The challenges keep rising for fault isolation and failure analysis (FIFA) for the advanced semiconductor devices fabricated via integrated processes. Perceiving that defects randomly occurred during IC manufacturing contribute primarily to the device failures in comparison to those caused by harsh service environmental, we focus our efforts on fixing the defect issues in the processes, expecting a significant portion of the device failures may be prevented. A case study here demonstrates the procedure for fixing an inline defect issue via improving tool maintenance for the chemical-mechanical polishing (CMP) process. Through a correlative physical and chemical analysis down to atomic scale, a 10 nm diamond particle and a 10 nm metallic debris damaging one of the metal interconnect layers were defined. The analysis led to pinpointing the issue to a metal CMP process. By examining the process operation and the tool configuration, we located the diamond-missing sites on a pad-conditioning disk made with embedded diamond grits in a metal matrix. Preventive countermeasure were implemented to avoid the same defect recurring via resetting the disk life and maintenance.

By using fluorocarbon gases for aluminum (Al) pad open plasma etch, the pad inevitably has a thin surface remnant layer of Al-oxyfluoride (AlOF) by-product. This layer is chemically stable and does not directly cause issues in chip testing or wire bonding. This is true until open Al pads were exposed to a humid environment causing pad corrosion over time. The F-assisted corrosion created so-called black mushroom (BM) defects on the Al pads according to the defects appearance, resulting in the non-stick pads for wire bonding. Experimental tests were carried out to induce the Al pad corrosion via placing random fab-out wafers in a cassette pod hosting about 90% RH over a period up to a week. Optical imaging revealed BMs nucleated, primarily at Al grain boundaries. BMs were found all to be composed of O, F, and Al. In the cross section, BMs were shown to have separations of F-rich region next to Al and O-rich region towards the surface. In addition, BMs were composed of small crystallites and were porous. The former indicates an ionic bonding involving in O, F, and Al. The latter indicates the corrosion generated gaseous byproduct. A moisture (H 2 O) involved cyclic chemical reaction incorporating these analyses has been formulated. Factors to prevent BM formation were discussed.

This paper presents the results of an investigation to gain a better understanding of the impact of wafer substrate copper (Cu) contamination on FinFET devices. A chip from a wafer free of Cu contamination and several chips near a Cu contaminated wafer edge were sampled for chemical, structural, and morphological analysis and electrical device performance testing. The contaminated wafer was also annealed at high temperature, trying to drive Cu diffusion further into the Si substrate. TEM analysis revealed that the Cu interacted with Si to form a stable η-Cu 3 Si intermetallic compound. SIMS analysis from the backside of the wafer detected no Cu even after most of the backside material was removed. Likewise, electrical nanoprobing showed no parametric drift in the FinFETs near the edge of the wafer, comparable to device behavior in a Cu-free Si substrate. These results indicate that the formation of η-Cu 3 Si with a well-defined crystalline structure and stable stoichiometry immobilizes Cu diffusion in the Si substrate. In other words, the impact of Cu diffusion in silicon has no effect on device performance as long as η-Cu 3 Si does not form in the FinFET channel or short any structures within the chip.

This paper explains how tunneling atomic force microscopy (AFM) was used to determine the cause of leakage in FinFETs along the boundary of SRAM cells. The leaking devices were electrically isolated using photoemission microscopy, but conventional FA techniques, including SEM and TEM imaging, found no structural abnormalities. Suspecting that the failures may be due to dopant-related issues, the authors obtained cross sections of both good and bad devices and scanned them in a tunneling AFM. The paper describes the sample preparation process and includes cross-sectional images showing the difference between good and bad transistors. In SRAM areas where no leakage occurred, the fins are well defined and evenly spaced. However, in the area where an emission spot was observed, two of the fins appear to be overlapping, the result of n-well implants that merged.