Richard Bersin, Amanda Horn, Han Xu, ULVAC Technologies Inc., Methuen, Massachusetts
Douglas Dopp, Motorola MOS-11, Austin, Texas
Mohamed Boumerzoug, Motorola MOS-12, Phoenix, Arizona
Photoresist removal following dry etching or high-dose ion implantation conventionally employs solvents and acids, sometimes preceded by a dry oxygen-based plasma ash. These costly, hazardous, and polluting wet chemicals are then disposed of through environmentally unfriendly waste-disposal processes, often contributing to global warming, substantial energy consumption, ground water contamination, etc. A new cleaning process (ENVIRO) that dry ashes etched resist and simultaneously renders remaining material 100% DI water soluble has been qualified in manufacturing and successfully used for more than 12 months. Normalized for a 10,000 wafer-starts/week fab, this process can save more than $5 million/year in solvent costs alone.
The removal of post via etch photoresist and etch residues can be quite challenging. The via etch process is designed to use polymer-forming gases to create a vertical etch profile and to etch oxide selectively while using a TiN barrier film as an etch stop layer. These chemistries are commonly a combination of CF4, CHF3, and Ar . Often, post-etch residues or "via veils" form along the via sidewalls as a byproduct of the etch chemistry (Fig. 1).
Figure 1. "Via veils" on via sidewalls after post via etch.
Via veils contain entrapped species of the materials etched, generally silicon-oxy-carbon compounds plus metal from the etch-stop layer. They may also contain fluorine. The photoresist is conventionally removed by reaction of molecular oxygen at >200°C wafer temperature. This ashing process serves to volatilize carbon chains of the photoresist, while oxidizing the exposed metal [2, 3].
The combination of polymer-forming via etch chemistries and high-temperature ash processing can create via veils that heretofore could only be removed in aggressive, wet chemical solvents.
Proprietary hydroxyl amine (HA) based chemical strippers have been found effective in removing via veils . However, these chemicals are most effective at >70°C. Water concentration must also be maintained to guarantee veil removal. Often the elevated temperature and equipment exhaust accelerates water evaporation from the chemical mixture, necessitating periodic refreshing to maintain temperature and water concentration.
While spray tools are often used for this process because of their throughput and floor-space advantages, spray processes are generally very difficult to monitor without extensive device testing. An improperly adjusted nozzle or a leaking valve may go undetected until ICs fail at probe and hundreds of wafers are in jeopardy.
The use of solvents is also costly. In addition to a proprietary solvent, an intermediate alcohol rinse, typically methanol or isopropyl alcohol (IPA), is done before final DI water rinse to prevent metal layer attack. The alcohol can become contaminated with the solvent and is difficult to recycle. Chemical dispense units, as well as dedicated chemical drains and reclaim tanks, must be provided for solvent and intermediate chemicals. These additional facilities add cost and complexity to the strip process (Fig. 2).
Figure 2. Two 5700-gallon stainless-steel reclaim vessels and associated plumbing lines to service HA and isopropyl chemicals used for conventional solvent de-veil processing.
Another alternative for this process is the "dry via de-veil ENVIRO" process that replaces the solvent wet chemistry discussed above with the dry chemistry of non-equilibrium gas plasmas, altering the chemistry of the polymer residues to render them water soluble (Fig. 3). Using free radicals created within the plasma initiates reactions at lower temperature. By maintaining a low temperature, this process avoids oxidation of metallized via veils, leaving them in water-soluble chemical forms such as fluorides.
In a joint effort, the authors placed this new process into manufacturing to replace an existing solvent cleaning process that was resulting in unacceptably high levels of scrapped wafers, which was attributed to the solvent cleaning step .
The joint effort
In our initial tests with the conventional process, we first stripped wafers in a commercially available dry-ash microwave-downstream tool using O2-N2 at 250°C. We then rinsed the wafers with HA for 20min using a solvent spray tool. We rinsed the wafers with IPA, rinsed again with DI water, and dried with heated N2.
Figure 3. Comparison of ENVIRO de-veil and conventional process flows.
