Search

The simultaneous-source technology for high-density seismic acquisition is a key solution to efficient seismic surveying. It is a cost-effective method when blended subsurface responses are recorded within a short time interval using multiple seismic sources. A following deblending process, however, is needed to separate signals contributed by individual sources. Recent advances in deep learning and its data-driven approach toward feature engineering have led to many new applications for a variety of seismic processing problems. It is still a challenge, though, to collect enough labeled data and avoid model overfitting and poor generalization performance over different datasets with a low resemblance from each other. In this article, we propose a novel self-supervised learning method to solve the deblending problem without labeled training datasets. Using a blind-trace deep neural network and a carefully crafted blending loss function, we demonstrate that the individual source-response pairs can be accurately separated under three different blended-acquisition designs.

Deep neural networks (DNNs) have achieved state-of-the-art performance in many important domains, including medical diagnosis, security, and autonomous driving. In domains where safety is highly critical, an erroneous decision can result in serious consequences. While a perfect prediction accuracy is not always achievable, recent work on Bayesian deep networks shows that it is possible to know when DNNs are more likely to make mistakes. Knowing what DNNs do not know is desirable to increase the safety of deep learning technology in sensitive applications; Bayesian neural networks attempt to address this challenge. Traditional approaches are computationally intractable and do not scale well to large, complex neural network architectures. In this paper, we develop a theoretical framework to approximate Bayesian inference for DNNs by imposing a Bernoulli distribution on the model weights. This method called Monte Carlo DropConnect (MC-DropConnect) gives us a tool to represent the model uncertainty with little change in the overall model structure or computational cost. We extensively validate the proposed algorithm on multiple network architectures and datasets for classification and semantic segmentation tasks. We also propose new metrics to quantify uncertainty estimates. This enables an objective comparison between MC-DropConnect and prior approaches. Our empirical results demonstrate that the proposed framework yields significant improvement in both prediction accuracy and uncertainty estimation quality compared to the state of the art.

A deep-learning-based workflow is proposed in this paper to solve the first-arrival picking problem for near-surface velocity model building. Traditional methods, such as the short-term average/long-term average method, perform poorly when the signal-to-noise ratio is low or near-surface geologic structures are complex. This challenging task is formulated as a segmentation problem accompanied by a novel postprocessing approach to identify pickings along the segmentation boundary. The workflow includes three parts: a deep U-net for segmentation, a recurrent neural network (RNN) for picking, and a weight adaptation approach to be generalized for new data sets. In particular, we have evaluated the importance of selecting a proper loss function for training the network. Instead of taking an end-to-end approach to solve the picking problem, we emphasize the performance gain obtained by using an RNN to optimize the picking. Finally, we adopt a simple transfer learning scheme and test its robustness via a weight adaptation approach to maintain the picking performance on new data sets. Our tests on synthetic data sets reveal the advantage of our workflow compared with existing deep-learning methods that focus only on segmentation performance. Our tests on field data sets illustrate that a good postprocessing picking step is essential for correcting the segmentation errors and that the overall workflow is efficient in minimizing human interventions for the first-arrival picking task.

The electron beam (e-beam) in the scanning electron microscopy (SEM) provides an appealing mobile heating source for thermal metrology with spatial resolution of ∼1 nm, but the lack of systematic quantification of the e-beam heating power limits such application development. Here, we systemically study e-beam heating in LPCVD silicon nitride (SiNx) thin-films with thickness ranging from 200 to 500 nm from both experiments and complementary Monte Carlo simulations using the CASINO software package. There is good agreement about the thickness-dependent e-beam energy absorption of thin-film between modeling predictions and experiments. Using the absorption results, we then demonstrate adapting the e-beam as a quantitative heating source by measuring the thickness-dependent thermal conductivity of SiNx thin-films, with the results validated to within 7% by a separate Joule heating experiment. The results described here will open a new avenue for using SEM e-beams as a mobile heating source for advanced nanoscale thermal metrology development.

