Since the invention of integrated circuits many advancements in processing technologies have been achieved. Electronic devices are used daily and therefore they have enormously impacted our lives and life without them would be unfathomable. Thus, the reliability issues connected to newly developed devices are of great technical and economic interest.
Integrated circuits (ICs) are an integral component of electronic devices such as computers, mobile phones, and other digital home appliances. ICs are simply very advanced electric circuits consisting of active and passive components. The most frequently implemented passive components are resistors and capacitors while the active components are diodes and transistors. Each component has a specific task to perform and they all have to work together to make the circuit perform as desired. Resistors are used to adjust signal levels, to limit current flow, bias active elements, and terminate transmission lines. Capacitors permit to temporarily store electrical energy in an electric field. Diodes allow current flow in only one direction and transistors are used to amplify or switch electronic signals and electrical power. These components are interconnected together by a single or multilevel metalization on a piece of single crystal silicon (Si) [4]. ICs range in complexity from simple logic modules to entire microcomputers, incorporating millions of components.
The first semiconductor-based electronic devices were constructed in 1947 for applications in medicine, military and other areas [4, 5, 6]. The first transistors were connected together with traces of metal that implemented different circuit functions. At that time these semiconductor devices could perform better than vacuum tubes or other electromechanical components. However, the real revolution was possible only with the development of the planar process which enabled the integration of multiple interconnect devices on one chip [4, 6].
Since their invention in 1959, ICs had a fast evolution. The first IC was composed of a few devices per chip, whereas now a typical IC has billions devices per chip, fabricated with a geometry size measured in nanometers [7]. In 1965, Gordon Moore made the important empirical observation that the number of components which fit on a single IC has doubled every year since their invention [8]. Subsequently, in 1975, looking forward to the next decade, he changed his prediction to a doubling every two years [9]. This technology revolution was supported by the exponential improvements of all the digital components, from processing speed and storage capacity to networking bandwidth and speed. The device miniaturization became strongly dependent on the process technology. The process technology and the number of transistors employed during the years are presented in the Table 1.1 [10, 11].
Name | Description | Year | Number of transistors | |
SSI | small-scale integration | 1964 | 1 to 10 | |
MSI | medium-scale integration | 1968 | 10 to 500 | |
LSI | large-scale integration | 1971 | 500 to 20 000 | |
VLSI | very large-scale integration | 1980 | 20 000 to 1 000 000 | |
ULSI | ultra large-scale integration | 1984 | 1 000 000 and more | |
ICs are fabricated by chemically or physically modifying the surface of a circular single-crystal Si called substrate (wafer). The build area of the IC is defined as a die.
To reduce the fabrication costs a large number of dies are processed on a single substrate and subsequently diced (die cutting). When a single die is packaged, it is then referred to as a chip. A single-crystal Si wafer is generally used as the base of ICs. Si is used because of its excellent electrical and physical properties fundamental for microelectronic devices compared to other semiconductor materials [12].
The IC is built by applying different chemical and mechanical processing steps on desired areas of the wafer. The basic processes employed to construct a device are:
In ICs, the electrical devices (capacitors, transistors, resistors, etc.) are built by applying the above described processes. All these steps are employed during IC fabrication and they are repeated until all the IC is completed. The manufacturing of an IC can require hundreds of processing steps.
Two different fabrication stages during IC manufacturing are defined, referred to as the Front End Of the Line (FEOL) and the Back End Of the Line (BEOL) [13]. These two stages have a very significant difference and therefore the semiconductor industry uses this terminology as a way to differentiate the stages of semiconductor processing.
FEOL refers to all the wafer processing steps of the initial stages of the IC processing. In this
stage transistors, capacitors, resistors, and local interconnects are manufactured. Local
interconnects are simple metal conductors, but at this stage they are used only as
local connectors between various devices; therefore, these processes do not include
the deposition of metal interconnect layer necessary to connect all the devices of the
FEOL process. On the other hand, BEOL are all the processing steps required for the
formation of the metal interconnects where the electrical devices get interconnect
wiring on the wafer (cf.
FEOL and BEOL stages are intensively studied in order to increase and maintain the performance of the IC necessary to continue device development.
The final stage of IC fabrication is the packaging. This last step is necessary because the ICs are small, fragile, susceptible to environmental damage and difficult to handle by the users. Ceramic and plastic are most commonly used materials used for the packaging technology. Plastic materials are preferred due to the lower cost compared to ceramics. A wide variety of IC packages are available for different IC architectures. This stage is the last assembly process before testing and shipping devices to the customers.
Because of the necessity to minimize the chip level features, the number of transistors per unit of area in IC continues to increase. All of the IC components must be electrically connected in order to provide the proper functionality. Over the years, different IC architectures were developed to maintain the miniaturization path.
Since the invention of the IC, the important role of the interconnects was already clear. With the increased complexity of the electronic devices the number of interconnects increased exponentially. At the beginning the connection of the electronic devices was difficult, expensive and unreliable. Through IC downscaling the transistor delay improved unlike the delay of caused by interconnect.
An evolution from two-dimensional (2D) to three-dimensional (3D) circuit fabrication was
necessary in order to avoid efficiency problems [14] (c.f.
