Matching Items (14)
Filtering by
- All Subjects: deep learning
- All Subjects: Materials Science
Description
Amorphous oxide semiconductors are promising new materials for various optoelectronic applications. In this study, improved electrical and optical properties upon thermal and microwave processing of mixed-oxide semiconductors are reported. First, arsenic-doped silicon was used as a model system to understand susceptor-assisted microwave annealing. Mixed oxide semiconductor films of indium zinc oxide (IZO) and indium gallium zinc oxide (IGZO) were deposited by room-temperature RF sputtering on flexible polymer substrates. Thermal annealing in different environments - air, vacuum and oxygen was done. Electrical and optical characterization was carried out before and after annealing. The degree of reversal in the degradation in electrical properties of the thin films upon annealing in oxygen was assessed by subjecting samples to subsequent vacuum anneals. To further increase the conductivity of the IGZO films, Ag layers of various thicknesses were embedded between two IGZO layers. Optical performance of the multilayer structures was improved by susceptor-assisted microwave annealing and furnace-annealing in oxygen environment without compromising on their electrical conductivity. The post-processing of the films in different environments was used to develop an understanding of mechanisms of carrier generation, transport and optical absorption. This study establishes IGZO as a viable transparent conductor, which can be deposited at room-temperature and processed by thermal and microwave annealing to improve electrical and optical performance for applications in flexible electronics and optoelectronics.
ContributorsGadre, Mandar (Author) / Alford, Terry L. (Thesis advisor) / Schroder, Dieter (Committee member) / Krause, Stephen (Committee member) / Theodore, David (Committee member) / Arizona State University (Publisher)
Created2011
Description
The increased use of commercial complementary metal-oxide-semiconductor (CMOS) technologies in harsh radiation environments has resulted in a new approach to radiation effects mitigation. This approach utilizes simulation to support the design of integrated circuits (ICs) to meet targeted tolerance specifications. Modeling the deleterious impact of ionizing radiation on ICs fabricated in advanced CMOS technologies requires understanding and analyzing the basic mechanisms that result in buildup of radiation-induced defects in specific sensitive regions. Extensive experimental studies have demonstrated that the sensitive regions are shallow trench isolation (STI) oxides. Nevertheless, very little work has been done to model the physical mechanisms that result in the buildup of radiation-induced defects and the radiation response of devices fabricated in these technologies. A comprehensive study of the physical mechanisms contributing to the buildup of radiation-induced oxide trapped charges and the generation of interface traps in advanced CMOS devices is presented in this dissertation. The basic mechanisms contributing to the buildup of radiation-induced defects are explored using a physical model that utilizes kinetic equations that captures total ionizing dose (TID) and dose rate effects in silicon dioxide (SiO2). These mechanisms are formulated into analytical models that calculate oxide trapped charge density (Not) and interface trap density (Nit) in sensitive regions of deep-submicron devices. Experiments performed on field-oxide-field-effect-transistors (FOXFETs) and metal-oxide-semiconductor (MOS) capacitors permit investigating TID effects and provide a comparison for the radiation response of advanced CMOS devices. When used in conjunction with closed-form expressions for surface potential, the analytical models enable an accurate description of radiation-induced degradation of transistor electrical characteristics. In this dissertation, the incorporation of TID effects in advanced CMOS devices into surface potential based compact models is also presented. The incorporation of TID effects into surface potential based compact models is accomplished through modifications of the corresponding surface potential equations (SPE), allowing the inclusion of radiation-induced defects (i.e., Not and Nit) into the calculations of surface potential. Verification of the compact modeling approach is achieved via comparison with experimental data obtained from FOXFETs fabricated in a 90 nm low-standby power commercial bulk CMOS technology and numerical simulations of fully-depleted (FD) silicon-on-insulator (SOI) n-channel transistors.