Accurate breast lesion risk estimation can significantly reduce unnecessary biopsies and help doctors decide optimal treatment plans. Most existing computer-aided systems rely solely on mammogram features to classify breast lesions. While this approach is convenient, it does not fully exploit useful information in clinical reports to achieve the optimal performance. Would clinical features significantly improve breast lesion classification compared to using mammograms alone? How to handle missing clinical information caused by variation in medical practice? What is the best way to combine mammograms and clinical features? There is a compelling need for a systematic study to address these fundamental questions. This paper investigates several multimodal deep networks based on feature concatenation, cross-attention, and co-attention to combine mammograms and categorical clinical variables. We show that the proposed architectures significantly increase the lesion classification performance (average area under ROC curves from 0.89 to 0.94). We also evaluate the model when clinical variables are missing.

Computer-aided diagnosis (CAD) systems must constantly cope with the perpetual changes in data distribution caused by different sensing technologies, imaging protocols, and patient populations. Adapting these systems to new domains often requires significant amounts of labeled data for re-training. This process is labor-intensive and time-consuming. We propose a memory-augmented capsule network for the rapid adaptation of CAD models to new domains. It consists of a capsule network that is meant to extract feature embeddings from some high-dimensional input, and a memory-augmented task network meant to exploit its stored knowledge from the target domains. Our network is able to efficiently adapt to unseen domains using only a few annotated samples. We evaluate our method using a large-scale public lung nodule dataset (LUNA), coupled with our own collected lung nodules and incidental lung nodules datasets. When trained on the LUNA dataset, our network requires only 30 additional samples from our collected lung nodule and incidental lung nodule datasets to achieve clinically relevant performance (0.925 and 0.891 area under receiving operating characteristic curves (AUROC), respectively). This result is equivalent to using two orders of magnitude less labeled training data while achieving the same performance. We further evaluate our method by introducing heavy noise, artifacts, and adversarial attacks. Under these severe conditions, our network’s AUROC remains above 0.7 while the performance of state-of-the-art approaches reduce to chance level.

We proposed a robust segmentation and picking workflow to solve the first arrival picking problem for seismic signal processing. Unlike traditional classification algorithm, image segmentation method can utilize the location information by outputting a prediction map which has the same size of the input image. A parameter-free nearest point picking algorithm is proposed to further improve the accuracy of the first arrival picking. The algorithm is test on synthetic clean data, synthetic noisy data, synthetic picking-disconnected data and field data. It performs well on all of them and the picking deviation reaches as low as 4.8ms per receiver. The first arrival picking problem is formulated as the contour detection problem. Similar to \cite{wu2019semi}, we use U-net to perform the segmentation as it is proven to be state-of-the-art in many image segmentation tasks. Particularly, a Lovasz loss instead of the traditional cross-entropy loss is used to train the network for a better segmentation performance. Lovasz loss is a surrogate loss for Jaccard index or the so-called intersection-over-union (IoU) score, which is often one of the most used metrics for segmentation tasks. In the picking part, we use a novel nearest point picking (NPP) method to take the advantage of the coherence of the first arrival picking among adjacent receivers. Our model is tested and validated on both synthetic and field data with harmonic noises. The main contributions of this paper are as follows: 1. Used Lovasz loss to directly optimize the IoU for segmentation task. Improvement over the cross-entropy loss with regard to the segmentation accuracy is verified by the test result. 2. Proposed a nearest point picking post processing method to overcome any defects left by the segmentation output. 3. Conducted noise analysis and verified the model with both noisy synthetic and field datasets.

With the development of industrial Internet technology, more and more devices are brought into the industrial big data platform. To improve the efficiency of device maintenance, the industrial big data platform needs to monitor the abnormal data of the device. However, most of the current anomaly detection algorithms are offline and they can’t be updated in real-time. To solve this problem, this paper proposes an anomaly detection model for the industrial stream. The model realizes anomaly detection by cooperatively calling 3 σ and DBSCAN algorithm. The model has the advantages of low cost, fast speed, and easy to use. On this basis, this paper presents a real-time update strategy for this model, which further improves the accuracy of the model. The experimental results of water pump equipment monitoring data show the effectiveness of this method.

