Evaluating the defects in CVD diamonds: A statistical approach to spectroscopy

Laboratory-grown diamonds produced by chemical vapor deposition (CVD) can have similar sizes and colors to natural diamonds, and their color can be enhanced by post-growth treatment. Despite considerable published research on CVD diamond growth, post-growth treatment remains difficult to detect in some CVD diamonds. To improve the detection of post-growth treatment, we apply multivariate statistics to photoluminescence (PL) emission spectra and surface fluorescence images from 11,606 gem-quality CVD diamonds within the D-Z color range. Unsupervised statistics indicate as-grown CVD diamonds commonly have 3H defects, with zero-phonon line (ZPL) at 503.5 nm, and relatively intense PL peaks at 467.6 nm. These features are absent or minor in diamonds treated under high-pressure, high-temperature (HPHT) conditions. HPHT-treated CVD diamonds tend to exhibit the H3 system (with ZPL at 503.2 nm) as well as larger 491.9 nm and 566.6 nm peak areas. A subset of CVD diamonds showing the H3 system could have experienced treatment at low-pressure high-temperature (LPHT) conditions (including inside the reactor during or after CVD growth). As-grown CVD diamonds commonly have strong red, magenta, or blue surface fluorescence when exposed to ultra-deep ultraviolet radiation, whereas treated stones tend to have strong green or aqua (greenish blue) fluorescence. Diamonds with yellow fluorescence appear to have experienced treatment and may reflect combinations of green and red fluorescence contributed by H3 and NV0, respectively. To improve identification of treatment in CVD diamonds using PL emission spectra we derive machine learning classifiers using the random forest statistical method. New classifiers in this study have error rates as low as 0.65 ± 0.04 % (2σ) and can classify up to 84 % of current "undetermined" CVD diamonds as "as-grown" or "treated." These statistical methods can be recalibrated rapidly in light of new data.