ContributorsSanchez Esqueda, Ivan (Author) / Barnaby, Hugh J (Committee member) / Schroder, Dieter (Thesis advisor) / Schroder, Dieter K. (Committee member) / Holbert, Keith E. (Committee member) / Gildenblat, Gennady (Committee member) / Arizona State University (Publisher)
Created2011
Description
Electromigration in metal interconnects is the most pernicious failure mechanism in semiconductor integrated circuits (ICs). Early electromigration investigations were primarily focused on aluminum interconnects for silicon-based ICs. An alternative metallization compatible with gallium arsenide (GaAs) was required in the development of high-powered radio frequency (RF) compound semiconductor devices operating at higher current densities and elevated temperatures. Gold-based metallization was implemented on GaAs devices because it uniquely forms a very low resistance ohmic contact and gold interconnects have superior electrical and thermal conductivity properties. Gold (Au) was also believed to have improved resistance to electromigration due to its higher melting temperature, yet electromigration reliability data on passivated Au interconnects is scarce and inadequate in the literature. Therefore, the objective of this research was to characterize the electromigration lifetimes of passivated Au interconnects under precisely controlled stress conditions with statistically relevant quantities to obtain accurate model parameters essential for extrapolation to normal operational conditions. This research objective was accomplished through measurement of electromigration lifetimes of large quantities of passivated electroplated Au interconnects utilizing high-resolution in-situ resistance monitoring equipment. Application of moderate accelerated stress conditions with a current density limited to 2 MA/cm2 and oven temperatures in the range of 300°C to 375°C avoided electrical overstress and severe Joule-heated temperature gradients. Temperature coefficients of resistance (TCRs) were measured to determine accurate Joule-heated Au interconnect film temperatures. A failure criterion of 50% resistance degradation was selected to prevent thermal runaway and catastrophic metal ruptures that are problematic of open circuit failure tests. Test structure design was optimized to reduce resistance variation and facilitate failure analysis. Characterization of the Au microstructure yielded a median grain size of 0.91 ìm. All Au lifetime distributions followed log-normal distributions and Black's model was found to be applicable. An activation energy of 0.80 ± 0.05 eV was measured from constant current electromigration tests at multiple temperatures. A current density exponent of 1.91 was extracted from multiple current densities at a constant temperature. Electromigration-induced void morphology along with these model parameters indicated grain boundary diffusion is dominant and the void nucleation mechanism controlled the failure time.
ContributorsKilgore, Stephen (Author) / Adams, James (Thesis advisor) / Schroder, Dieter (Thesis advisor) / Krause, Stephen (Committee member) / Gaw, Craig (Committee member) / Arizona State University (Publisher)
Created2013
Description
This dissertation addresses challenges pertaining to multi-junction (MJ) solar cells from material development to device design and characterization. Firstly, among the various methods to improve the energy conversion efficiency of MJ solar cells using, a novel approach proposed recently is to use II-VI (MgZnCd)(SeTe) and III-V (AlGaIn)(AsSb) semiconductors lattice-matched on GaSb or InAs substrates for current-matched subcells with minimal defect densities. CdSe/CdTe superlattices are proposed as a potential candidate for a subcell in the MJ solar cell designs using this material system, and therefore the material properties of the superlattices are studied. The high structural qualities of the superlattices are obtained from high resolution X-ray diffraction measurements and cross-sectional transmission electron microscopy images. The effective bandgap energies of the superlattices obtained from the photoluminescence (PL) measurements vary with the layer thicknesses, and are smaller than the bandgap energies of either the constituent material. Furthermore, The PL peak position measured at the steady state exhibits a blue shift that increases with the excess carrier concentration. These results confirm a strong type-II band edge alignment between CdSe and CdTe. The valence band offset between unstrained CdSe and CdTe is determined as 0.63 eV±0.06 eV by fitting the measured PL peak positions using the Kronig-Penney model. The blue shift in PL peak position is found