified proportion of the population of individual observations within a given confidence level. A confidence interval is an interval of plausible values for a population parameter, such as the population mean (mu), with a specified confidence level equal to the degree of plausibility.

The endpoints of a tolerance interval are called tolerance limits. The tolerance limits may be compared with the specification limits to judge the process behavior (e.g., are most of the values falling inside the specification limits?).  Normality of the data is a strict requirement to construct tolerance intervals because our focal point is the tail areas of the distribution.

If we collect three samples at each MTO from 0 to 5, we should expect to see a normal distribution. If we construct a tolerance interval with 95% confidence covering 95% of the population, then at any time over the electroless nickel bath life we can sample and test for phosphorous and we should be within our interval 95% of the time. In other words, with the tolerance interval we can state that we are 95% confident that 95% of the population (of the phosphorous % by weight in the electroless nickel deposit) is between x1 and x2.

Let’s take some actual data and see how it applies. 

Figure 5: Phosphorous readings from an electroless nickel production bath between 0 to 5 MTO (three readings at each MTO). 

We can check the normality of the data by graphing it on a probability plot. 

Figure 6: The normal probability plot of the phosphorous readings.

The p-value is significantly above 0.05 and the data fits the diagonal line well (the diagonal line represents a perfect fit), so we can conclude that there is simply not enough evidence to reject the normality requirements and we can proceed with the calculation of the tolerance interval. That calculated tolerance interval is:

Tolerance Interval Type: Two-sidedConfidence Level: 95%Percent of Population in Interval: 95%N: 18Mean: 8.4 StDev: 0.73    Tolerance Interval: 6.3 to 10.4

We match up very well to the ASTM definition of a mid-phosphorous electroless nickel bath, and are well within the empirical range determined for excellent solder joint integrity [3, 4 & 5]. Now we can randomly sample at any time, measure the phosphorous and quickly determine the pulse of the process. Any value outside of the tolerance interval should be investigated.

For our process, we are not only concerned with acceptance of all boards produced since the last sample, but we are also concerned with both the between- and within-variance of the nickel baths. That is, every time we make up a new nickel bath we would like to measure the between-variance of these baths, along with the within-variance of each bath. The control chart for this application is the Between-Within Chart [13, 15].

We’ll sample at 0, 2½ and 5 MTO and use these as a subgroup of three. Strict XFR measurement protocol must be used, which cannot be overemphasized [9]. Data collected is shown below:

Figure 7: The phosphorous readings from 15 nickel baths at 0, 2½ and 5 MTO.

Now we can plot the data on the between-within chart and ascertain the stability of the process in regards to the phosphorus content in the electroless nickel bath.

Figure 8: The Between-Within control chart.

We can apply numerous statistical tests to the control chart to determine if there are any detectable special causes of variation [12, 13]. Minitab® software allows the user to apply up to eight specific tests for special causes of variation [15]. These statistical tests look for special causes (non-random variation), with each individual test looking for a specific pattern in the data, which if detected should be investigated.

The subgroup mean chart (top) has no violations of the eight special causes of variation tests; the between-bath moving range chart (middle) and the within- bath moving range chart (bottom) have no violations. Furthermore, the between-within charts show good stability. Common cause variation appears to be the only source of variation within the plotted data; hence, we can conclude the process is in control. 

Electroless nickel baths are complex process systems with many inputs that influence the phosphorous content in the nickel deposit, that is, Y = f(x1, x2…xn), where n = numerous factors and interactions.  

Electroless nickel baths require good process control for proper operation and consistency from bath to bath. Understanding, measuring and tracking the phosphorous in the electroless nickel deposit gives us a good overall indication of how well the process is being controlled, and good control is directly related to reliability.

References:

1. IPC Statistical Program for the Worldwide Process Consumables Industries, Results for fourth quarter 2010.2. ASTM International B733-04 (2009) Standard Specification for Autocatalytic (Electroless) Nickel-Phosphorous Coating on Metal.3. Dong-Jun Lee, Hyo S. Lee, “Major factors to the solder joint strength of ENIG layer in FC BGA package.” Microelectronics Reliability 46 (2006) 1119-1127.4. F.D. Bruce Houghton, “Solving the ENIG Black Pad Problem: An ITRI Report on Round 2.” Proceedings of IPC Works ’99, October 23-28, 1999, Minneapolis, MN, pp. S-04-3-1 to S-04-3-9.5. Mei, Zequin, et. al. “The effect of Electroless Ni/Immersion Au Plating Parameters on PBGA Solder Joint Attachment Reliability.” IPC National Conference Proceedings: A summit on PWB Surface Finishes and Solderability, September 22-23, 1998, Austin, Texas, pp. 19-42.6. Mallory, G., Hajdu, J. et al., (1979). Electroless Plating Fundamentals and Applications, Noyes Publications/William Andrew Publishing, LLC, Norwich, NY.7. OMG internal testing and analytical methods development.8. Conversations with Lisa Gamza, OMG Senior Materials Scientist.9. IPC-4552 ENIG Specification.10. Ryan, T. (2009). Modern Regression Methods, 2nd Edition, Wiley & Sons Inc., Hoboken, NJ.11. Launsby, R., and Schmidt, S. (2005).  Understanding Industrial Designed Experiments, 4th Edition, Air Academy Press, Colorado Springs, CO.12. Ryan, T. (2000).  Statistical Methods for Quality Improvement, 2nd Edition, Wiley & Sons Inc., Hoboken, NJ.13. Montgomery, D. (2009).  Introduction to Statistical Quality Control, 6th Edition, Wiley & Sons Inc., Hoboken, NJ.14. Yourstone, S.  A. and W. J. Zimmer, “Non-normality and the Design of Control Charts for Averages.” Decision Sciences 23 (1992) 1099-1113.15. Minitab ® 16 Software.

Patrick Valentine is the North American PCB and EP&F business manager for OMG Electronic Chemicals and has been with the company since 1991. He holds a Master’s degree in business from Regent University, earned his Six Sigma master black belt certification from Arizona State University, and is an ASQ certified Six Sigma black belt.  Contact Valentine at patrick.valentine@omgi.com.