Real fab comparisons reveal advantage to inorganic-based polymer removal
12/01/2002
Overview
Amid the clamor about the transition to copper-based interconnect processing, we perhaps lose sight of the fact that the industry still predominantly uses aluminum-based processing. Within aluminum processing, there is still a growing need to improve performance and to control associated manufacturing costs. Engineers at AMD's Fab 25 have done this with a comparison that replaced an established organic-based batch clean with a more functional inorganic single-wafer clean for post dry-etch applications.
AMD and SEZ have been collaborating to implement an SEZ-developed, single-wafer inorganic process at AMD's Fab 25. We have found that for use in aluminum-based processing, inorganic cleaning solutions offer a technically viable and cost-enhancing alternative to proprietary organic chemical-containing mixtures for post dry-strip polymer-residue removal [1].
 |
We have now collected data for one year in a fully ramped manufacturing operation. The resulting performance and cost enhancements are significant.
Previously at AMD, our process of record (POR) used a batch process tool and two proprietary organic solvent-based chemistries. Our desire for a new process was driven by chemical cost, marginal chemical performance, and equipment-related defect issues.
In addition, the POR needed two proprietary chemistries that were available from only one source and required dedicated waste streams. We also needed to move from batch to single-wafer processing to reduce fab line yield loss risk.
The chemistry
Our new process uses Kanto Chemical's DSP+ (dilute sulfuric peroxide, plus) — a blend of DI water (>80%), hydrogen peroxide, sulfuric acid, and hydrofluoric acid, the latter in extremely low proprietary concentrations. The ability of more concentrated sulfuric-acid hydrogen-peroxide mixtures to remove photoresist has been known for a long time [2, 3]. Its use in a dilute form for post dry-strip residue removal is relatively new, however.
Although IBM and Infineon have reported the use of similar chemical mixtures in batch wet-bench systems [4], these applications were less robust than the SEZ single-wafer process. The use in a batch system is much less robust due to a very narrow concentration window for chemical components within which the process works. The same narrow margin is not present with the SEZ single-wafer Spin-Processor, and a more aggressive formulation can be used.
An important aspect of this chemistry, compared to organic-containing chemistries for the same process, is that inorganic chemicals can be disposed of via existing diluted acid waste drains, thereby avoiding the need and expense of a dedicated drain and disposal system.
 Figure 1. Grain boundary pitting on an aluminum pad after using a) batch chemical and b) DSP+. |
null
Process, performance comparison
Overall, we found that our single-wafer spin-processor tool with its inorganic acid process could compete with batch tools on throughput for post-dry strip-residue removal, while providing enhanced process performance (Table 1). In fact, we found that the inherent hydrodynamics of the spin-processing system provides increased processing efficiency and shorter process times. The risk of wafer surface damage is therefore greatly diminished when using more aggressive chemistries. Furthermore, the chemistry in direct contact with the wafer is constantly replenished with fresh chemistry, improving process efficiency; the wafer's chemistry exposure time is significantly lower when using a single-wafer tool vs. a batch system. With the single wafer spin processor, we were able to clean metal lines in typically 30 sec total exposure to the chemicals compared to 8–20 min exposures with our previous batch process.
For more challenging via cleans, we found that the most effective cleaning sequence alternated the application of DSP+ with very short (i.e., a few seconds) intermediate DI rinses. Even for the most difficult clean (i.e., Via 6) the amount of time that each wafer sees the cleaning chemistry is 1/6th that in our previous batch process.
Using blanket-film monitor wafers for measurements, we found that aluminum removal (aka: etch rate) and particles added were both better with the single-wafer spray process (Table 2). In fact, particle-adder reduction was dramatically improved at even smaller particle sizes compared to our previous batch process. Measurement of smaller particle size reflects a tightening of specifications within the AMD process flow and highlights the advantage of this process both in terms of the absolute particle numbers and under more stringent selection criteria.
We also observed improved particle performance on patterned wafers — a 10% reduction in particles detected on via-structures and a 25% reduction on metal-lines. Furthermore, we noted a standard deviation decrease in the number of particles, indicative of improved process consistency.
There are several likely reasons for our data about particles:
- Because wafers are not immersed in a solution, there is no mechanism for particles to transfer from one wafer to another as happens in an immersion system. The same is true for particle transport from the front of one wafer to the back of another wafer. While the chemistry is re-circulated in the spray-cleaning tool for re-use, it is passed through filtration to remove particles.
