Combining high-throughput data streaming, multivariate analysis, and custom analytical algorithms to transform in-situ metrology into closed-loop process control.
Modern thin film deposition processes generate rich streams of in-situ metrology data yet these remain largely underutilized. Translating this data into actionable process decisions requires more than measurement hardware, it demands intelligent data infrastructure.
At SVC TechCon, k-Space Technical Sales Director, Heidi Olson, will show how researchers can maximize the impact of their metrology data. Her presentation, “Toward Intelligent In Situ Metrology: From Real-Time Data to Actionable Process Control,” takes place on Tuesday, April 28.
Based on work by Dr. Barry Wissman and Dr. Michael De Zeeuw at k-Space, Heidi Olson will discuss how researchers can combine their metrology data with AI and machine learning algorithms to unlock new materials and new processes. The presentation will outline k-Space’s effort to build a framework combining high-throughput data streaming, multivariate analysis, and custom analytical algorithms to transform in-situ metrology into closed-loop process control.
Metrology and AI
Modern computing power, when paired with purpose-built algorithms, can extract process-relevant information from complex metrology signals in real time. For example, a high-throughput image streaming architecture delivers full-frame RHEED data — up to 1.58 MP at 70 fps — directly to user analysis environments or back into the deposition tool for real-time correction, expanding the volume and fidelity of data for process inference.
Next, is Principal Component Analysis. Principal Component Analysis (PCA) plays a key role by reducing high-dimensional RHEED image data to a compact set of physically meaningful components while preserving dominant process information. Applied to representative MBE growth datasets, PCA projection coefficients resolve intensity oscillations, identify surface phase transitions and reconstructions as changepoints, and reveal crystallographic symmetry in rotating RHEED patterns. PCA reprojection error serves as a sensitive, real-time anomaly indicator. When surface behavior departs from the learned baseline, error increases sharply, providing an automated flag for unexpected process excursions.
Finally, kSA FitTool illustrates how custom algorithms can be embedded directly into this infrastructure. It performs iterative optical model fitting on spectral reflectivity and transmissivity data (375-900 nm) in real time to extract multilayer film thickness. With incoherent layer support for transparent substrates and fitting times ranging from ~10 ms for single-layer films to under 100 ms for complex multi-layer stacks.
Together these capabilities define a scalable intelligence layer for in-situ metrology with AI/ML-ready data outputs, resulting in a clear path toward autonomous closed-loop control.
Click here for the complete SVC TechCon program.