to be primarily caused by the band bending effect based on self-consistent solutions of the Schrödinger and Poisson equations. Secondly, the design of the contact grid layout is studied to maximize the power output and energy conversion efficiency for concentrator solar cells. Because the conventional minimum power loss method used for the contact design is not accurate in determining the series resistance loss, a method of using a distributed series resistance model to maximize the power output is proposed for the contact design. It is found that the junction recombination loss in addition to the series resistance loss and shadowing loss can significantly affect the contact layout. The optimal finger spacing and maximum efficiency calculated by the two methods are close, and the differences are dependent on the series resistance and saturation currents of solar cells. Lastly, the accurate measurements of external quantum efficiency (EQE) are important for the design and development of MJ solar cells. However, the electrical and optical couplings between the subcells have caused EQE measurement artifacts. In order to interpret the measurement artifacts, DC and small signal models are built for the bias condition and the scan of chopped monochromatic light in the EQE measurements. Characterization methods are developed for the device parameters used in the models. The EQE measurement artifacts are found to be caused by the shunt and luminescence coupling effects, and can be minimized using proper voltage and light biases. Novel measurement methods using a pulse voltage bias or a pulse light bias are invented to eliminate the EQE measurement artifacts. These measurement methods are nondestructive and easy to implement. The pulse voltage bias or pulse light bias is superimposed on the conventional DC voltage and light biases, in order to control the operating points of the subcells and counterbalance the effects of shunt and luminescence coupling. The methods are demonstrated for the first time to effectively eliminate the measurement artifacts.
ContributorsLi, Jingjing (Author) / Zhang, Yong-Hang (Thesis advisor) / Tao, Meng (Committee member) / Schroder, Dieter (Committee member) / Vasileska, Dragica (Committee member) / Arizona State University (Publisher)
Created2012
Description
As the world energy demand increases, semiconductor devices with high energy conversion efficiency become more and more desirable. The energy conversion consists of two distinct processes, namely energy generation and usage. In this dissertation, novel multi-junction solar cells and light emitting diodes (LEDs) are proposed and studied for high energy conversion efficiency in both processes, respectively. The first half of this dissertation discusses the practically achievable energy conversion efficiency limit of solar cells. Since the demonstration of the Si solar cell in 1954, the performance of solar cells has been improved tremendously and recently reached 41.6% energy conversion efficiency. However, it seems rather challenging to further increase the solar cell efficiency. The state-of-the-art triple junction solar cells are analyzed to help understand the limiting factors. To address these issues, the monolithically integrated II-VI and III-V material system is proposed for solar cell applications. This material system covers the entire solar spectrum with a continuous selection of energy bandgaps and can be grown lattice matched on a GaSb substrate. Moreover, six four-junction solar cells are designed for AM0 and AM1.5D solar spectra based on this material system, and new design rules are proposed. The achievable conversion efficiencies for these designs are calculated using the commercial software package Silvaco with real material parameters. The second half of this dissertation studies the semiconductor luminescence refrigeration, which corresponds to over 100% energy usage efficiency. Although cooling has been realized in rare-earth doped glass by laser pumping, semiconductor based cooling is yet to be realized. In this work, a device structure that monolithically integrates a GaAs hemisphere with an InGaAs/GaAs quantum-well thin slab LED is proposed to realize cooling in semiconductor. The device electrical and optical performance is calculated. The proposed device then is fabricated using nine times photolithography and eight masks. The critical process steps, such as photoresist reflow and dry etch, are simulated to insure successful processing. Optical testing is done with the devices at various laser injection levels and the internal quantum efficiency, external quantum efficiency and extraction efficiency are measured.