- Because wafers see a very short, efficient process, there is less opportunity for particle retention.
The latter is also the reason there is less aluminum removal with the spray process; the etch rate using DSP+ is only 25Å wafer pass vs. 60–85Å with the batch chemistries. Using DSP+, we also noted a reduction in grain boundary pitting on aluminum pads (Fig. 1). Decreased pitting results in less likelihood of a decrease in a device's electrical characteristics.
With our development work, we observed several solid examples where the improvement in process time, particle contamination, and etch characteristics associated with the DSP+ clean yielded improved polymer removal capability. On metal lines (Fig. 2), for example, the improvement was particularly evident at the interface with the capping layer. This area typically suffers from the presence of a stubborn crust formed after plasma ash. The SEM images in Fig. 2 show insufficient removal of this crust using our previous chemistry compared to processing with DSP+. We noted similar improvements when cleaning sidewalls and the bottoms of vias.
 Figure 2. Metal lines cleaned with a) batch chemical and b) DSP+ based processing. |
null
Electrical comparison
While SEMs and particle numbers reveal very important information regarding process efficiency, ultimate success is determined by comparing electrical parametrics. With the DSP+ based process, we observed a 3% reduction in contact resists with a noticeable improvement in standard deviation (Fig. 3). Metal resistivity showed a 10% normalized reduction (Fig. 4).
The data in Figs. 3 and 4 show that individual devices have better performance, but they do not suggest an improvement in overall die yield. To confirm the latter, we have observed an increase in functional yield as more polymer removal steps are performed using the DSP+ process. To date, 3 of 13 layers cleaned with our new process have resulted in a 3% wafer functional yield increase.
 Figure 3. Normalized contact resistance. |
null
Chemical cost and lifetime
We compared the costs of organic batch and inorganic DSP+ chemicals for a nine-week period (Fig. 5). Our data revealed as much as a 75% reduction associated with the inorganic chemistry. This translates directly into cost-of-ownership of equipment and cost/die. At present, the time between chemical change-outs is 16 hours, but we have observed effective polymer removal characteristics for times well in excess of this, and we are extending the time between bath changes to 24 hrs.
 Figure 4. Normalized metal resistivity. |
null
There are three potential pathways for the chemical breakdown of DSP+:
- Once a certain number of wafers have been processed, the rate of polymer removal decreases to an unacceptable level due to the neutralization of active components after reaction with materials on the wafer's surface. This is dictated by the amount and nature of the polymer produced by dry strip steps and by the etch characteristics of the chemistry on the metal.
- We must also consider the breakdown of H2O2 and H2SO4 over time due to exposure to air and atmospheric conditions.
- In addition, HF is lost to vaporization over time.
The last two issues can be diminished by very tight process control during the formulation of the DSP+ and effective chemical storage. When we assayed the stability (i.e., normalized concentration vs. time) of H2O2 and H2SO4 in our chemistry, both components' concentrations underwent only minor changes in a 27-hr period. A separate test showed that there was little change in process performance during this test.
As stated earlier, trace metal build-up associated with metal etch can be a cause of diminished chemical lifetime and contamination concern (Fig. 6). As expected, we found that aluminum was the most prevalent, but its levels were well within specifications. Copper and titanium were also present in significantly lower concentrations. Overall, our work has shown that the single-wafer approach provides process-enabling performance that makes it a clear choice over traditional batch tools.
 Figure 5. Normalized chemicals cost/wafer for 13 layers of a six-layer metal process. |
null
 Figure 6. DSP+ trace metals concentration vs. bath age. |
null
References
- 1.L. Archer, et al., MICRO, p. 95, June 2001.
- 2.W. Kern, Journal of Electrochemical Society, Vol. 197, No. 6, p. 1887, 1990.
- 3.L.H. Kaplan, B.K. Bergin, Journal of Electrochemical Society, Vol. 127, No. 2, p. 986, 1980.
- 4.D.L. Rath, R. Ramachandran, European Patent Application No. EP 918081 A1, 1998.
Leo Archer received his PhD in inorganic chemistry at the University of New Mexico. He is a senior process engineer at SEZ America, 4829 S. 38th St., Phoenix, AZ 85040; ph 602/437-5050, fax 602/437-4949, [email protected].
Terri Couteau received her BS in physics at Sam Houston State University. She is a member of the technical staff at Advanced Micro Devices, Austin, TX.