ABSTRACT. This talk focuses on various EDA tools used in design of digital systems at various levels of abstraction. In general, design automation tools are used for design verification, optimization, automatic hardware generation, test and testability analysis, and dependability and security analysis. At more abstract design levels, design automation tools are merely used for suggesting high-level design decisions such as the type of processing or hardware-software partitioning in a DSE (design space exploration) environment. On the other hand, EDA tools provide detailed analysis and hardware optimization and placement at the lower abstraction levels. Artificial Intelligence (AI) has made significant improvements in new generation of EDA tools. At the abstract system level, AI focuses on prediction of specifics of the implemented hardware based on design requirements and constraints. Whereas, at the lower-levels of abstraction, AI-based tools provide more efficient decisions on hardware implementation and optimization. After a presentation of a hierarchical design flow from system description to hardware implementation, this talk will discuss tools and environments used at various abstraction levels. The role of AI in such tools will be discussed.

ABSTRACT. Greenhouse gases are one of the most significant problems created by humanity to improve living standards. Along with the industrial revolution, the growth of cities around the world is increasing with an increase in vehicle use, industrialization and corresponding shortcomings in planning and environmental regulations. Measurement of greenhouse gas emissions is vital for tackling global warming, assessing carbon budgets, improving climate models, and assessing climate feedbacks of natural emissions. In this study a greenhouse gas and meteorology measurement station is developed to monitor ozone, methane, ammonia, nitrogen dioxide, chlorine, smoke, carbon monoxide, temperature, wind and humidity in Ankara, Turkey. Station is developed using low-cost sensors and IOT platform. Station is proved to be reliable and efficient. Measurement data were stored between 17.01.2021-02.05.2021, which includes the full closure bans due to the pandemic and the processes in which the bans were lifted, and the analysis results are presented in this study. This dataset can be downloaded from https://doi.pangaea.de/10.1594/PANGAEA.938486 [1].

ABSTRACT. This paper is devoted to the study of the application of machine learning (ML) algorithms for power prediction at floorplan stage of physical design of integrated circuits (ICs). Reduction of the power consumption can lead to additional iterations in the design flow since a realistic calculation of power is possible at the final stages. Unlike the existing works, the proposed method considers the variations of process, supply voltage, temperature, threshold voltage, frequency, and floorplan related parameters. For estimation of the proposed method, the circuits from IWLS05 benchmark and 8 ML algorithms were used. The mean absolute percentage error is approximately 13% for unseen designs with mentioned parameters’ variations. The disadvantage of the method is the necessity to go through the full design flow of the selected designs with a selected range of parameters to collect data necessary for training ML models, which requires additional machine and time resources.

ABSTRACT. Wireless sensor networks (WSNs) are self-organizing networks consisting of many wireless sensor nodes distributed in space and intended for monitoring environmental characteristics and managing objects located in it. A self-organized network is a network in which the number of nodes is a random value at each moment of time, varying from 0 to some value Nmax. Connections that occur between nodes in such networks are also random in time because they are formed to achieve any goal or transfer data to a public communication network or to other types of networks. The paper is devoted to the investigation of the latest innovative type of the particular case of WSNs: the so-called WBSM (Wireless Body sensor network). It is based on service-oriented architecture (SOA), uses additional fog and edge computing networks as middleware and implements blockchain technologies to ensure the confidentiality of data received and processed. The proposed solutions were tested by creating a prototype of a mobile medical intelligent platform WBSM, which enables remote interaction of patients and doctors, which is especially critical in the conditions of a pandemic to identify risk groups of patients with chronic diseases through timely round-the-clock medical diagnostics. network.

ABSTRACT. In this study, various experiments were carried out for fire and smoke detection by using YOLOv5 and YOLOv7 pre-trained weights. As a result of the experiments, the performances of the models were evaluated separately and compared with each other. A total of 7 weights, including small, medium and large, were used for pre-trained models. In the experiments with YOLOv5, the model with the best mAP.5 score was yolov5m with 0.79, while the model with the best mAP.5 score in the YOLOv7 models was yolov7 with 0.78. Since the YOLOv7 model performed worse than expected, a hyper-parameter search was performed for yolov7 using Genetic Algorithm (GA). After hyper-parameter search with yolov7, the precision and mAP.5 value of the model increased by 0.01, while the recall value increased by 0.02. It has been observed that tuned hyper-parameters strong claim the default hyper-parameters for fire-smoke detection.

ABSTRACT. In this study, an alternative data sampling method is proposed for deep learning models. In the suggested method, 3-stage sampling is done by using sequential frames. In the sampling process, a 5-minute 30 fps video is re-sampled to 5 fps and a total of 1500 frames are obtained. Optical flow and random sampling operations are performed sequentially on the frames obtained. In optical flow methods, 4 different thresholds are used while determining the magnitude vector. When performing random sampling, a total of 4 different data sets were created with 3 different sample rates. As a result of experiments with different sample rates and thresholds, a maximum of 1459 and a minimum of 757 frames are obtained from 1500 frames. The number of data obtained for each sub-data set is different. Obtained sub-data sets can be used for model generalization according to the needs of the environment in which the deep learning model will be used. To evaluate the performance of a deep learning model from different perspectives, experiments were carried out with the obtained data sets. As a result of the experiments, it was seen that the learning model cannot be directly associated with the amount of data. According to different evaluation metrics, it has been evaluated which data set is more suitable for various environments.

