AI-FPGA Driven Gas Chromatography for Automated and Enhanced Detection of Trace Diborane in Electronic Grade Nitrogen

Abstract

Accurate measurement of trace impurities such as diborane (B₂H₆) in ultra - high - purity nitrogen (N₂) holds utmost significance in semiconductor manufacturing because even minuscule amounts of contamination could have a negative impact on device performance and processing dependability; traditional Gas Chromatography (GC) methods face major challenges like unstable baselines, reduced signal - to - noise ratios at parts - per - billion (ppb) levels, and the simultaneous elution of interfering substances which lead to frequent manual readjustments and lower analytical accuracy; this research presents an automated GC system boosted by Artificial Intelligence (AI) for accurately measuring diborane in nitrogen samples; based on advancements in non - targeted chromatographic profiling, this study uses a Convolutional Neural Network (CNN) that was trained on a large dataset of chromatograms covering different concentration ranges and chromatographic situations; the CNN model simplifies the processes of identifying peaks, adjusting the baseline, and determining trace concentrations; designed for use on an FPGA - based inference platform, the system enables high - speed, real - time analytical functions; the results showed that the AI - improved system reached a diborane detection limit under 5 ppb, having a 95% reduction in false positives compared to standard integration methods and also made possible predictive maintenance via continuous monitoring of chromatographic performance indicators; this research demonstrated the feasibility of integrating AI directly into analytical procedures to maintain strict purity standards during the production of electronic specialty gases thus enhancing consistency and reducing the necessity for expert operators' involvement.