Abstract

Ohmic metal semiconductor contacts are indispensable part of a semiconductor device. These are characterized by their specific contact resistivity (ρ_c) in expressed in Ohm-cm^2, defined as the inverse slope of current density versus voltage curve at origin. Engineering and measurement of specific contact resistivity (ρ_c) is becoming of increasing importance in the semiconductor industry. Devices ranging from integrated circuits to solar cells use contact resistivity as a measure of device performance. Novel methods such as contact silicidation, doped-metal contacts, dipole inserted contacts etc. are continually being developed to reduce specific contact resistivity and improve device performance. The Transmission Line Measurement (TLM) method is most commonly used to extract the specific contact resistivity for such applications. This method is, however, not fully understood and modeled to understand the flow of current and behavior of charge carriers for contacts of different dimensions. It has often been observed in literature that applications that involve smaller TLM geometries most often than not, show low values of ρ_c and applications that involve ρ_c extraction through larger TLM geometries show significantly larger values. A perfect example of this would be the inconsistencies observed in extracted ρ_c's from integrated circuit applications where TLM geometries range from 0.1 um to 10 um and extracted ρ_c is of the order of 10^{-8} to 10^{-6} Ohm-cm^2 and photovoltaic applications where geometries are around 50 um to 1000 um and ρ_c is of the order of 10^{-5} to 10^{-2} Ohm-cm^2. The transfer length or L_T which is the characteristic length that the charge carriers travel beneath the contact before flowing up into the contact. It has also been seen that in certain cases of TLM device dimensions, the extracted L_T is greater than the actual length of the contact. This occurence cannot be effectively explained through the conventional TLM analysis.

In this project, the inconsistencies observed in literature were initially attributed to the error in measurement. Equations for relative uncertainty due to systematic error were optimized to obtain values of optimum TLM widths for application specific values of ρ_c. TLM structures with varying widths were fabricated and tested. Underlying doped regions were created through methods of ion implantation and spin-on-doping targeted for particular values of sheet resistance. The contacts were fabricated on high and low values of sheet resistances using Aluminum, NiSi and TiSi_2 metals. This was used to experimentally compare the experimental and simulated values of the optimum widths. The devices were also fabricated with changing contact length in order to try to explain the occurence of the transfer length to be greater than the length of the contact. The experimental mask design had test structures with constant width and varying TLM lengths. Scaling structures where both the length and width of the TLM geometry were also increased proportionally to evaluate the scaling effect of the TLM length and width on the extracted transfer length.

The fabricated TLM structures were then tested and the data was analysed to obtain values of the transfer length (L_T) and ρ_c. The relative uncertainty due to systematic error in ρ_c was also evaluated. The experimental values of the optimum widths for the least amount of measurement error were a close match to those obtained through simulations. It was also observed that for a contact made with a particular metal on a doped layer of a particular sheet resistance, the L_T increased as the width of the TLM structure increased. Many cases were observed where the extracted L_T was greater than the length of the contact, indicative of current crowding. This was the first time this relation was observed and this prompted a mask design with changing TLM lengths. A similar linear relation was observed on constant width and changing the length of the contacts. The scaled structures showed that on simultaneously increasing the length and width of the TLM contacts, the transfer length proportionally increased. There is, therefore, a geometric dependence of L_T extracted from the measurement of the TLM structures. Through the use of the exact field solution modeling, L_T is underestimated in the integrated circuit application space due to current crowding effects and overestimated in the case of silicon photovoltaics. There is no "one-size-fits-all" geometry that can be used for any particular application space. Due to the observed underestimations, it was also concluded that the TLM method is not an appropriate method to determine ρ_c for nanoscale contact applications.

Publication Date

12-2016

Document Type

Thesis

Student Type

Graduate

Degree Name

Materials Science and Engineering (MS)

Department, Program, or Center

School of Chemistry and Materials Science (COS)

Advisor

Santosh Kurinec

Advisor/Committee Member

Michael Jackson

Advisor/Committee Member

Robert Pearson

Campus

RIT – Main Campus

Share

COinS