I was assigned to the Metal Deposition department as an Equipment/Process engineering Intern focused on the physical vapor deposition (PVD) step of wafer fabrication. I designed, tested, and implemented an end-to-end analytics solution using production data to generate actionable insights.
The wafer process is highly sensitive-just a single spec of dust could render a $10,000 wafer unusable. Identifying root causes can take weeks, slowing down production with the allocation of people and resources. In this particular case, the Metals department was under scrutiny for potential hardware malfunctions in other departments after 77 wafers had been scrapped within the span of a few months. A lack of database was preventing the team from being able to rapidly identify the issue and set a plan for diagnosis and repairs on the hardware.
To investigate, I ran eight non-product wafers (NPWs) through PadAl chambers and performed TEM imaging at two locations per wafer. These were then compared to known defects, edge composition standards, and expected particle distributions.
I analyzed over 10,000 wafer images—both product and NPW—to identify defect signatures. I then developed a comprehensive database of known wafer defects, integrating Visedge, TEM, and VSEM imaging. This tool allows engineers to match new defects with known signatures and root causes, enabling faster diagnosis and resolution.
This is an example of the type of wafer images that were analyzed, keeping an eye out for areas where wafers are scratched, particles are concentrated, these are referred to as "signatures."
This is an example of an Endura PVD toolset which my work was centered around.
I recommended a thorough chamber inspection for issues such as flaking shield components or a worn electrostatic chuck. Once those were ruled out, the comparison of imaging data against known defect signatures, along with the careful inspection of the toolset confirmed that the Metals Department was not at fault.