In contrast, the dry de-veil treatment takes place in a series of plasma process steps. The process in this work used NF3, O2, and H2N2. In each step, process gases, microwave and RF power, operating pressure, wafer platen temperature, and automated optical endpoint can be selected. The use of fluorine-containing gases is minimal, typically <5% of total flow. We monitored gas concentrations in the process exhaust for a typical de-veil process with a residual gas analyzer (RGA).
Following the dry resist strip and veil treatment, we removed residues using a spin rinse dryer (SRD) a 5 min DI water rinse at room temperature with a heated N2 dry.
The dry de-veil process is a multiple step RIE process done at or near room temperature. We were concerned that the addition of a RIE process would result in particle contamination from ion bombardment of the chamber walls and the potential of device damage from charging effects. These and other qualification concerns were addressed through numerous runs of split lots to allow for the direct comparison of the new dry de-veil process to the existing wet solvent process. Only the ash and solvent cleaning steps were varied. All other processing was identical across the splits.
In addition to split lots, the new process was slowly ramped into production over several months while we monitored lot yield and performance data for any anomalies.
To compare the cost of the two processes, we modeled a 10,000 wafer starts/week factory, assuming a triple-layer metal IC where each wafer would receive two passes through the via de-veil process. Solvent usage, waste disposal, and consumable part costs were based on actual manufacturing data and scaled to 10,000 wafer starts. Where manufacturing data were not available, we made estimates based on similar processes and equipment. Tool costs are rough estimates based on the equipment manufacturer's list prices.
Split lot results
We found that ICs made using the new dry de-veil process were equivalent to those processed using the wet solvent process (Table 1). No particle contamination or end-of-line charging damage was detected. Ramp lot data showed an improvement in Kelvin, or individual via, resistance. In addition, three split lots showed no 168 hr burn-in failures for either process.
With cross-sectional SEMs and TEMs, we observed that the dry de-veil process completely removed via veils over a relatively wide range of de-veil tool process conditions. In addition, we found that the metal via glue deposition was more uniform, presumably due to complete veil removal with the absence of underlying metal attack.
Improved yields, better IC performance, and substantially reduced scrap combine to generate significant cost savings.
In addition, the complete elimination of solvents in itself generates a major reduction in manufacturing costs.
We performed a simplified cost analysis based on initial manufacturing data. These costs were scaled to a factory running 1,040,000 via de-veil wafers/yr, or 10,000 wafer starts/week each having two via layers.
Table 2 tabulates operational or raw material costs for each process. The resultant dry de-veil process cost/wafer is a fraction of the wet solvent process.
Table 3 tabulates overhead or equipment costs. Seven single-chamber dry de-veil tools and two spin rinse dryers are required to run at the 10,000 wafer start level.
Combining the data from these tables, the dry de-veil process results in a very significant net direct manufacturing cost savings of approximately $5 million dollars annually. The dry de-veil equipment return on investment can therefore be estimated at <10 months. In addition, with water as the only wet solvent used, the de-veil process markedly reduces hazard to human health when compared to acids, HA-based solvents, or NMP-based chemical strippers. Overall, DI water use is reduced and residual contaminants introduced into DI water are <0.2ppm weight fraction. Thus, recycling and reuse of this water can further reduce consumption of DI water.
Extension to other process cleans
Based on our work, the de-veil process has now been extended to wafer processing cleans and moved into manufacturing. Polysilicon and metal etch residues, as well as heavily ion-implanted photoresist are being removed using similar fluorinated process schemes, thus further reducing the use of hazardous, costly, and polluting solvents and acids. We are evaluating the resultant cost savings from these processes. The wide variation in the manner of fab setup and manufacturing-line configuration among manufacturers requires that individual cost analyses be performed to obtain comprehensive cost-savings data.
Sustainability of processes
The cost and environmental impact of sustaining various semiconductor-manufacturing processes is becoming a worldwide concern. Table 4 compares conventional wet and ENVIRO dry processes, the two used in our study reported here specifically the quantity of chemicals required.
The conventional process outlined in Table 4 results in ashing 71.8 lbs of photoresist/yr with CO2 and water byproducts of combustion in the plasma. Assuming that the resist is 100% carbon, the maximum emission would be 36 x 10-9 MMTCE (million metric tons of carbon equivalent as CO2). Thus, >99% of post-etch resist is ashed via dry plasma processing with the generation of a minuscule contribution to global warming. The problem with this approach is that the subsequent conventional solvent cleaning of ash residue results in a need to dispose of a vast quantity of chemicals. From the standpoint of resources, the ashing portion of the conventional process clearly is exceedingly more efficient than the final chemical cleaning portion.