The 2D IC approach consists of connecting different discrete devices with their packages using a
printed circuit board (mechanical support with conductive tracks) (cf.
In 2D ICs the dies can also be directly connected by wire bonding (cf.
The beginning of the 1990s was the time for the advent of Multi-Chip Modules (MCMs), where digital-only dies were mounted on the same package substrate [15].
The next step was the System-On-Chip (SOC) technology where all discrete devices were
implemented on a single die [16]. This mixed-technology design was innovative but it
lead to an increase in the chip area and as a consequence to transmission delays (cf.
The System-in-Package (SiP) was the new technology available at the start of the 2000s
(c.f.
All the single dies can be fabricated using the most appropriate technology process and subsequently integrated in the package.
By using the third dimension, the large number of long interconnects, needed in 2D structures, are replaced by vertical interconnects which have maintained, and even increased, the performance of the device. The vertical direction is used to stack and connect the logic gates using Through Silicon Vias (TSVs) [14, 18, 19]. The TSV is effectively a conductor connecting the top and bottom of a silicon substrate of the stacked dies. The conductor is electrically insulated from the substrate by a dielectric layer and its function is to connect the metal wires of the stacked dies. The dies are connected by employing solder bumps. Bumps are interconnections which enable the face-to-face electrical connection between two devices. TSVs and bumps are used in 2.5D and 3D IC architectures [14, 20].
To explain the 2.5D IC architecture we refer to the 2D IC/SiP. The main difference between
these two configuration is that, for the 2.5D IC/SiP an interposer is used between the
SiP substrate and the dies (c.f.
The development of the 3D TSV architecture depends on the progress of several technologies. All 3D IC fabrication processes comprise three basic steps [14, 18]:
These three processing steps are described in more detail in
A further distinction arises at the chip fabrication stage during which the via is formed.
Three different approaches are industrially defined. In the Via First or Via Last technology the
vias are fabricated before or after tier bonding, respectively, (tier is a structure comprising one
or more IC die) and in the Via Middle technology, the vias are fabricated after the FEOL
processes but before the BEOL interconnect metalization. The choice of approach depends on
the device architecture. Via Middle is the preferred manufacturing approach for 3D IC
applications because it is the least cost-intensive one. Via Last technology is mainly used for
complementary metal-oxide-semiconductor (CMOS) image sensors and Via First technology has
limitations mainly due to the via resistance [14]. In
3D system integration is the key to the realization of the “More than Moore” system fabrication where SiP and vertical integrations enable heterogeneous integration of different technologies [21].
3D integration, which enhances the performance of the IC by reducing the chip area and reducing the RC delay is made possible by TSVs. Stacking ICs and densely interconnecting them vertically gives several benefits [18].
One benefit is due to the heterogeneous functionalities including processing, sensing, memory, and data transmission which can be incorporated in a single die using TSVs. This opens new opportunities for efficient system integration. Dies can be built in different process technologies from different vendors and can be bonded at the later stage by a third party.
A second benefit is the capability of integration of incompatible technologies, leading to advantages in performance and form factor. In particular, the trend of smart technologies, including smart-watches, phones, and tablets need increasing functionality in a decreasing amount of space. Only by stacking can this combination of dense packing and increased functionality be enabled.
A third benefit lies 3D structure geometry itself. In a 3D structure the average wiring length is shorter than in a 2D structure and therefore the signal paths between dies is shorter, making the circuit faster. This contributes to a power decrease and a faster data transmission.
Another important benefit is related to the production cost, the chip stacking enables more cost-efficient integration.
Stacking dies in 3D using TSVs as interconnects results in a variety of different advantages as detailed above. However, in order to fabricate 3D TSVs many separate processing steps are necessary. In the following, a simplified TSV fabrication procedure is described.
The
The
The
The
The
Once the TSV fabrication is completed the dies or the interposer layer have to be bonded together in pairs. The bonding requires a perfect alignment, where TSVs must be correctly placed to establish the electrical connection between dies and/or interposer layer. Any offset may result in increased local stresses and degraded performance.
The TSV geometry differ slightly for 3D IC and 2.5D interposer applications. Usually, for 3D
IC applications TSVs have a small dimension, 1-5
In
Copper (Cu) and tungsten (W) are the materials most frequently employed as metalization in the TSV structures. For Via First technology poly-silicon can be employed as well. The frequently used materials with their features are listed below:
The choice of material also depends on the geometric dimensions of the TSV. The coefficient of thermal expansion (CTE) mismatch between the TSV metalization and the surrounding Si is a cause of thermal stress in the surrounding Si, causing reliability issues for the TSV structure and surrounding devices, when subject to different thermal loads. Since TSV stresses increase as the TSV diameter increases, the choice of the metal for TSVs can be determined by the required depth and aspect ratio of TSVs demanded for a specific architecture.