The classification of Antibody Mediated Rejection (AMR) in kidney transplant remains challenging even for experienced nephropathologists; this is partly because histological tissue stain analysis is often characterized by low inter-observer agreement and poor reproducibility. One of the implicated causes for inter-observer disagreement is the variability of tissue stain quality between (and within) pathology labs, coupled with the gradual fading of archival sections. Variations in stain colors and intensities can make tissue evaluation difficult for pathologists, ultimately affecting their ability to describe relevant morphological features. Being able to accurately predict the AMR status based on kidney histology images is crucial for improving patient treatment and care. We propose a novel pipeline to build robust deep neural networks for AMR classification based on StyPath, a histological data augmentation technique that leverages a light weight style-transfer algorithm as a means to reduce sample-specific bias. Each image was generated in \(1.84 \pm 0.03\) s using a single GTX TITAN V gpu and pytorch, making it faster than other popular histological data augmentation techniques. We evaluated our model using a Monte Carlo (MC) estimate of Bayesian performance and generate an epistemic measure of uncertainty to compare both the baseline and StyPath augmented models. We also generated Grad-CAM representations of the results which were assessed by an experienced nephropathologist; we used this qualitative analysis to elucidate on the assumptions being made by each model. Our results imply that our style-transfer augmentation technique improves histological classification performance (reducing error from 14.8% to 11.5%) and generalization ability.

Capsule Networks (CapsNets) have demonstrated to be a promising alternative to Convolutional Neural Networks (CNNs). However, they often fall short of state-of-the-art accuracies on large-scale high-dimensional datasets. We propose a Detail-Oriented Capsule Network (DECAPS) that combines the strength of CapsNets with several novel techniques to boost its classification accuracies. First, DECAPS uses an Inverted Dynamic Routing (IDR) mechanism to group lower-level capsules into heads before sending them to higher-level capsules. This strategy enables capsules to selectively attend to small but informative details within the data which may be lost during pooling operations in CNNs. Second, DECAPS employs a Peekaboo training procedure, which encourages the network to focus on fine-grained information through a second-level attention scheme. Finally, the distillation process improves the robustness of DECAPS by averaging over the original and attended image region predictions. We provide extensive experiments on the CheXpert and RSNA Pneumonia datasets to validate the effectiveness of DECAPS. Our networks achieve state-of-the-art accuracies not only in classification (increasing the average area under ROC curves from 87.24% to 92.82% on the CheXpert dataset) but also in the weakly-supervised localization of diseased areas (increasing average precision from 41.7% to 80% for the RSNA Pneumonia detection dataset).

Deep Neural Networks (or DNNs) must constantly cope with distribution changes in the input data when the task of interest or the data collection protocol changes. Retraining a network from scratch to combat this issue poses a significant cost. Meta-learning aims to deliver an adaptive model that is sensitive to these underlying distribution changes, but requires many tasks during the meta-training process. In this paper, we propose a tAsk-auGmented actIve meta-LEarning (AGILE) method to efficiently adapt DNNs to new tasks by using a small number of training examples. AGILE combines a meta-learning algorithm with a novel task augmentation technique which we use to generate an initial adaptive model. It then uses Bayesian dropout uncertainty estimates to actively select the most difficult samples when updating the model to a new task. This allows AGILE to learn with fewer tasks and a few informative samples, achieving high performance with a limited dataset. We perform our experiments using the brain cell classification task and compare the results to a plain meta-learning model trained from scratch. We show that the proposed task-augmented meta-learning framework can learn to classify new cell types after a single gradient step with a limited number of training samples. We show that active learning with Bayesian uncertainty can further improve the performance when the number of training samples is extremely small. Using only 1% of the training data and a single update step, we achieved 90% accuracy on the new cell type classification task, a 50% points improvement over a state-of-the-art meta-learning algorithm.