ContributorsWu, Songnan (Author) / Zhang, Yong-Hang (Thesis advisor) / Menéndez, Jose (Committee member) / Ponce, Fernando (Committee member) / Belitsky, Andrei (Committee member) / Schroder, Dieter (Committee member) / Arizona State University (Publisher)
Created2010
Description
To extend the lifetime of complementary metal-oxide-semiconductors (CMOS), emerging process techniques are being proposed to conquer the manufacturing difficulties. New structures and materials are proposed with superior electrical properties to traditional CMOS, such as strain technology and feedback field-effect transistor (FB-FET). To continue the design success and make an impact on leading products, advanced circuit design exploration must begin concurrently with early silicon development. Therefore, an accurate and scalable model is desired to correctly capture those effects and flexible to extend to alternative process choices. For example, strain technology has been successfully integrated into CMOS fabrication to improve transistor performance but the stress is non-uniformly distributed in the channel, leading to systematic performance variations. In this dissertation, a new layout-dependent stress model is proposed as a function of layout, temperature, and other device parameters. Furthermore, a method of layout decomposition is developed to partition the layout into a set of simple patterns for model extraction. These solutions significantly reduce the complexity in stress modeling and simulation. On the other hand, semiconductor devices with self-feedback mechanisms are emerging as promising alternatives to CMOS. Fe-FET was proposed to improve the switching by integrating a ferroelectric material as gate insulator in a MOSFET structure. Under particular circumstances, ferroelectric capacitance is effectively negative, due to the negative slope of its polarization-electrical field curve. This property makes the ferroelectric layer a voltage amplifier to boost surface potential, achieving fast transition. A new threshold voltage model for Fe-FET is developed, and is further revealed that the impact of random dopant fluctuation (RDF) can be suppressed. Furthermore, through silicon via (TSV), a key technology that enables the 3D integration of chips, is studied. TSV structure is usually a cylindrical metal-oxide-semiconductors (MOS) capacitor. A piecewise capacitance model is proposed for 3D interconnect simulation. Due to the mismatch in coefficients of thermal expansion (CTE) among materials, thermal stress is observed in TSV process and impacts neighboring devices. The stress impact is investigated to support the interaction between silicon process and IC design at the early stage.
ContributorsWang, Chi-Chao (Author) / Cao, Yu (Thesis advisor) / Chakrabarti, Chaitali (Committee member) / Clark, Lawrence (Committee member) / Schroder, Dieter (Committee member) / Arizona State University (Publisher)
Created2011
Description
The object of this study is to investigate and improve the performance/stability of the flexible thin film transistors (TFTs) and to study the properties of metal oxide transparent conductive oxides for wide range of flexible electronic applications. Initially, a study has been done to improve the conductivity of ITO (indium tin oxide) films on PEN (polyethylene naphthalate) by inserting a thin layer of silver layer between two ITO layers. The multilayer with an optimum Ag mid-layer thickness, of 8 nm, exhibited excellent photopic average transmittance (~ 88 %), resistivity (~ 2.7 × 10-5 µ-cm.) and has the best Hackee figure of merit (41.0 × 10-3 Ω-1). The electrical conduction is dominated by two different scattering mechanisms depending on the thickness of the Ag mid-layer. Optical transmission is explained by scattering losses and absorption of light due to inter-band electronic transitions. A systematic study was carried out to improve the performance/stability of the TFTs on PEN. The performance and stability of a-Si:H and a-IZO (amorphous indium zinc oxide) TFTs were improved by performing a systematic low temperature (150 °C) anneals for extended times. For 96 hours annealed a-Si:H TFTs, the sub-threshold slope and off-current were reduced by a factor ~ 3 and by 2 orders of magnitude, respectively when compared to unannealed a-Si:H TFTs. For a-IZO TFTs, 48 hours of annealing is found to be the optimum time for the best performance and elevated temperature stability. These devices exhibit saturation mobility varying between 4.5-5.5 cm2/V-s, ION/IOFF ratio was 106 and a sub-threshold swing variation of 1-1.25 V/decade. An in-depth study on the mechanical and electromechanical stress response on the electrical properties of the a-IZO TFTs has also been investigated. Finally, the a-Si:H TFTs were exposed to gamma radiation to examine their radiation resistance. The interface trap density (Nit) values range from 5 to 6 × 1011 cm-2 for only electrical stress bias case. For "irradiation only" case, the Nit value increases from 5×1011 cm-2 to 2×1012 cm-2 after 3 hours of gamma radiation exposure, whereas it increases from 5×1011 cm-2 to 4×1012 cm-2 for "combined gamma and electrical stress".
