Thermal analysis is increasingly critical in modern integrated circuits, where non-uniform power dissipation and high transistor densities can cause rapid temperature spikes and reliability concerns. Traditional methods, such as FEM-based simulations, offer high accuracy but are computationally prohibitive for early-stage design, often requiring multiple iterative redesign cycles to resolve late-stage thermal failures. To address these challenges, we propose ‘2D-ThermAl’, a physics-informed generative AI framework that effectively identifies heat sources and estimates full-chip transient and steady-state thermal distributions directly from input activity profiles. ThermAl employs a hybrid U-Net architecture enhanced with positional encoding and a Boltzmann regularizer to maintain physical fidelity. Our model is trained on an extensive dataset of heat dissipation maps for over 200 circuit configurations, ranging from simple logic gates (e.g., inverters, NAND, XOR) to complex designs, generated using COMSOL and Cadence EDA flows. The dataset captures diverse activity patterns, and we note that material-dependent thermal properties may require targeted fine-tuning to ensure accuracy across different fabrication contexts. Experimental results demonstrate that 2D-ThermAl provides precise temperature mappings for large circuits, with a root mean squared error (RMSE) of only 0.71°C and outperforms conventional FEM tools by running up to ∼ 200× faster. We analyze performance across diverse layouts and workloads and discuss its applicability to large-scale EDA workflows. Although thermal reliability assessments often extend beyond 85°C for post-layout signoff, our focus here is on early-stage hotspot detection and thermal pattern learning. To ensure generalization beyond the nominal operating range (25−55°C), we additionally performed cross-validation on an extended dataset that spans 25−95°C, maintaining a high accuracy (< 2.2% full-scale RMSE) even under elevated temperature conditions representative of the peak power and stress scenarios. Limitations such as 2D-only modeling and real-world validation are addressed with concrete future directions, including 3D extension, generalization across technology nodes, and transfer learning strategies. The code and dataset are publicly available at: https://github.com/soumyadeepchandra/2D-ThermAl.

Spiking neural networks (SNNs) offer a promising alternative to traditional analog neural networks (ANNs), especially for sequential tasks, with enhanced energy efficiency. The internal memory in SNNs obtained through the membrane potential equips them with innate lightweight temporal processing capabilities. However, the unique advantages of this temporal dimension of SNN s have not yet been effectively harnessed. To that end, this article delves deeper into the what, why and where of SNNs. By considering event-based optical flow as an exemplary task in vision-based navigation, we highlight that the true potential of SNNs lies in sequential tasks. The event-driven recurrent dynamics of a spiking neuron merged harmoniously with event camera inputs enables SNNs to outperform corresponding ANNs with a lower number of parameters for optical flow. Furthermore, we demonstrate that SNNs can be synergistically combined with ANNs to form SNN-ANN hybrids to obtain the best of both worlds in terms of accuracy, energy, memory, and training efficiency. Additionally’ the emergence of various near-memory and in-memory computing techniques has propelled efficient implementation of these approaches. Overall, the immediate future of SNNs looks exciting, as we discover the niche of SNN s, comprising sequential tasks with low power requirements.

Procedure learning refers to the task of identifying the key-steps and determining their logical order, given several videos of the same task. For both third-person and first-person (egocentric) videos, state-of-the-art (SOTA) methods aim at finding correspondences across videos in time to accomplish procedure learning. However, to establish temporal relationships within the sequences, these methods often rely on frame-to-frame mapping, or assume monotonic alignment of video pairs, leading to sub-optimal results. To this end, we propose to treat the video frames as samples from an unknown distribution, enabling us to frame their distance calculation as an optimal transport (OT) problem. Notably, the OT-based formulation allows us to relax the previously mentioned assumptions. To further improve performance, we enhance the OT formulation by introducing two regularization terms. The first, inverse difference moment regularization, promotes transportation between instances that are homogeneous in the embedding space as well as being temporally closer. The second, regularization based on the KL-divergence with an exponentially decaying prior smooths the alignment while enforcing conformity to the optimality (alignment obtained from vanilla OT optimization) and temporal priors. The resultant optimal transport guided procedure learning framework (`OPEL') significantly outperforms the SOTA on benchmark datasets. Specifically, we achieve 22.4\% (IoU) and 26.9\% (F1) average improvement compared to the current SOTA on large scale egocentric benchmark, EgoProceL. Furthermore, for the third person benchmarks (ProCeL and CrossTask), the proposed approach obtains 46.2\% (F1) average enhancement over SOTA.

