Deep Learning-Based Predictive Modeling for Thermal-Aware and Reliability-Optimized VLSI System Design
Keywords:
Deep Learning, VLSI Design, Thermal-Aware Optimization, Reliability Modeling, Predictive Modeling, Electronic Design Automation (EDA).Abstract
As the chips continuously advance in complexity and principle of tightly integrating into the VLSI channels, the thermal mechanism and dependability issue is among design considerations that are strongly sought after. The hotspots are due to high heat levels in localised areas and these processes result in imperfection in the system as well as electromigration and temperature change caused by bias and ultimately shortens the life of the system. In the thermal and reliability modelling practises that have been established, they typically involve physics-based modelling effort-consuming computationally with very little flexibility in adapting to dynamic design situations. In this article, a predictive modelling scheme of thermal-conscious and reliability-optimal design of VLSI systems is proposed on the basis of deep learning. High accuracy of a prediction of temperature distribution, high reliability at the initial design phases is the aim in order to optimise it in advance. It applies a hybrid deep learning model, which applies Convolutional Neural Networks (CNN) to derive Montage features and Long Short-Term Memory (LSTM) to derive temporal relationships. This is training of the model using the benchmark data sets that were created in the simulation environments that included power density, voltage and layout parameters. The experimental evidence suggests that the proposed method is characterised by high predictive performance and low root mean square error of thermal prediction is cheaper in nature than the conventional methods. In addition to it, the framework improves the metrics of reliability because it will be able to detect thermal hotspots in the initial phases due to the increased mean time to failure (MTTF). The proposed solution presents a good and scalable solution to the introduction of AI-based predictive analysis into VLSI design cycles which will be useful to thermal control and long-term reliability.
