Lambda-decoupled T-ray Imaging and Profile Mapping for Wafer-scale Metrology and Improved Yield

A. Rahman
Applied Research & Photonics, Inc.,
United States

Keywords: terahertz (T-ray), Volume imaging, Profile mapping, Yield improvement


Billions of wafers/year are fabricated at the foundries around the world. While the process for a given IC is perfected before mass production, in some cases the rejection rate of fully processed wafers is as high as 30%, especially for the 300 mm wafers. Since the dies must be singularized and packaged, it is of paramount importance to test the functionality before the laborious and expensive task of packaging. A whole wafer is rejected when the percentage of defective dies exceeds a threshold. The yield is directly dependent on the number of good dies. This paper introduces an advanced terahertz (T-ray) technique of wafer-scale metrology via a novel cameraless volume imaging route [1]. The technique allows wafer-scale profile mapping for identifying good dies on a patterned wafer; thus, helps sorting the dies that will pass the subsequent test. A criterion is proposed to identify the good dies via the _magnitude variation of profile wrt a known good die’s profile, resulting in a significant improvement of the yield. The concept of ‘profile mapping’ is introduced for the first time that allows for rapid sorting of all dies on a patterned wafer. “Dendrimer dipole excitation (DDE)” [1] is a new route for T-ray generation from an electro-optic dendrimer. The DDE T-ray source generates wider bandwidth (0.1 THz to 30 THz) and >200 milliwatts CW power that enables many applications for semiconductors. A T-ray nanoscanning spectrometer and 3D imager (TNS3DI), designed around the DDE source, was used for the present work. Here, the wafer remains stationary while the T-ray scans it over the user specified volume or area. The scanned traces of the reflected intensity are utilized to generate the T-ray profile and image via the inverse distance to power equation algorithm [1]. Areas of good and bad dies are visualized. The magnitude difference at a given location of each die (magnitude = peak intensity – trough intensity) is used as a criterion to compare all dies wrt the magnitude of a known good die. The reflected intensity is a physical parameter for the arrangement of materials on a die, thus, a chosen location provides the intensity profile. The disagreement between the magnitude of individual dies wrt the reference die is computed, Δ_i=|M_ref-M_Di |, where M_ref is the magnitude of the reference die and M_Di is the magnitude of the ith die. Ideally, all Δ_i are expected to be identical. In practice, the lithography process creates differences between the same features on different dies. While such difference will be small for a good die compared to the reference, a limit is established beyond which the functionality of a given die will deteriorate. That is, for a good die a criterion is set as, Δ_i≤∆_max=|M_ref-M_Di |, (1) where, ∆_max is the maximum acceptable value of the magnitude beyond which a die must be eliminated from further packaging. From the traces along the mid-point, ∆_max is computed as a criterion to sort the dies that will pass functionality test.