Controlling and Understanding Sensitized Ammonium Nitrate Hazards in Semiconductor Facility Ductwork
Stefan Uhlenbrock
Shock sensitive materials have had a significant impact on process safety related to the semiconductor industry. In 2014, a deadly explosion at a Mitsubishi Materials Corporation facility manufacturing hexachlorodisilane, resulted in 5 fatalities and 13 injuries [1]. This precursor is used for depositing polysilicon, silicon nitride, and silicon oxide films and can form shock-sensitive “poppy-gel” on partial hydrolysis if exposed to atmosphere [2], [3]. Similarly, epitaxial silicon deposition processes yield low vapor pressure oligo/poly chlorosilanes that deposit in the tool exhaust and upon partial hydrolysis from atmospheric moisture can also become highly shock sensitive [4]. Ammonium nitrate (AN) is a common by-product formed from various wet process steps and can condense in exhaust ducts of semiconductor facilities. For AN, the presence of small quantities of sensitizers can significantly enhance the shock sensitivity of this otherwise very stable material.
Understanding which chemistries sensitize AN and the concentration range in which this happens are crucial for establishing a safe production process and in determining the effectiveness of cleans to provide guidance for intrusive maintenance. Therefore, it is pertinent to test various by-products from thin film deposition and etch processes for energetic content as this can potentially have a significant impact on tool and process uptime and associated process safety. In 2015 Micron established the Byproduct Analysis and Sampling of Exhaust Systems (BASES) program. It is a web-based interface created for access to a specific business process, to upload analytical data, and to access procedures and supporting documentation. The BASES program focusses on two areas, facility exhaust ductwork and semiconductor tool pump exhausts lines. It covers guidelines for sampling to be performed at all scheduled and unscheduled preventive maintenance activities. For ductwork, a visual inspection is often followed by sample collection if a residue is observed. This is then submitted for chemical analyses. Based on the analytical results preventative maintenance protocols are developed, often resulting in a water flush of the ductwork. In addition, the BASES program enables optimization of sampling frequency. The BASES program has been in place for 6 years and it has taught us that by-products like AN can show up in very unexpected locations. This helps us to build a comprehensive and predictive program for controlling by byproducts such as AN. Micron currently utilizes primarily IC, ICP-MS, DSC, TGA, and FTIR to analyze any residue found as part of the BASES program However, an internal analysis on the shock sensitivity of a specific residue was not possible. A shipment of potentially energetic materials for analytical tests is very complicated or may not be possible at all. In addition, this is often a time-consuming step when immediate actions are required. Access to instant on-site testing capabilities would not only enable relaxation of conservative preventative maintenance protocols but also lead to higher tool availability.
Furthermore, the goal is to have these tests eventually evolve into a repository for different samples and analytical techniques. This will help to quickly identify chemistries which could potentially lead to high-risk situations. To enable these testing capabilities, Micron decided to acquire a BAM-Fallhammer instrument. The BAM-Fallhammer is the United Nations (UN) recommended instrument used to determine the sensitivity of a solid, gel, or paste to a mechanical force. The BAM-Fall hammer is an atypical piece of analytical equipment with regards to a semiconductor manufacturer. Most if not all analytical equipment for this industry is traditionally focused on quality of materials and contamination of wafers. For a typical Fallhammer experiment, approximately 40 mg of a sample is confined between two rollers and a collar. This setup is then placed onto the anvil of the Fallhammer. The instrument functions by allowing standard weights of 1 kg, 5 kg, and 10 kg to fall onto this assembly from an adjustable height.
Using the equation for potential energy (U = m·g·h, where g is approximated to 10 m/s2), the impact energy in Joules can be calculated. The test investigates for the presence of a visible flash, a detectable burn smell, a burn residue, and an audible detonation sound. A positive result for either one of those four parameters by itself classifies the material to be shock sensitive at the given impact energy. Several different sensitizers which could potentially be present in the ductwork of semiconductor facilities were mixed with pure AN and investigated for shock impact sensitivity. Mixtures for all experiments were prepared immediately prior to the impact testing. The maximum impact energy of the BAM-Fallhammer is limited to 100 J. This value was used as the screening threshold for all experiments to identify mixtures of concern. Additional tests are ongoing to identify the exact Minimum Impact Energy (MIE) and the associated influence towards intrusive maintenance or other process safety concerns. The Japanese Standard JIS K 4810 classifies the sensitivities of explosives into eight different categories with class 1 showing a MIE of < 2.5 J, and class 8, exhibiting a MIE of >25 J. Pure AN doesn’t exhibit shock sensitivity at 100 J. However, glucose is a known sensitizer for AN. Therefore, mixtures of AN/glucose at different ratios were investigated to establish a testing procedure and to design experiments for actual mixtures found in facility ducts. Although glucose isn’t a chemical used in semiconductor manufacturing, AN/glucose mixtures clearly demonstrated how the presence of small quantities of sensitizer dramatically lowers the MIE for AN. Adding only 5 % glucose drops the shock sensitivity to 50 J. For a 82 % AN/18 % glucose ratio, the MIE is only 10 J. As a reference point, very well-known explosives, like pentaerythritol tetranitrate (PETN) or cyclotetramethylene tetranitramine (HMX) have MIE of 12-17 J [5], [6]. We investigated the impact sensitivities for various mixtures of 80% AN/20 % sensitizers for residues potentially found in facilities exhaust ductwork at 100 J.
Even though the precise MIE for these chemistries is still under evaluation, it clearly demonstrates how small quantities of contaminants can lower the shock sensitivity of AN. Our tests demonstrate the value of quickly identifying potentially shock sensitive mixtures to evaluate process safety hazards related to various deposition and etch systems. Knowledge and understanding how these hazards are formed will enable Micron to better prepare for operational safety and improve tool uptime. Utilizing the BASES program with all its details results in quick identification of hazardous sample deposition and safe remediation of potentially hazardous circumstances. References: [1] M. Wilson, “Explosion and Fire at Yokkaichi Plant – Explanation, Consequences and Action items from the Incident that had Five Fatalities,” Kristiansand, Norway, Jun. 2016, pp. 179–188. [2] Y.-J. Lin et al., “Characterization of Shock-Sensitive Deposits from the Hydrolysis of Hexachlorodisilane,” ACS Omega, vol. 4, no. 1, pp. 1416–1424, Jan. 2019, doi: 10.1021/acsomega.8b03103. [3] X. Zhou, M. A. Wanous, X. Wang, D. V. Eldred, and T. L. Sanders, “Study on the Shock Sensitivity of the Hydrolysis Products of Hexachlorodisilane,” Ind. Eng. Chem. Res., vol. 57, no. 31, pp. 10354–10364, Aug. 2018, doi: 10.1021/acs.iecr.8b01241. [4] J. V. Gompel, “Complete Exhaust Management for SiGe EPI Processes,” presented at the SESHA, Scottsdale, AZ, Apr. 2004. [5] P. S. Wang and G. F. Hall, “Friction, impact, and electrostatic discharge sensitivities of energetic materials,” Monsanto Research Corp., Miamisburg, OH (USA). Mound, MLM–3252, May 1985. doi: 10.2172/5667780. [6] V. W. Manner, D. N. Preston, C. J. Snyder, D. M. Dattelbaum, and B. C. Tappan, “Tailoring the sensitivity of initiating explosives,” AIP Conf. Proc., vol. 1793, no. 1, p. 040036, Jan. 2017, doi: 10.1063/1.4971530.