Machine learning based effective linear regression model for TSV layer assignment in 3DIC

On the integration of 3D IC design, thermal management issues play a significant role. So, it is required to implement an effective approaches and solutions for integrating 3DIC. The TSV causes problems with the distinct coefficients of thermal expansion that induces mismatch strains and stresses. The major drawback of 3DIC is the thermal management issues which increases the power consumption through the current crowding, perhaps the temperature upraised by the slacked layers due to its heat generation. Several research has not been undergone in 3DIC utilizing machine learning approaches which is highly complicated. This paper firstly proposes an efficient ML model to achieve better reduction in wire length and temperature. An efficient linear regression model is preferred here in order to achieve significant performances in TSV layer assignment. The linear regression utilized gradient based approach where the error is predicted at every instance through tracing gradient cost function. An optimized TSV layer assignment is achieved with this flexible ELRM. The performance analysis of data shows that the proposed ELRM based TSV assignment achieved better wire length and temperature. The ISPD98 Circuit Benchmark Suite is utilized for result evaluation and it achieves improved TSV layer assignment through reducing wire length and temperature.