The application of TSVs can be split into two main areas [18]:
CMOS image sensors are the one of the first devices to use 3D integration in high volume manufacturing. Since 2007-2008, Toshiba, Aptina, ST Microelectronics, and other companies have commercialized CMOS image sensors which implement TSVs. The use of TSVs resulted in an increased lateral compactness permitting the design of miniature devices [30]. This advantage led to a wide implementation of CMOS image sensors in cellular phones and tablets. Usually, a Via Last technology is used for the processing of TSVs within the fabrication of CMOS image sensors. An Austrian sensors manufacturer, ams AG developed Open TSV structure for CMOS image sensors. This open TSV technology allowed for an electrical connection between a thinned wafer on top and a CMOS image sensor, bonded to the bottom [29]. The bottom and top wafers are electrically connected by TSVs. In this application the use of open TSVs results an improved thermal behavior compared to filled TSVs.
In order to increase the performance of DRAM a shrinking of the dimensions while maintaining a sufficient amount of capacitance in a memory cell and low level of leakage current is required. In 2014 Samsung started the mass production of double data rate-4 (DDR4) registered dual n-line memory modules which use the 3D TSV packaging technology [30]. In the future, Samsung plans to continue to work on improving 3D TSV technologies in order to create even higher density DRAM modules supporting the transition from DDR3 to DDR4.
In 2015 Toshiba announced the development of the first stacked NAND flash memories utilizing TSVs [31]. NAND are flash memories which enable the largest bit density among solid-state memory devices. The application of TSV technology in this device produces an increase in the speed of reading and writing data while reducing the power consumption.
Micron Technology (memory manufacturer) and Intel (CPU manufacturer) have jointly built a new breed of integrated architectures based on 3D chip packaging [30]. TSVs are used in order to resolve many problems regarding the mismatch in the clock speed between CPUs and memory.
Microelectromechanical systems (MEMS) can also incorporate TSVs. MEMS are mainly used for microsensors and microactuators where piezo-capacitive components are used to measure a physical phenomenon and convert it to an electrical signal. Usually the signal level is low and noise sensitive TSVs can be employed to reduce the parasitic capacitance. Since 2011, ST Microelectronics began to replace traditional wiring with short TSVs in MEMS devices such as smart sensors and multi-axis inertial modules [30].
The continued demand for electronic devices having a more compact form factor, more versatility, higher performance and lower power consumption will lead to further innovation and continues development of 3D technology.
In microelectronics, interconnect structures bear the task of electrically connecting several components of an IC. It is evident that, along with the reliability of the device themselves, the reliability of the interconnect structures must be thoroughly investigated in order to limit the probability of IC or chip failure. By definition, reliability is the probability that a device, operating under defined conditions, will continue to operate under those conditions for a defined period of time [32]. On the other hand, failure can be defined as any condition that causes a device or circuit to cease to operate in a desired manner [32]. A failure can appear instantly after a shock or grow slowly and decrease the device functionality over time until arriving at a failure condition. Each failure mode is the result of a certain failure mechanism in which specific combinations of material properties and the surrounding environment act together.
Interconnects can mechanically fail due to three main mechanisms [32]: tensile rupture (fracture due to mechanical overloading), creep failure (damage caused by a long-lasting permanent load or stress), and fatigue (damage caused by cyclical loads or stresses). These mechanisms affect the interconnection simultaneously. The thermomechanical stress is the main source of mechanical failure (fatigue, delamination, creep, etc.). However other factors can also affect the failure of a device, such as electrical and chemical actions. Interconnections exposed to high electrical current densities are subject to electromigration-induced voiding. Corrosion phenomena metals can accelerate fatigue and delamination failure. In the list below the most frequent mechanisms leading to failure in devices are outlined [33]:
The finite element method (FEM) (also called finite element analysis (FEA)) is a numerical method for solving a system of partial differential equations (PDEs). Most physical problems (fatigue, electromigration, thermally induced failure mechanism, delamination, etc.) can be described using a set of PDEs. FEM is employed to solve these PDEs and obtain an accurate analysis of many different physical problems [32].
In a continuum problem of any dimensions an unknown field variable (displacement,
potential, etc.) has infinitely many values because it is a continuous function of generic points in
the body that generates an infinite number of unknowns. The main idea of FEM is to discretize
a domain into a finite number of elements in a way that the entire domain is approximated by
the union of these discrete elements (cf.
The unknown field variable is expressed by assuming approximating functions (interpolation
functions) within each element. At specified nodes or nodal points of the domain, the
approximating functions are defined in terms of values of the field variables which are the
fundamental unknowns. Nodes are usually located on the element boundaries, where adjacent
elements are connected (cf.
FEM is widely employed in microelectronics. A multitude of different physical phenomena can be handled by choosing the desired geometry and using the material data required. The FEM tool permits the determination or prediction of the stress limits, or the lifetime of specific configurations, fundamental for the reliability study.
In this document a detailed analysis of the mechanical failure of open TSV is presented. Finite element simulations are employed to investigate the possible causes which lead to mechanical failure. Different simulation schemes, material, and mechanical models are applied. By means of simulations, different scenarios are studied leading to a comprehensive analysis of the mechanical behavior of TSVs. The goal of this work is to improve and analyze the reliability of the TSV interconnections and thus of the entire device leading to a deeper understanding of TSVs. The study can help developers to improve their processes.
The dissertation is divided into six chapters, including this introductory chapter,