Copper Cleaning Made Easy Oxidation reduction for wafer bumping
By Terence Collier,
Clean metal surfaces are critical to bumping and soldering processes. Poor cleaning can result in delamination of plated metal layers and lackluster wetting during soldering. An etching process exists that attacks only oxides and leaves base metal intact, or with a protective layer that prevents oxide regrowth.
Bumping begins with either aluminum or copper features. In a typical process, aluminum pads are converted to a solderable, bump-ready layer by adding an under bump metallization (UBM) layer. For better reliability, it’s important to add a UBM layer, rather than nickel-gold to bare copper, to eliminate the impact of copper-tin intermetallic compounds, copper consumption, and poor wetting due to copper oxidation. Nickel serves as both a barrier and adhesion layer, and can improve long term reliabilty particularly with multiple reflows.
Unfortunately, most metals oxidize, and most etchants/cleaners readily etch both base metal and metal oxide. The result is a rough and pitted surface after the etching process (defects and variation in the oxide can lead to etching of some regions faster than adjacent areas). A better etching process would attack the oxide only, leaving the base metal intact, or with a protective layer that prevents the regrowth of thick oxide layers.
The standard bumping process begins with cleaning to remove organic residues and other undesirable films that result in poor metal adhesion. The cleaning stage can be a dry process involving plasma, a wet process, or a combination. The second stage typically involves a micro-etch to etch back the metal to a given thickness to remove oxides, additional contaminants, and in some cases, to roughen the surface a bit for better adhesion. For aluminum this is typically a mild phosphoric, acetic, and nitric acid (PAN) etch.
An additional cleaning and etch-back stage is required for copper pillar processes, as a small pillar (post) is first grown on the aluminum prior to growing the full pillar, which can range in height from a few microns up to 30 to 90 µm (Figure 1). After pillar growth, either a pure solderable cap, (typically pure Sn) or a full bump is grown. For copper, the etch back process is a solution of sodium persulfate and sulfuric acid (SPS) which can be formulated to have a target etch rate based on the process requirements.
The problem is that both PAN and SPS etch oxides and the base metal. Even though the etch rates are well-known, variables such as temperature, bath life, oxide thickness, and metal thickness contribute to process variation and yield loss. A solution that addresses the process variation to provide better process control of etching would be desirable for cycle time, scrap reduction, and long-term reliability.
Figure 1. Variations of bump processing: bump with copper stud, UBM only, and copper pillar with Sn cap.
Typically, a bumping house receives information from the internal/external customer indicating the metal thickness with little or no information on the oxide thickness (and exposure conditions prior to sealing and shipping). The information provided by the customer is used to baseline the cleaning recipes and etch process. When the information is not accurate (for example the metal thickness is listed as 5 µm, and the goal is to etch back 2 µm to 2.5 µm, but the thickness is actually only 2 µm to 3 µm) the result is over-etching leading to catastrophic thinning of the metal.
Both PAN and SPS etchants penetrate via grain boundaries, surface defects, and after etching through oxide layers. Since oxide thickness varies, the final metal thickness varies as the etch penetrates the areas of thin oxide first. A better process is to use etchant that attacks only oxide with little or no impact on the metal. A metal etch can be included on those occasions when necessary. (Metal etch rates in the angstroms would be useful.)
Two solutions* etch metal at less than 4 to 7 Å/min but are aggressive on the oxide. BPS172 is a targeted oxide remover that works well on copper pads, pillars, and bond surfaces. It removes the thick oxide layers from the metal, and then works to deposit a 3-4 Å “mono” layer of oxide on the copper. This monolayer can be measured 24 hours later and still only measure 10 Å. Figure 2 demonstrates that the oxide re-growth rate levels out at 5 to 8 hours at the 10 Å thickness. Ongoing test are under review to document extended periods to reflect longer lead times.
Figure 2. Oxide re-growth rate levels out at 5 to 8 hours at the 10 Å thickness.
Now the process engineer knows there is a conformal layer of oxide on the pad, rather than peaks and valleys of thick and thin layers of oxide. With minimal variation and better control on surface morphology (no etching on base metal), improvement in short- and long-term reliability can be demonstrated. Similarly, BPS 100 will etch the aluminum oxide on the bond pad without attacking the base metal like PAN etches. Neither material attacks organic passivation, nitrides, oxides, or glasses. Both are safe for various wafer chemistries.
A modified wet UBM process for copper would be the organic cleaner, followed by the BPS172 etch and finally on to UBM processing. Experimental data demonstrates that UBM layers are more conformal, eliminating the worry about incorrect measurements on the base metal thickness. Figure 2 provides data showing the copper oxide re-growth rate and demonstrates the control that the process engineer can implement when the oxide thickness is well characterized minimizing the need to overetch the base metal.
This solution can also be used on bumped die with SnPb and lead-free alloys that have thick oxide layers (it similarly reduces the oxide thickness on the Sn without attacking the Sn). Thick oxide on solder bumps results in poor reflow. To correct this problem, wafers are fluxed, sent through reflow a second time, and cleaned. In some cases solderability is still not improved and the wafer is scrapped. Similar to copper, this solution deposits a thin protective layer of oxide on the Sn to prevent re-growth of thick oxide layers, and can save time and money when compared to reflowing or scrapping wafers. The net result is better control during bump fabrication, testing, and assembly.
BPS100 and BPS172 from Air Products
Terence Collier, contributing editor, maybe contacted at CVInc. 850 S. Greenville, Suite 108, Richardson, Texas 75081; 214/557-1568; E-mail: email@example.com.