ABSTRACT. In the context of smart cities, real-time vehicle speed detection is a research area that can be examined under the categories of traditional and hybrid methods. Traditional methods encompass existing Radar and Lidar studies, while hybrid methods can be explored within the framework of artificial intelligence-supported systems. In this study, we propose a hybrid method that combines various vehicle speed detection techniques available in the literature. Our proposed method for determining road capacity and individual vehicle speeds in smart cities comprises three stages. In the first stage, we employ the YOLOv7 algorithm for vehicle detection, and in the second stage, we use the Centroid Tracker for tracking the detected vehicles. In the final stage, we calculate the displacement of each tracked vehicle within the frame through depth estimation, utilizing the Adabins algorithm. Additionally, we apply the 'Triangle Method' from the literature to find the displacement of each tracked vehicle one second after depth estimation. To evaluate our approach, we conducted sequential experiments using a data set of 6000 images, which were subdivided into training (85\%), validation (10\%), and testing (5\%) sets. Our trained model recognizes two classes: 'car' and 'truck,' achieving precision rates exceeding 90\% for each class.

ABSTRACT. The proposed paper analyses the modern approaches to decision-making, based on reinforcement learning with practical implementation of Q-learning for tasks of robotics. Theoretical aspects of reinforcement learning are considered in comparison to other decision-making technologies. From practical point, Q-learning is proposed to use for robot’s actions observation, for reduction of decision-making process duration and as a decision-making model for mobile robot movements inside complex workspace. Materials of paper include the results of numerical experiments, which simulate workspaces of robot with different levels of complexity for paths with various starting points.

ABSTRACT. The Monte Carlo Method was used to calculate the coverage factors for the composition of two Student's distribution laws with a different number of degrees of freedom and the ratio of standard deviations. Approximating expressions for the coverage factor of this composition are given. The relative errors of applying these approximating expressions are calculated. The choice of an expression that provides the most accurate approximation of the obtained numerical values is substantiated. A technique for finding the number of degrees of freedom of the resulting composition is proposed.

ABSTRACT. In this study, experiments were conducted on obtaining depth maps from RGB images using deep learning methods. Depth maps were obtained from indoor and outdoor images using AdaBins, MiDas, and ZoeDepth methods. When using AdaBins, versions trained with NYU and KITTI datasets were used, and when using MiDaS, versions dpt_swin2_tiny_256, dpt_swin2_large_384, and dpt_beit_large_512 were used, resulting in a total of 6 comparisons. Since the images are real-world images, there is no ground-truth to evaluate the results. Therefore, the results were evaluated by two different researchers using observational methods. The results showed that XXX from the MiDaS models was the most optimal method compared to other methods.

ABSTRACT. A design methodology for aging aware foldback current limiting is presented. Traditional constant and foldback limiting techniques are reviewed in the context of hot carrier induced degradation. Aging analysis is performed in 14nm finFET process and the results are used to design an LDO with foldback operating envelope that avoids high stress conditions entirely. Simulation results confirm the functionality and effectiveness of the proposed design.

ABSTRACT. The correctness and stability of distributed data storage systems is one of the key and, at the same time, special conditions for the overall performance of a wide variety of domain-oriented networked computer systems. In this regard, information-dynamic aspects of ensuring the reliability of the functioning of distributed data storage systems based on analysis, verification and prediction, including statistical ones, of their states and behavior are of interest. The present paper discusses the development of a verification model for distributed data storage systems based on a state diagram (machine), statistical analysis and behavioral automaton experiments, which has the features of identifying states and neighborhoods of behavior, behavior recognition, using logistic regression and Bayes formulas, in particular, for machine learning. The verification model makes it possible to increase the completeness and accuracy of analysis to ln(|S|2*|L|), verification and forecasting, and the methods based on it reduce the computational complexity to ln(|S|!) and increase the efficiency of predicting the reliable functioning of distributed data storage systems.

ABSTRACT. In this paper a complete flow, from design toward ASIC implementation of a fully synthesizable 32-bit microcontroller in 180nm CMOS technology is presented. This microcontroller, referred to as HIRMA, features the open-source RISC-V IM processor mounted through customized busses for communication processes. The microcontroller contains a 4kB-SRAM, SPI flash controller for inserting instructions from external flash to SRAM, a UART transmitter and receiver module, a 32-bit timer, and capability of external off-chip accelerators. All peripherals are controlled by a RISC-V processor, and an SPI master interface that is used for programming the SRAM in the system. In this paper, we exhibit the design of our proposed microcontroller and present our design flow from Register-Transfer Level design to generating an ASIC layout. An affordable and easy to implement platform for post-manufacturing testing is also introduced. A total power density is reported as 10.7091mW in 50 MHz and the area for this RISC-V microcontroller has a reduced footprint of 1mm × 2mm including I/O pad modules.

ABSTRACT. The growing complexity of SoCs requires more accurate and faster test and verification in the early ‎stages of the design. Fault injection plays an important role in ensuring safety, security, and fault-‎tolerance system specifications are met at various stages of the design. The SystemC language ‎standard provides the possibility of system-level design modeling and early evaluation in the design ‎flow. In this paper, we propose a simulation-based multi-level fault injection framework in ‎SystemC. Our framework makes use of Universal Verification Methodology (UVM), a well-known ‎language independent standard developed to unify the verification flow, to systematize the use and ‎reuse of the test sequences for different test scenarios. To evaluate our framework, we study the ‎impact of faults on a RISCV-like processor described in three levels of abstraction (ISS, RTL, and ‎Gate-level) for both permanent and transient faults.‎