The 3376 lbs of O2 used in the alternative dry process step is essentially also used to ash the 71.8 lbs of resist/year. The relatively small amount of NF3 is used to facilitate total solubilization and subsequent cleaning of the residues using only a DI water rinse. In the end, the conclusion reached in the analysis of Table 4 is that for removing just 0.718 lbs of resist strip residue over the course of one year 2,779,920 lbs of recoverable DI water is modest compared with 6,580,096 lbs of non-recoverable chemicals. Not only is use of 3,800,176 lbs of chemicals eliminated, but the remaining DI rinse water can be decontaminated and reused because of the very low level of residue contamination. This emphasizes the important advantages of using dry processes wherever possible to conserve resources and achieve greater sustainability.
The ENVIRO process does include the addition of a small amount of NF3 to the plasma chemistry. Since the global warming potential for NF3 is 8000x that for CO2, we should be concerned about the possible negative impact of introducing NF3. Table 5 summarizes the important aspects of the influence of the effluent gases on global warming. The analysis in Table 5 shows that even if all fifty-two 8-in wafer fabs worldwide were to adopt these processes, the MMTC equivalent would be increased by only 0.27%.
Also significant, the reduction of wet processing substantially reduces fab construction costs by eliminating plumbing and sinks (Fig. 2). A more complex aspect of solvent-based processes arises when decommissioning a wafer fab in the future. The very complex problems of disassembly and disposal of plumbing installations used for toxic, explosive, and hazardous liquids have to be considered.
Industrial ecology methods are most beneficially satisfied when manufacturing process schemes can be developed that eliminate the generation of hazardous waste emissions and disposal requirements, rather than employ such "end-of-pipe solutions" as abatement. In the best situation, the new dry de-veil process discussed here results in not only substantial cost savings to a manufacturer, but also a major reduction in environmental impact.
Complete elimination of solvent processing is a major shift that companies may be hesitant to undertake. Fortunately, we have found that the dry process discussed here can be gradually introduced into a conventional solvent process line with minimal interference and substantial cost benefit.
ENVIRO is a trademark of ULVAC Technologies Inc.
- K. Mocala, et al., "Characterization, Properties and Analysis of Via Veils," The Electrochemical Society Proceedings, Vol. 95-20, pp. 395-400.
- P. Singer, "Plasma Ashing Moves Into the Mainstream," Semiconductor International, p. 83, Aug. 1996.
- R.L. Bersin, et al., "Residue-Free Dry Stripping Photoresist Without Solvents or Acids: Time for a Paradigm Shift," Semiconductor Fabtech, Sixth Edition, pp. 341-347.
- M. Weling, I. Harvey, "Optimization of a Deep Sub-Micron Via Etch and Strip Process Stopping on TiN," The Electrochemical Society Proceedings, Vol. 96-12, pp. 435-443.
- The "via veil" portion of this paper is abstracted from a paper presented at the May 2000 meeting of the ECS in Toronto, Canada. This work comprised a joint effort between Motorola MOS-11 and ULVAC Technologies.
Richard Bersin received his BS in physics from MIT and his MS in physics from Northeastern University. Bersin is currently senior technical staff member at ULVAC Technologies Inc., Methuen, MA; ph 978/686-7550, fax 978/689-6302, e-mail email@example.com.
Amanda Horn is a process engineer at ULVAC Technologies.
Han Xu received his PhD in physical chemistry from University of Washington, and did postdoctoral research work on gas phase silicon wafer cleaning at MIT and surface chemistry and analysis at Harvard University. He is currently Director of Process Technology at ULVAC Technologies.
Mohamed Boumerzoug received his PhD from Institut National de la Recherche Scientifique in Quebec, Canada. Boumerzoug is a senior process engineer at Motorola MOS-12, Phoenix, AZ.
Douglas Dopp received his BS in chemical engineering from the University at Buffalo and MS in materials engineering from Rensselaer Polytechnic Institute. He is manufacturing manager at Motorola MOS-11, Austin, TX.