DSpace Collection:

ReLANCE: A Resource-Efficient Low-Latency Cortical Neural Acceleration Engine

Thu, 01 Jan 2026 00:00:00 GMT

Title: ReLANCE: A Resource-Efficient Low-Latency Cortical Neural Acceleration Engine Authors: Kumar, Sonu; Nair, Arjun S.; Chaudhary, Bhawna; Lokhande, Mukul; Vishvakarma, Santosh Kumar Abstract: This brief presents a cortical neural pool (CNP) architecture incorporating a high-speed, resource-efficient CORDIC-based Hodgkin–Huxley (RCHH) neuron. The design employs modular CORDIC stages with a latency–area tradeoff and introduces a constraint-aware modular parallelism (CAMP) scheme with precision and stability handling. The FPGA implementation achieves 24.5% lower LUT utilization and 35.2% faster execution than prior designs while reducing normalized root-mean-square error (NRMSE) by 70%. The CNP engine provides 2.85× higher throughput (12.69 GOPS) than a functionally equivalent CORDIC-based DNN accelerator with only 0.35% accuracy degradation on MNIST. These results demonstrate a biologically accurate, resource-efficient cortical neural acceleration engine (NCE) that employs modular CORDIC stages with a latency–area tradeoff, making it suitable for resource-constrained edge-AI systems. The implementation is publicly available at https://github.com/mukullokhande99/CNP_RCHH © 1993-2012 IEEE.

Integrating Memristor-Based Median Filtering at the Sensor Front-End for Biomedical Image Enhancement

Thu, 01 Jan 2026 00:00:00 GMT

Title: Integrating Memristor-Based Median Filtering at the Sensor Front-End for Biomedical Image Enhancement Authors: Hindoliya, Lokesh Kumar; Jyoti, Kumari; Paul, Animesh; Kumar, Mohit; Yadav, Saurabh; Pachori, Ram Bilas; Mukherjee, Shaibal Abstract: Camera sensors often struggle to capture images in low-light conditions, leading to reduced brightness, contrast, and color fidelity, and increased noise that degrades the performance. Many methods have emerged for image enhancement but they often require slow processing and blur image, making them imperfect for real-world scenarios. This paper presents the first-ever Y2O3-based transmission gate memristor comparator-based median filter (TG-MCBMF) for on-sensor image enhancement in biomedical imaging systems such as X-ray, computed tomography (CT), and magnetic resonance imaging (MRI), designed using Verilog-A. The current system performs front-end noise suppression directly at the sensor output stage, effectively removing salt and pepper (SAP) noise that is introduced during signal acquisition from sensors. The denoised images were reconstructed in MATLAB, and performance was evaluated using quality assessment metrics such as peak signal-to-noise ratio (PSNR), mean square error (MSE), and mean absolute error (MAE). The proposed filter demonstrated superior performance compared to traditional methods such as adaptive median filter (AMF), switch median (SM), and threshold and weighted median filter (TWMF), achieving PSNR values of 46.36 dB for brain CT and 43.84 dB for COVID-19 X-ray, alongside reduced MSE and MAE values of 1.5 and 29.53 for brain CT and 2.67 and 43.84 for COVID-19 X-ray, respectively. The findings indicate the potential of memristor-based filters for next-generation biomedical sensors. © 2017 IEEE.

3 D-HQAM Constellation Design and Performance Evaluation Under AWGN

Thu, 01 Jan 2026 00:00:00 GMT

Title: 3 D-HQAM Constellation Design and Performance Evaluation Under AWGN Authors: Sukhsagar; Bhatia, Vimal Abstract: This letter proposes a simple and effective method for constructing higher-order three-dimensional (3D) signal constellations. The proposed approach introduces a novel 3D hexagonal quadrature amplitude modulation (3D-HQAM) scheme, where constellation points are systematically arranged in a 3D signal space to form structured lattice configurations. To address the increased decision complexity resulting from a larger number of constellation points, a dimension reduction (DR) technique is introduced, allowing the derivation of an analytical approximation of symbol error probability (SEP) under additive white Gaussian noise (AWGN) conditions. Theoretical SEPs closely match simulation results, thereby validating accuracy of the proposed method. The minimum Euclidean distance (MED) of the 3D constellations shows a minimum increase of 12.14% over 2D constellation for 8-HQAM, reaching up to 160.81% for 1024-HQAM constellations. This significant improvement in MED leads to enhanced error performance. Additionally, the average SEP performance of the proposed system is analyzed over Rayleigh faded channel. Therefore, the proposed 3D constellations are promising candidates for high-quality and reliable next-generation digital communication systems. © 2012 IEEE.

Scalable oxide-based memcapacitive crossbar arrays for 1 Kb neuromorphic memory

Wed, 01 Jan 2025 00:00:00 GMT

Title: Scalable oxide-based memcapacitive crossbar arrays for 1 Kb neuromorphic memory Authors: Paul, Animesh; Yadav, Saurabh; Hindoliya, Lokesh Kumar; Dubey, Mayank; Mukherjee, Shaibal Abstract: Memcapacitors are being investigated as potential candidates for high-density data storage. However, developing high-density memcapacitive devices for complex applications is challenging due to higher cycle-to-cycle (C2C) and device-to-device (D2D) variations. In this work, we demonstrate the fabrication of high-density (32 × 32) memcapacitor crossbar arrays achieving device sizes as small as 10 µm × 10 µm using yttrium oxide (Y2O3) as the switching material, deposited via dual ion beam sputtering system. The arrays exhibit low C2C variability (1.01% for VSET and 2.56% for VRESET) and low D2D variability (1.70% for VSET and 4.83% for VRESET). The Y2O3-based crossbar arrays also display robust switching behavior, with a high on/off current ratio (IRATIO > 150), excellent endurance (∼18 000) cycles, long retention ∼160 000 s) and low power consumption of 17 pW. Electrochemical impedance spectroscopy has been utilized to examine the electrical behavior, providing insights into device performance. Neuromorphic functionalities are further demonstrated through potentiation (learning) and depression (forgetting) mechanisms. Moreover, a (16 × 16) array subset is employed to electrically encode random alphabet patterns and exhibit neuromorphic learning capabilities, underscoring the potential of these devices for analog and neuromorphic applications. © 2025 IOP Publishing Ltd. All rights, including for text and data mining, AI training, and similar technologies, are reserved.