Spiking Neural Networks (SNNs) offer a promising alternative to traditional deep learning, since they provide higher computational efficiency due to event-driven information processing. SNNs distribute the analog values of pixel intensities into binary spikes over time. However, the most widely used input coding schemes, such as Poisson based rate-coding, do not leverage the additional temporal learning capability of SNNs effectively. Moreover, these SNNs suffer from high inference latency which is a major bottleneck to their deployment. To overcome this, we propose a time-based encoding scheme that utilizes Discrete Cosine Transform (DCT) to reduce the number of timesteps required for inference (DCT-SNN). DCT decomposes an image into a weighted sum of sinusoidal basis images. At each time step, a single frequency base, taken in order and modulated by its corresponding DCT coefficient, is input to an accumulator that generates spikes upon crossing a threshold. We use the proposed scheme to train DCT-SNN, a low-latency deep SNN with leaky-integrate-and-fire neurons using surrogate gradient descent based backpropagation. We achieve top-1 accuracy of 89.94%, 68.30% and 52.43% on CIFAR10, CIFAR-100 and TinyImageNet, respectively using VGG architectures. Notably, DCT-SNN performs inference with 2-14X reduced latency compared to other state-of-the-art SNNs, while achieving comparable accuracy to their standard deep learning counterparts. The dimension of the transform allows us to control the number of timesteps required for inference. Additionally, we can trade-off accuracy with latency in a principled manner by dropping the highest frequency components during inference. The code is publicly available.1

Spiking Neural Networks (SNNs) are a promising alternative to traditional deep learning methods since they perform event-driven information processing. However, a major drawback of SNNs is high inference latency. The efficiency of SNNs could be enhanced using compression methods such as pruning and quantization. Notably, SNNs, unlike their non-spiking counterparts, consist of a temporal dimension, the compression of which can lead to latency reduction. In this paper, we propose spatial and temporal pruning of SNNs. First, structured spatial pruning is performed by determining the layer-wise significant dimensions using principal component analysis of the average accumulated membrane potential of the neurons. This step leads to 10-14X model compression. Additionally, it enables inference with lower latency and decreases the spike count per inference. To further reduce latency, temporal pruning is performed by gradually reducing the timesteps while training. The networks are trained using surrogate gradient descent based backpropagation and we validate the results on CIFAR10 and CIFAR100, using VGG architectures. The spatiotemporally pruned SNNs achieve 89.04% and 66.4% accuracy on CIFAR10 and CIFAR100, respectively, while performing inference with 3-30X reduced latency compared to state-of-the-art SNNs. Moreover, they require 8-14X lesser compute energy compared to their unpruned standard deep learning counterparts. The energy numbers are obtained by multiplying the number of operations with energy per operation. These SNNs also provide 1–4% higher robustness against Gaussian noise corrupted inputs. Furthermore, we perform weight quantization and find that performance remains reasonably stable up to 5-bit quantization.

Autonomous aerial surveillance using drone feed is an interesting and challenging research domain. Along with surveillance, an autonomous device should be capable of detecting device malfunction or abnormality in real time. However, the inherent uncertainty embedded within the type and level of abnormality makes supervised techniques unsuitable since the adversary may present a unique anomaly for intrusion. To counter this, in this paper, we propose an unsupervised ensemble anomaly detection system to detect device anomaly of an unmanned drone analyzing multimodal data like images and IMU (Inertial Measurement Unit) sensor data synergistically. We have proposed AngleNet and used autoencoder to analyze image and IMU data, respectively and later ensembled the two pipelines for predicting degree of abnormality of the device. Furthermore, we have applied adversarial attack to test the robustness of the proposed approach and integrated defense mechanism. The proposed method performs satisfactorily on the IEEE SP Cup-2020 dataset with an accuracy of 97.8%.

Due to the rise of autonomous vehicles like drones and cars anomaly detection for better and robust surveillance becomes prominent for real-time recognition of normal and abnormal states. But the whole system fails if the unmanned device is unable to detect its own device's anomaly in real-time. Considering the scenario, we can make use of various data of autonomous vehicles like images, video streams, and other digital or analog sensor data to detect device anomaly. In this paper, we have demonstrated a heterogeneous system that estimates the degree of an anomaly in unmanned surveillance drone by inspecting IMU (Inertial Measurement Unit) sensor data and real-time image in an unsupervised approach. We've used AngleNet for detecting images taken in an abnormal state. On top of that, an autoencoder fed by the IMU data has been ensembled with AngleNet for evaluating the final degree of the anomaly. This proposed method is based on the result of the IEEE SP Cup 2020 which achieved 97.3 percent accuracy on the provided dataset. Besides, this approach has been evaluated on an in-house setup for substantiating its robustness.

