The back-end process: Step 2 - Wafer bumping Low-alpha lead considerations
02/01/2002
BY KRISHNA K. NAIR, POOJA JINDAL, AND GLENN A. RINNE
As a result of decades of manufacturing flip chip packages, there exists a large database of reliability information about lead solder bumps. IBM was the first to warn us about the elusive issues of alpha particle emission from the solder. The "spontaneous" soft errors that occurred in devices led to several months of investigation. The problem was traced to an air jet used by a nitric acid vendor to clean bottles. The air jet contained a small amount of alpha-emitting material that contaminated a very small number of bottles.
Soft errors result when an alpha particle penetrates a device, creating a trail of electron-hole pairs. While this process seldom causes permanent damage, it changes the state of the device. Alpha particles are most often produced by the radioactive decay of various atomic species in the solder. A simplified approach would identify lead (Pb) as the source of the problem, but the problem is multi-faceted. Accordingly, an effective low-alpha strategy should include more than a single approach.
Mechanism of Radioactive Decay of Lead
Figure 1 shows the sequence of events involved in the radioactive decay of lead. Naturally occurring lead contains several isotopes, including 210Pb. The 210Pb decays with a half-life of 22 years to 210Bi and then quickly to 210Po, emitting a beta particle with each transition. The 210Po decays with a half-life of 138 days to 206Pb, emitting an energetic alpha particle in the process. The 210Po can be removed from the solder but the decay of 210Pb keeps creating new 210Po. The two processes eventually reach equilibrium, which in practical terms means that there will be an appreciable amount of 210Po causing alpha emission throughout the working life of the device. The maximum level of soft errors that can be tolerated for today's integrated circuits has been translated to a requirement of 0.01 alphas/hr/cm2 (or cph/cm2) to 0.002 cph/cm2 for more stringent applications, such as dynamic logic. Naturally occurring lead has an alpha particle emission rate in the range of 1 to 30 cph/cm2, which is mostly because of polonium. Thus, the concentration of 210Po in the lead must be reduced by a factor of 100 to 15,000.
Availability of Low-alpha Lead
Low-alpha lead can be found in lead ore that was refined many years ago, because the 210Pb in the metal is no longer being replenished by the decay of uranium isotopes. Such "antique lead" can be found in shipwrecks and other artifacts that are more than 200 years old.
 Figure 1. Radioactive decay of 210Pb. |
The alternative to antique lead is to selectively remove the 210Po and 210Pb from natural lead. 210Po can be removed chemically, whereas 210Pb requires special techniques, such as laser isotope separation. To achieve an alpha emission rate of 0.002 cph/cm2, the level of 210Pb needs to be about one part in 1017. The laser separation process is, however, expensive and limited in capacity.
Currently, it only produces enough low-alpha lead for approximately 120,000 wafers per month.
A Strategy for Low-alpha Bumping
Bumping on wafers is currently done by three main techniques: electroplating, screen printing and vacuum evaporation. Irrespective of the bumping technique, implementation of a low-alpha process involves the same elements: (a) control of the starting materials, (b) control of the process, and (c) control of the monitoring techniques.
Control of Starting Materials
For evaporation of Sn/Pb bumps, the starting material comes in the form of solder slugs that are tested for alpha emission. For screen printing and electroplating, source materials of the solder alloy are in the form of powder. Vendors usually certify the powders to have a specified maximum alpha count, such as 0.01 cph/cm2. It is critical to make sure that these materials have reached secular equilibrium, the point at which the rate of decay of Po has begun to exceed the rate at which it is created in the sample. If this has not happened, the alpha activity is likely to increase with time. Testing should be done on these materials at a certain frequency - every six months for example - to ensure there is no increase in alpha activity.
Control of the Process
In the case of evaporated Sn/Pb metallurgy, because the metals are not mixed with any other material, chances of contamination by alpha emitters are small. However, because tooling parts reach high temperatures in high vacuum, trace impurities present in the materials used in the tooling (such as lead impurity in alumina insulators) may evaporate and become entrapped in the solder films. Such exposure can be minimized by selecting proper materials and design for this environment. Frequent monitoring of blanket deposited films provides additional process control.
 Figure 2. Typical system for counting alpha particles. |
In the case of electroplating of Sn/Pb solders, a number of chemical processes are required to form the proper plating chemistry for each solder. Typically, low-alpha lead and tin monoxides are converted to lead methane sulfonic acid (Pb-MSA) and tin methane sulfonic acid (Sn-MSA), which are used to mix the plating baths. Although the chemical processes do not affect the behavior of the heavy isotopes, appropriate controls are needed. The plating baths are tested to ensure that no alpha-producing contaminants have entered the plating system. Testing is done on a sample plated onto a steel substrate in the respective plating bath. The Sn/Pb anodes used in electroplating are also tested for alpha emission.