Pyrolytic carbon (PyC) is an important material used in many applications including thermal management of electronic devices and structural stability of ceramic composites. Accurate measurement of physical properties of structures containing textured PyC layers with few-micrometer thickness poses new challenges. Here a laser-based thermoreflectance technique is used to measure thermal conductivity in a 30-μm-thick textured PyC layer deposited using chemical vapor deposition on the surface of spherical zirconia particles. Raman spectroscopy is used to confirm the graphitic nature and characterize microstructure of the deposited layer. Room temperature radial and circumferential thermal conductivities are found to be 0.28 W m−1 K−1 and 11.5 W m−1 K−1, corresponding to cross-plane and in-plane conductivities of graphite. While the anisotropic ratio of the in-plane to cross-plane conductivities is smaller than previous results, the magnitude of the smallest conductivity is noticeably smaller than previously reported values for carbon materials and offers opportunities in thermal management applications. Very low in-plane and cross-plane thermal conductivities are attributed to strong grain boundary scattering, high defect concentration, and small inter-laminar porosity. Experimental results agree with the prediction of thermal transport model informed by the microstructure information revealed by Raman spectroscopy.

Recent first-principles modeling reported a decrease of in-plane thermal conductivity (k) with increased thickness for few layered MoS2, which results from the change in phonon dispersion and missing symmetry in the anharmonic atomic force constant. For other 2D materials, it has been well documented that a higher thickness could cause a higher in-plane k due to a lower density of surface disorder. However, the effect of thickness on the k of MoS2 has not been systematically uncovered by experiments. In addition, from either experimental or theoretical approaches, the in-plane k value of tens-of-nm-thick MoS2 is still missing, which makes the physics on the thickness-dependent k remain ambiguous. In this work, we measure the k of few-layered (FL) MoS2 with thickness spanning a large range: 2.4 nm to 37.8 nm. A novel five energy transport state-resolved Raman (ET-Raman) method is developed for the measurement. For the first time, the critical effects of hot carrier diffusion, electron-hole recombination, and energy coupling with phonons are taken into consideration when determining the k of FL MoS2. By eliminating the use of laser energy absorption data and Raman temperature calibration, unprecedented data confidence is achieved. A nonmonotonic thickness-dependent k trend is discovered. k decreases from 60.3 W m-1 K-1 (2.4 nm thick) to 31.0 W m-1 K-1 (9.2 nm thick), and then increases to 76.2 W m-1 K-1 (37.8 nm thick), which is close to the reported k of bulk MoS2. This nonmonotonic behavior is analyzed in detail and attributed to the change of phonon dispersion for very thin MoS2 and a reduced surface scattering effect for thicker samples.

Steady state Raman spectroscopy is the most widely used opto-thermal technique for measuring a 2D atomic-layer material's thermal conductivity. It requires the calibration of temperature coefficients of Raman properties and measurement/calculation of the absolution laser absorption in 2D materials. Such a requirement is very laborious and introduces very large measurement errors (of the order of 100%) and hinders gaining a precise and deep understanding of phonon–structure interactions in 2D materials. In this work, a novel nanosecond energy transport state resolved Raman (ns ET-Raman) technique is developed to resolve these critical issues and achieve unprecedented measurement precision, accuracy and ease of implementation. In ns ET-Raman, two energy transport states are constructed: steady state and nanosecond thermal transport and Raman probing. The ratio of the temperature rise under the two states eliminates the need for Raman temperature calibration and laser absorption evaluation. Four suspended MoS2 (45–115 nm thick) and four suspended MoSe2 (45–140 nm thick) samples are measured and compared using ns ET-Raman. With the increase of the sample thickness, the measured thermal conductivity increases from 40.0 ± 2.2 to 74.3 ± 3.2 W m−1 K−1 for MoS2, and from 11.1 ± 0.4 to 20.3 ± 0.9 W m−1 K−1 for MoSe2. This is attributed to the decreased significance of surface phonon scattering in thicker samples. The ns ET-Raman features the most advanced capability to measure the thermal conductivity of 2D materials and will find broad applications in studying low-dimensional materials.

