MACHINE LEARNING-DRIVEN POWER ESTIMATION FOR STATIC RANDOM ACCESS MEMORY CELLS USING REGRESSION TECHNIQUES

The growing complexity of modern integrated circuits and the demand for memory with low power have made efficient power estimation techniques essential in VLSI design. Static Memory (SRAM), using 6-Transistor cell, is currently used in cache systems due to its low latency and small size. However, traditional simulation-based power analysis using Electronic Design Automation (EDA) tools is time-consuming and computationally expensive. [9][10]Here in our paper, we proposed AI/ML based approach to predict the power consumption estimation of a 6T memory. The SRAM cell is designed and simulated using the Virtuoso tool and a dataset is generated by sweeping various design parameters such as supply voltage, transistor dimensions, capacitance, and operating frequency. These parameters are used as input features, while the corresponding average power consumption obtained from simulations serves as the target output. [9][10] different regression models such as Extra Tree Regressor and Random Forest, are implemented and evaluated based on performance metrics. The simulation results indicates that machine learning models can predict power consumption accurately with reduced computation time compared to conventional methods. [9][10]