ContributorsIndluru, Anil (Author) / Alford, Terry L. (Thesis advisor) / Schroder, Dieter (Committee member) / Krause, Stephen (Committee member) / Theodore, David (Committee member) / Arizona State University (Publisher)
Created2011
Description
Millimeter-wave (mmWave) and sub-terahertz (sub-THz) systems aim to utilize the large bandwidth available at these frequencies. This has the potential to enable several future applications that require high data rates, such as autonomous vehicles and digital twins. These systems, however, have several challenges that need to be addressed to realize their gains in practice. First, they need to deploy large antenna arrays and use narrow beams to guarantee sufficient receive power. Adjusting the narrow beams of the large antenna arrays incurs massive beam training overhead. Second, the sensitivity to blockages is a key challenge for mmWave and THz networks. Since these networks mainly rely on line-of-sight (LOS) links, sudden link blockages highly threaten the reliability of the networks. Further, when the LOS link is blocked, the network typically needs to hand off the user to another LOS basestation, which may incur critical time latency, especially if a search over a large codebook of narrow beams is needed. A promising way to tackle both these challenges lies in leveraging additional side information such as visual, LiDAR, radar, and position data. These sensors provide rich information about the wireless environment, which can be utilized for fast beam and blockage prediction. This dissertation presents a machine-learning framework for sensing-aided beam and blockage prediction. In particular, for beam prediction, this work proposes to utilize visual and positional data to predict the optimal beam indices. For the first time, this work investigates the sensing-aided beam prediction task in a real-world vehicle-to-infrastructure and drone communication scenario. Similarly, for blockage prediction, this dissertation proposes a multi-modal wireless communication solution that utilizes bimodal machine learning to perform proactive blockage prediction and user hand-off. Evaluations on both real-world and synthetic datasets illustrate the promising performance of the proposed solutions and highlight their potential for next-generation communication and sensing systems.
ContributorsCharan, Gouranga (Author) / Alkhateeb, Ahmed (Thesis advisor) / Chakrabarti, Chaitali (Committee member) / Turaga, Pavan (Committee member) / Michelusi, Nicolò (Committee member) / Arizona State University (Publisher)
Created2024
Description
Deep neural networks (DNN) have shown tremendous success in various cognitive tasks, such as image classification, speech recognition, etc. However, their usage on resource-constrained edge devices has been limited due to high computation and large memory requirement.
To overcome these challenges, recent works have extensively investigated model compression techniques such as element-wise sparsity, structured sparsity and quantization. While most of these works have applied these compression techniques in isolation, there have been very few studies on application of quantization and structured sparsity together on a DNN model.
This thesis co-optimizes structured sparsity and quantization constraints on DNN models during training. Specifically, it obtains optimal setting of 2-bit weight and 2-bit activation coupled with 4X structured compression by performing combined exploration of quantization and structured compression settings. The optimal DNN model achieves 50X weight memory reduction compared to floating-point uncompressed DNN. This memory saving is significant since applying only structured sparsity constraints achieves 2X memory savings and only quantization constraints achieves 16X memory savings. The algorithm has been validated on both high and low capacity DNNs and on wide-sparse and deep-sparse DNN models. Experiments demonstrated that deep-sparse DNN outperforms shallow-dense DNN with varying level of memory savings depending on DNN precision and sparsity levels. This work further proposed a Pareto-optimal approach to systematically extract optimal DNN models from a huge set of sparse and dense DNN models. The resulting 11 optimal designs were further evaluated by considering overall DNN memory which includes activation memory and weight memory. It was found that there is only a small change in the memory footprint of the optimal designs corresponding to the low sparsity DNNs. However, activation memory cannot be ignored for high sparsity DNNs.
To overcome these challenges, recent works have extensively investigated model compression techniques such as element-wise sparsity, structured sparsity and quantization. While most of these works have applied these compression techniques in isolation, there have been very few studies on application of quantization and structured sparsity together on a DNN model.