Lung cancer is the leading cause of cancer death among both men and women. The solitary pulmonary nodule (SPN) is frequently seen on radiographs and computed tomography (CT) and often provokes additional clinical and imaging activities as an SPN alerts possible early stage lung cancer. However, reliable diagnosis of malignant lung nodules in the workup of SPN is challenging, making it difficult for early stage cancer management to avoid morbidity due to malignant cancer, patient distress, and increased costs caused by more invasive and unwarranted procedures for benign cases. We developed a deep learning model based on convolutional neural networks (CNNs) to predict lung cancer risk for incidental SPN. The model can help better management of incidental lung cancer by providing a reliable characterization of SPNs. Deep learning models currently are applied to nodule detection in low-dose CT scans of lung cancer screening for at-risk population. Our work aims to investigate whether deep learning can differentiate benign from malignant incidental SPN during normal dose CT exams, where the number of patients with SPN is an order more than that of lung screening population. We collected an incidental SPN dataset containing 139 CT scans to generate the nodule malignancy prediction model using 3D CNN. This dataset was collected from either contrast or non-contrast CT scans with regular doses. All 139 patients went through surgeries to remove the tumors. CT images were collected within one year before the surgery. 90 cases are malignant and the rest are benign. Ground truth labels were obtained from biopsies. Data were preprocessed by resampling, normalization, and cropping to unique volume around the SPN. The radiologists annotated the SPN location. Since the two nodule classes are imbalanced, we did online data augmentation to increase the variety of lung nodule images and balance the two classes. We used 80% of the data for training and 20% for testing. The deep learning structure adapted is a 3D Resnet with 68 layers and 33M parameters. The area under the receiving operating characteristics curve (AUC) of the prediction result was 0.81 while the accuracy was 79.6%. Applying GradCam to visualize important regions in CT scans contributing to the final predictions, we found that our trained ResNet model focuses on meaningful areas in lung nodules and detects malignant SPNs in most cases. Our preliminary results shows the feasibility of using deep learning to predict incidental lung nodules and improve diagnostic speed and accuracy. Citation Format: Pengyu Yuan, Tiancheng He, Hien Nguyen, Stephen T. Wong. A deep learning model-based lung cancer risk assessment for incidental pulmonary nodules [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2021; 2021 Apr 10-15 and May 17-21. Philadelphia (PA): AACR; Cancer Res 2021;81(13_Suppl):Abstract nr 2614.

Quantitative understanding of 2D atomic layer interface thermal resistance (R) based on Raman characterization is significantly hindered by unknown sample-to-sample optical properties variation, interface-induced optical interference, off-normal laser irradiation, and large thermal-Raman calibration uncertainties. In this work, we develop a novel energy transport state resolved Raman (ET-Raman) to resolve these critical issues, and also consider the hot carrier diffusion, which is crucial but has been rarely considered during interface energy transport study. In ET-Raman, by constructing two steady heat conduction states with different laser spot sizes, we differentiate the effect of R and hot carrier diffusion coefficient (D). By constructing an extreme state of zero/negligible heat conduction using a picosecond laser, we differentiate the effect of R and material’s specific heat. In the end, we precisely determine R and D without need of laser absorption and temperature rise of the 2D atomic layer. Seve...