Heart rate (HR) is one of the key indicators of the physiological condition of a person and Photoplethysmography (PPG) is a non-invasive and low-cost method for its estimation. However, the PPG signal is severely corrupted in the presence of motion artifacts which makes HR estimation difficult. A real-time technique for estimating HR is presented here using a cascaded architecture of adaptive filter and recurrent neural networks (RNN) with HR estimation tracking. The NLMS based adaptive filter reduces the motion artifact from PPG signal, then HR is estimated from the cleaned PPG signal using RNN. Finally, the HR tracking approach provides the eventual estimates and prevents the sudden falls in tracking due to estimation errors. The dataset of 2015 SP Cup organized by IEEE, is used for validation of our proposed approach and mean absolute error of 0.98 beats per minute (BPM) is achieved which is lower compared to other cutting edge techniques. From analysis of experimental results, the robustness and suitability for real-time applications are also validated.

Heart rate (HR) monitoring using photoplethysmographic (PPG) signals recorded from wearers' wrist greatly facilitates design of wearable devices and maximizes user experience. However, placing PPG sensors in wrist causes much stronger and complicated motion artifacts (MA) due to loose interface between sensors and skin. Therefore, developing robust HR estimation algorithms for wrist-type PPG signals has significant commercial values. In this paper, we propose a robust HR estimation algorithm for wrist-type PPG signals using multiple reference adaptive noise cancellation (ANC) technique-termed here as "MURAD." The main challenge of using ANC for MA reduction is to devise a qualified reference noise signal (RNS) to the adaptive filter. We propose a novel solution by using four RNSs, namely, the three-axis accelerometer data and the difference signal between the two PPG signals. For each RNS, we get a different version of the cleaned PPG signal. Then, a set of probable HR values is estimated using all of the cleaned PPG signals, and then, the value that is closest to the estimated HR of the previous time window is chosen to be the HR estimate of the current window. Then, some peak verification techniques are employed to ensure accurate HR estimations. The proposed technique gives lower average absolute error compared to state-of-the art methods. So, MURAD method provides a promising solution to the challenge of HR monitoring using PPG in wearable devices during severe MA conditions.

Wheelchairs provide immense impact for the physically challenged as they offer mobility with relative ease. But in cases like severe disability or seeking relief from physical exertion, there is a necessity for automated electric wheelchairs which are very expensive. In this paper, an innovative design and implementation of a robust and user-friendly smart electric wheelchair are presented, where the product cost is minimized considering the affordability of a large group of people. A notable feature of the proposed scheme is `Destination Mapping' by which the smart wheelchair learns the destinations accessed by the user and autonomously reaches those destinations afterward using speech recognition. There are also various smart features such as rough surface detection, torque adjustments, slope and obstacle detection which promise a comfortable and safe ride. The proposed system takes input from a joystick, microphone, rotary encoders and sonar sensors, processes them through a microcontroller and eventually drives the motors through a motor driver. After much study on selecting suitable motors for implementation, windshield wiper motors were used as they produce quite enough torque at low rpm and are locally available as recycled parts. Experimentation has been carried out with 50 disabled persons of various age groups and subsequent improvements were undertaken based on their feedback over the timeframe of 1 year. Satisfactory results were obtained which implies that the proposed wheelchair can really serve as a low-cost smart alternative to aid the physically challenged, which has a tremendous potential impact on society.

Elderly people in our society may be visually challenged, physically weak and thus face troubles during day-to-day life movement. Due to poor vision, often they fail to detect obstacles around them and hurt themselves. Due to aging, abnormal involuntary movements and lack of a timely rescue system, the magnitude of injury and other adverse effects worsens. One possible way to reduce these old age sufferings could be to design an assistive device which can take corrective measures based on position, neighborhood information and the event of fall (if any). The aim of this research is to develop a cost effective, comfortable and easily producible smart-hat for safe and smooth movement of the elderly people. Since wearing a hat is a common phenomenon for elderly people, smart-hat will get ease of acceptance by them. Some of the key features to be offered by the smart-hat are — obstacle detection, providing aid in the movement in a straight path and fall detection. No such technology exists at present that incorporates all these features in a single device. The proposed device uses a very simple methodology. This system takes reading from an accelerometer and differentiates a fall from other normal activities using an adaptive threshold set in triaxial directions. This device aids an elderly person to walk straight using the angular displacement of head movement obtained from a gyroscope. Lastly, few sonar sensors are attached to the hat to sense obstacles around the user covering a 360-degree range. During an event of fall, the system keeps sending request for help to pre-assigned caregivers by using a GSM module, until any of them responds. Also, the device makes wireless arrangement for opening the lock of the front door for the ease of rescue operation when the victim is at home. The device showed an overall success rate of nearly 86.25%. The implications of this safe and smooth movement assistant would be to reduce the number of accidents, decrease the time of arrival of medical attention, increase supervision capability, thus reducing injuries as well as mortality rate; which in turn will ensure independent living of the target group.