This thesis co-optimizes structured sparsity and quantization constraints on DNN models during training. Specifically, it obtains optimal setting of 2-bit weight and 2-bit activation coupled with 4X structured compression by performing combined exploration of quantization and structured compression settings. The optimal DNN model achieves 50X weight memory reduction compared to floating-point uncompressed DNN. This memory saving is significant since applying only structured sparsity constraints achieves 2X memory savings and only quantization constraints achieves 16X memory savings. The algorithm has been validated on both high and low capacity DNNs and on wide-sparse and deep-sparse DNN models. Experiments demonstrated that deep-sparse DNN outperforms shallow-dense DNN with varying level of memory savings depending on DNN precision and sparsity levels. This work further proposed a Pareto-optimal approach to systematically extract optimal DNN models from a huge set of sparse and dense DNN models. The resulting 11 optimal designs were further evaluated by considering overall DNN memory which includes activation memory and weight memory. It was found that there is only a small change in the memory footprint of the optimal designs corresponding to the low sparsity DNNs. However, activation memory cannot be ignored for high sparsity DNNs.
ContributorsSrivastava, Gaurav (Author) / Seo, Jae-Sun (Thesis advisor) / Chakrabarti, Chaitali (Committee member) / Berisha, Visar (Committee member) / Arizona State University (Publisher)
Created2018
Description
In recent years, conventional convolutional neural network (CNN) has achieved outstanding performance in image and speech processing applications. Unfortunately, the pooling operation in CNN ignores important spatial information which is an important attribute in many applications. The recently proposed capsule network retains spatial information and improves the capabilities of traditional CNN. It uses capsules to describe features in multiple dimensions and dynamic routing to increase the statistical stability of the network.
In this work, we first use capsule network for overlapping digit recognition problem. We evaluate the performance of the network with respect to recognition accuracy, convergence and training time per epoch. We show that capsule network achieves higher accuracy when training set size is small. When training set size is larger, capsule network and conventional CNN have comparable recognition accuracy. The training time per epoch for capsule network is longer than conventional CNN because of the dynamic routing algorithm. An analysis of the GPU timing shows that adjusting the capsule structure can help decrease the time complexity of the dynamic routing algorithm significantly.
Next, we design a capsule network for speech recognition, specifically, overlapping word recognition. We use both capsule network and conventional CNN to recognize 2 overlapping words in speech files created from 5 word classes. We show that capsule network achieves a considerably higher recognition accuracy (96.92%) compared to conventional CNN (85.19%). Our results show that capsule network recognizes overlapping word by recognizing each individual word in the speech. We also verify the scalability of capsule network by increasing the number of word classes from 5 to 10. Capsule network still shows a high recognition accuracy of 95.42% in case of 10 words while the accuracy of conventional CNN decreases sharply to 73.18%.
In this work, we first use capsule network for overlapping digit recognition problem. We evaluate the performance of the network with respect to recognition accuracy, convergence and training time per epoch. We show that capsule network achieves higher accuracy when training set size is small. When training set size is larger, capsule network and conventional CNN have comparable recognition accuracy. The training time per epoch for capsule network is longer than conventional CNN because of the dynamic routing algorithm. An analysis of the GPU timing shows that adjusting the capsule structure can help decrease the time complexity of the dynamic routing algorithm significantly.
Next, we design a capsule network for speech recognition, specifically, overlapping word recognition. We use both capsule network and conventional CNN to recognize 2 overlapping words in speech files created from 5 word classes. We show that capsule network achieves a considerably higher recognition accuracy (96.92%) compared to conventional CNN (85.19%). Our results show that capsule network recognizes overlapping word by recognizing each individual word in the speech. We also verify the scalability of capsule network by increasing the number of word classes from 5 to 10. Capsule network still shows a high recognition accuracy of 95.42% in case of 10 words while the accuracy of conventional CNN decreases sharply to 73.18%.
ContributorsXiong, Yan (Author) / Chakrabarti, Chaitali (Thesis advisor) / Berisha, Visar (Thesis advisor) / Weng, Yang (Committee member) / Arizona State University (Publisher)
Created2018