We report a novel approach for non-contact simultaneous determination of the hot carrier diffusion coefficient (D) and interface thermal resistance (R) of sub-10 nm virgin mechanically exfoliated MoS2 nanosheets on c-Si. The effect of hot carrier diffusion in heat conduction by photon excitation, diffusion, and recombination is identified by varying the heating spot size from 0.294 μm to 1.14 μm (radius) and probing the local temperature rise using Raman spectroscopy. R is determined as 4.46–7.66 × 10−8 K m2 W−1, indicating excellent contact between MoS2 and c-Si. D is determined to be 1.18+0.30−0.23, 1.07+0.37−0.26, 1.20+0.34−0.27 and 1.62+0.30−0.23 cm2 s−1 for 3.6 nm, 5.4 nm, 8.4 nm, and 9.0 nm thick MoS2 samples, showing little dependence on the thickness. The hot carrier diffusion length (LD) can be determined without knowledge of the hot carrier's life-time. The four samples LD is determined as 0.344+0.041−0.036 (3.6 nm), 0.327+0.052−0.043 (5.4 nm), 0.346+0.046−0.042 (8.4 nm), and 0.402+0.036−0.030 μm (9.0 nm). Unlike previous methods that are implemented by making electrical contact and applying an electric field for D measurement, our technique has the advantage of being truly non-contact and non-invasive, and is able to characterize the electron diffusion behavior of virgin 2D materials. Also it points out that hot carrier diffusion needs to be taken into serious consideration in Raman-based thermal property characterization of 2D materials, especially under very tightly focused laser heating whose spot size is comparable to the hot carrier diffusion length.

We report a systematic investigation of interfacial thermal conductance (Gk) between few to tens -layered mechanical exfoliated molybdenum disulfide (MoS2) and crystalline silicon (c-Si). Based on Raman spectroscopy, we find Gk at room temperature increases with increased layer numbers of MoS2 from 0.974 MW m−2 K−1 to 68.6 MW m−2 K−1. The higher Gk of thicker samples reveals their better interface contact with the substrate, leading to accordingly improved interfacial energy coupling. Molecular dynamics (MD) simulations are conducted to interpret and compare with the experimental observations. MD simulations predict a thermal conductance in the range of 53–77 MW m−2 K−1, which agrees well with the upper bound Gk measured in our work. The thickness dependence of measured Gk reflects the improved interface spacing for thicker MoS2 samples. This phenomenon is further confirmed by the Raman intensity enhancement study by the interface spacing and local optical interference calculations.

Organic thermoelectric materials and generators are a promising field of research due to numerous attractive characteristics that suggest the ability to overcome the shortcomings of existing inorganic TE systems. This book aims to reprise some of the key directions in this field and provide some context on the key developments and opportunities that remain to be explored. Therefore, to facilitate comprehension of the book, this introduction chapter will present a brief overview of essential topics spanning the fields of chemistry, materials science, engineering, and physics, including a brief historical overview of organic TE materials. TE measurement techniques, including carrier-energy and thermal transport in organic TE systems, will also be briefly discussed, as they are relevant to understanding what challenges remain in understanding and optimizing the performance of these organic TE materials.

We report a novel approach for non-contact, simultaneous determination of hot carrier diffusion coefficient (D) and interface thermal resistance (R) of sub-10 nm virgin mechanically exfoliated MoS2 nanosheets on c-Si. The effect of hot carrier diffusion in heat conduction by photon excitation, diffusion, and recombination is identified by varying the heating spot size from 0.294 μm to 1.14 μm (radius) and probing the local temperature rise using Raman spectroscopy. R is determined as 4.46~7.66×10-8 K·m2/W, indicating excellent contact between MoS2 and c-Si. D is determined to be 1.07~1.62 cm2/s, showing little dependence on the thickness. Unlike previous methods of making electrical contacts and applying an electric field for D measurement, our technique has the advantage of being truly non-contact and noninvasive, and is able to characterize the electron diffusion behavior of virgin 2D materials. Also, it points out that hot carrier diffusion needs to be taken into serious consideration in Raman-based thermal properties characterization of 2D materials, especially under very tightly focused laser heating whose spot size is comparable to the hot carrier diffusion length.