An indispensable process control vehicle consists of blanket solder films deposited by the respective process. In evaporation and electroplating processes, it is fairly easy to prepare such blanket films, whereas in screen printing it is a challenge. Large areas and long testing times are required to determine the alpha emission density to acceptable confidence levels. (See sidebar: "Sampling Statistics for Low Alpha Control.") Testing on bumps does not offer the same confidence level because of the greatly reduced surface area. There is no known method to make a blanket film with paste, so there is no satisfactory means of testing the paste.
The question of homogeneity is another important consideration in process control. In electroplating, the baths are constantly recirculated, mixed and agitated to ensure uniformity. Any isotope atoms are uniformly distributed through the bath, which will guarantee meaningful alpha counts. However, in screen printing, homogeneity may be difficult to maintain. Another possible problem is that the powders may come from different sources (especially when different metals are combined) that are not necessarily homogenized while mixing the paste.
Control of Monitoring Techniques
There are many types of alpha detection and counting equipment, but the ionization detector is commonly used. In a typical counting system, there is a sample chamber and a detection chamber (Figure 2). The detection chamber contains an ionizing gas composed of methane and argon. The detection chamber is also provided with an anode and a cathode. The two chambers are separated by a thin Mylar film (typically 2 µm in thickness), which confines the ionizing gas to the detection chamber. Any alpha particle that enters the detection chamber causes an instant ionization in the gas. This causes an instant current between the anode and cathode, which is registered as an event of alpha emission. To test for 0.01 cph/cm2, the sample must have an area of the order of 1,000 cm2, and counting should be done for about 24 hours (see sidebar).
Because of the nature of the testing and the costs involved, testing is often done by specialized testing services. Understanding and interpreting the testing results is critical for the bumping vendor as well as the customer. A number of factors can affect the testing results. For example, the surface of the blanket-deposited solder film is usually very rough, the degree of which depends on the process. Because the emission of alpha particles is a surface phenomenon, the actual surface area needs to be estimated and entered into the calculation. If such correction is not employed, the measured emission density could be larger than the actual emission density by a factor of two or more.
 Table 1. There is a significant difference between testing on blanket films and bumped wafers because of the difference in actual solder area. |
Similarly, when testing is done on silicon wafers, some overlapping needs to be done to maximize the tested area and to accommodate the geometry of the wafers in the limited area of the chamber. The overlapped area needs to be determined accurately to obtain the effective area. The difference in geometries between the wafers and the sample tray also causes some area of the tray to be exposed, which is undesirable because of possible cross-contamination from previous samples. This problem could be significant when using vendor services where they may be testing high-alpha emitting samples. For the levels required for today's electronics industry, even a milligram of ordinary lead (like a smear) could produce erroneous high counts. Also, because the sample sits very close to the Mylar film, the Mylar film can get contaminated by high-alpha samples. This factor is often overlooked.
Does "Lead-free" Mean "Alpha-free"?
null
Several lead-free solder alternatives are being developed in an effort to solve alpha emission issues and to address the environmental issues. While lead-free solders address the environmental concerns, they require careful scrutiny regarding the alpha emission issues. Typical alloys considered for lead-free are Sn-Ag and Sn-Ag-Cu. There are also alloy systems containing Bi and Sb. Sn, Ag, Bi and Sb could all be contaminated with small amounts of Pb and Po that could cause alpha emission. In addition, wherever recycled tin is used, there is added risk of lead contamination.
However, the same rigorous process could be instituted for lead-free solders as for present solders to result in low alpha emission.
AP
Krishna K. Nair, principal technologist, Pooja Jindal, process engineer, and Glenn A. Rinne, vice president of research and development, can be contacted at Unitive Inc., P.O. Box 14584, Research Triangle Park, NC 27709-4585; 919-941-0606; Fax: 919-941-5097; E-mail: [email protected], [email protected] and [email protected].
Sampling Statistics for Low-alpha Control
null
Because emission of an alpha particle is a random event, prediction of its effects on a device is described statistically. The sample size determines the confidence level to which the measured level can be predicted. If N is the sample size (number of alpha emissions counted), the error in measure-ment E is given by E = 1 / N1/2, and the per- cent confidence level is obtained (or calcu-lated) as (1-E) x 100 percent.
For example, if a sample of area 1 cm2 is counted for 500 hours (approximately 3 weeks) and five counts are registered, we may predict the alpha emission density to be 0.01 cph/cm2, but only with a confi-dence level of 55 percent. To predict the alpha activity at any appreciable confi-dence level, the sample size should be increased by increasing the sample area or the counting time or both. Typically, blanket films of an area approximately equal to 1,000 cm2 are used and counting is performed for 24 hours. If bumped wafers are used for testing, substantially less area would be obtained and the results would be much less reliable. Table 1 shows a hypothetical comparison of the two cases. If blanket films (with an alpha emission density of 0.01 cph/cm2) on wafers of total area 1000 cm2 were tested for 24 hours, we would obtain a count of 240, typically. Then we could predict the alpha emission density to be 0.01 cph/cm2 with a 93.5 percent confidence level. However, if bumps on wafers of the same size were tested, we would obtain a count of 47, typically. Then, the predicted alpha emission density would still be 0.01 cph/cm2, but with only a confidence level of 85.4 percent.
null