Ab Al-Hadi Ab Rahman

Spectral Convolutional Neural Networks (SpCNNs) represent a critical domain in signal processing and machine learning, where the integration of activation functions (AFs) presents significant computational and theoretical challenges. This study systematically evaluates six spectral-domain AFs, employing a comprehensive methodology that assesses their performance across multiple critical metrics including signal quality, computational efficiency, and spectral domain characteristics. Using the MNIST dataset and a high-performance computational infrastructure, we conducted a rigorous analysis of AFs including ACReLU, SReLU, Split-tanh, Split-LReLU, PhaseReLU, and ModReLU. Our findings reveal substantial variations in signal processing capabilities, with novel AFs demonstrating marked improvements in signal quality metrics. Notably, Split-tanh and Split-LReLU exhibited significant noise reduction, achieving Signal-to-Noise Ratio (SNR) and Peak Signal-to-Noise Ratio (PSNR) values approaching 0 dB and 9.47 dB, respectively, compared to traditional AFs. The research provides critical insights into the trade-offs between computational efficiency and signal transformation capabilities, highlighting the potential of advanced spectral-domain AFs in improving neural network performance across complex signal processing tasks.

Optimizing wirelength involves predicting the total length of wires needed to connect different components within a chip during cell placement. It is a fundamental challenge in very-large-scale integration (VLSI) of integrated circuit (IC) design, as it directly impacts the overall performance and manufacturability of chips. Accurate wire-length estimation in the early stages of the design process is critical for guiding subsequent optimization tasks. This paper proposes a novel hybrid linear regression wirelength (hybrid-LRWL) method that combines the strengths of existing methods rectilinear Steiner minimal tree (RSMT) for low-degree nets and a statistical learning-based approach for high-degree nets. Additionally, it compares the performance of three well-established wirelength estimation techniques: half-perimeter wirelength (HPWL), rectilinear minimum spanning tree (RMST), and RSMT. The methods were evaluated using the International Symposium on Physical Design (ISPD) 2011 benchmark suite, considering accuracy and computational efficiency. The experimental results demonstrated that the proposed hybrid method achieves superior accuracy, with a mean error of less than 0.05% in total wirelength, closely approximating RSMT results. The proposed method reduces computational time up to 3.6 times faster than traditional RSM-based methods. The results establish a strong framework for accurate and efficient wirelength estimation in VLSI design for modern, high-performance ICs.

Medium access control (MAC) protocol and routing strategy significantly influence Wireless Network-on-Chip (WiNoC) performance and energy characteristics. Conventional WiNoC MAC typically utilizes a daisy-chained ring topology, which limits wireless channel performance due to maximum waiting times when a radio hub misses the token before packet arrival. Additionally, indiscriminate use of wireless transmission results in channel congestion even when efficient wired paths are available. This paper proposes DA+BMAC, a Distance-Aware (DA) Bidirectional Medium Access Control (BMAC) scheme for mesh WiNoC architecture. BMAC employs bidirectional links allowing token reversal when it overshoots an idle radio hub, reducing worst-case token wait time. DA+BMAC implements distance-aware routing to reserve single-hop wireless transmission exclusively for source-destination pairs beyond a specific distance threshold, while dedicating wired interconnects with higher aggregate bandwidth to short-range communication. Comprehensive validation using a cycle-accurate Noxim simulator on an 8×8 mesh with 16 radio hubs demonstrates up to 11.49× throughput improvement and 15% energy savings compared to a baseline token-ring WiNoC. Scalability analysis confirms performance benefits extend to larger networks (256 and 1024 cores), making DA+BMAC suitable for future many-core systems. The proposed approach has been validated using both synthetic traffic distributions (random, shuffle, transpose, and hotspot) and real-application traces (Barnes and Fluidanimate) from PARSEC and SPLASH-2 benchmark suites, confirming its effectiveness across diverse workloads.