This work reports on the discovery of a high thermal conductivity (κ) switch-on phenomenon in high purity graphene paper (GP) when its temperature is reduced from room temperature down to 10 K. The κ after switch-on (1732 to 3013 W m−1 K−1) is 4–8 times that before switch-on. The triggering temperature is 245–260 K. The switch-on behavior is attributed to the thermal expansion mismatch between pure graphene flakes and impurity-embedded flakes. This is confirmed by the switch behavior of the temperature coefficient of resistance. Before switch-on, the interactions between pure graphene flakes and surrounding impurity-embedded flakes efficiently suppress phonon transport in GP. After switch-on, the structure separation frees the pure graphene flakes from the impurity-embedded neighbors, leading to a several-fold κ increase. The measured κ before and after switch-on is consistent with the literature reported κ values of supported and suspended graphene. By conducting comparison studies with pyrolytic graphite, graphene oxide paper and partly reduced graphene paper, the whole physical picture is illustrated clearly. The thermal expansion induced switch-on is feasible only for high purity GP materials. This finding points out a novel way to switch on/off the thermal conductivity of graphene paper based on substrate-phonon scattering.

The thermal conductivity (k) of supported graphene is a critical property that reflects the graphene-substrate interaction, graphene structure quality, and is needed for thermal design of a graphene device. Yet the related k measurement has never been a trivial work and very few studies are reported to date, only at the μm level. In this work, for the first time, the k of giant chemical vapor decomposition (CVD) graphene supported on poly(methyl methacrylate) (PMMA) is characterized using our transient electro-thermal technique based on a differential concept. Our graphene size is ∼mm, far above the samples studied in the past. This giant graphene measurement eliminates the thermal contact resistance problems and edge phonon scattering encountered in μm-scale graphene k measurement. Such mm-scale measurement is critical for device/system-level thermal design since it reflects the effect of abundant grains in graphene. The k of 1.33-layered, 1.53-layered, 2.74-layered and 5.2-layered supported graphene is measured as 365 W m(-1) K(-1), 359 W m(-1) K(-1), 273 W m(-1) K(-1) and 33.5 W m(-1) K(-1), respectively. These values are significantly lower than the k of supported graphene on SiO2, and are about one order of magnitude lower than the k of suspended graphene. We speculate that the abundant C atoms in the PMMA promote more ready energy and momentum exchange with the supported graphene, and give rise to more phonon scattering than the SiO2 substrate. This leads to a lower k of CVD graphene on PMMA than that on SiO2. We attribute the existence of disorder in the sp(2) domain, graphene oxide (GO) and stratification in the 5.2-layered graphene to its more k reduction. The Raman linewidth (G peak) of the 5.2-layered graphene is also twice larger than that of the other three kinds of graphene, indicating the much more phonon scattering and shorter phonon lifetime in it. Also the electrical conductivity of the 5.2-layered graphene is about one-fifth of that for the other three. This further confirms the poor graphene quality of sample 4S, explaining its much lower k.

Recent first-principles modeling reported a decrease of in-plane thermal conductivity (k) with increased thickness for few layered MoS2, which results from the change in phonon dispersion and missing symmetry in the anharmonic atomic force constant. For other 2D materials, it has been well documented that a higher thickness could cause a higher in-plane k due to a lower density of surface disorder. However, the effect of thickness on the k of MoS2 has not been systematically uncovered by experiments. In addition, from either experimental or theoretical approaches, the in-plane k value of tens-of-nm-thick MoS2 is still missing, which makes the physics on the thickness-dependent k remain ambiguous. In this work, we measure the k of few-layered (FL) MoS2 with thickness spanning a large range: 2.4 nm to 37.8 nm. A novel five energy transport state-resolved Raman (ET-Raman) method is developed for the measurement. For the first time, the critical effects of hot carrier diffusion, electron-hole recombination, and energy coupling with phonons are taken into consideration when determining the k of FL MoS2. By eliminating the use of laser energy absorption data and Raman temperature calibration, unprecedented data confidence is achieved. A nonmonotonic thickness-dependent k trend is discovered. k decreases from 60.3 W m-1 K-1 (2.4 nm thick) to 31.0 W m-1 K-1 (9.2 nm thick), and then increases to 76.2 W m-1 K-1 (37.8 nm thick), which is close to the reported k of bulk MoS2. This nonmonotonic behavior is analyzed in detail and attributed to the change of phonon dispersion for very thin MoS2 and a reduced surface scattering effect for thicker samples.