International Business Machines Corporation

Pravin Kamdar

David Wells Winston

Tong Li

Richard Daniel Kimmel

David James Hathaway

Bhavna Agrawal

Karl K Moody, Iii

Peng Peng

Methods, systems, and computer program products for machine learning model decision boundary enhancement are provided. Aspects include determining a decision boundary associated with a first machine learning model, wherein the first machine learning model is trained with an initial training set of features and associated classifications from a verification model, obtaining a plurality of new feature sets, analyzing, by the first machine learning model, the plurality of new feature sets to determine that a subset of feature sets have classification predictions within a threshold range of the decision boundary associated with the first machine learning model, inputting the subset of feature sets into the verification model to determine enhanced training data based on outputs of the verification model, and creating an enhanced machine learning model by further training the first machine learning model with the initial training data and the enhanced training data.

Decision boundary enhancement for learning models

Systems and methods of developing an integrated circuit implement selecting a desired yield for a circuit used in the integrated circuit. The desired yield corresponds to a desired failure probability of the circuit. The method includes determining a parameter threshold value that corresponds with the desired yield. The circuit passes if a parameter associated with the circuit is below the parameter threshold value and the desired yield indicates a percentage of instances of the circuit that pass according to the parameter threshold value. The method also includes using the parameter threshold value that corresponds with the desired yield during testing and improvement of a design of the integrated circuit, and providing the design of the integrated circuit for fabrication.

Iterative approach to determine failure threshold associated with desired circuit yield in integrated circuits

Methods, systems, can computer program products for optimizing statistical method of computing output yields are provided. Aspects include determining a target criteria of a system for statistical analysis, based on the target criteria, determining a statistical analysis algorithm for the simulation, determining a block size for a plurality of statistical samples of the system for a parallelization of the statistical analysis algorithm, generating the plurality of statistical samples of the system, simulating the plurality of statistical samples of the system to determine one or more output yields, calculating a confidence interval for each of the one or more output yields, wherein the confidence interval comprises a lower bound, comparing the lower bound to a threshold standard deviation of a probability density function, and adjusting the block size for the plurality of statistical samples based on determining that the lower bound is less than the threshold standard deviation.

Adaptive accelerated yield analysis

Disclosed are embodiments of a system and of an associated method for estimating the leakage current of an electronic circuit. The embodiments analyze a layout of an electronic circuit in order to identify all driven and non-driven nets within the electronic circuit, to identify all of the driven net-bounded partitions within the electronic circuit (based on the driven and non-driven nets), and to identify, for each driven net-bounded partition, all possible states of the electronic circuit that can leak. Then, using this information, the embodiments estimate the leakage current of the electronic circuit. This is accomplished by first determining, for each state of each driven net-bounded partition, a leakage current of the driven net-bounded partition and a probability that the state will occur in the driven net-bounded partition during operation of the electronic circuit. Then, for each state of each driven net-bounded partition, the leakage current of the driven net-bounded partition and the state probability are multiplied together. The results are then aggregated.

Growing community of inventors

Leander, TX, United States of America

Pravin Kamdar