On the reliability estimation of nano-circuits using neural networks

As the integrated circuit geometries shrink, it becomes important for the designers to take into consideration the reliability of the circuits. Different techniques can be used for reliability calculation or estimation. Some of these techniques are accurate but time-consuming while others are quick but not accurate. For example, using a set of mathematical equations for reliability estimation is very fast but not precise enough for large systems. Alternatively, Monte Carlo simulations are highly accurate, but very time-intensive. This work presents three different neural network models for estimating circuit reliability. The models provide better prediction accuracies than the mathematical technique. A reasonably large number of combinational circuits were simulated over a wide range of device reliabilities to collect the training data for the models. Multiple slices of an ISCAS-85 benchmark circuit were used to validate the models' prediction results.