Synopsys, Inc.

Vinod Kumar Nakkala

Sankaranarayanan Srinivasan

Jianfeng Huang

Yanhua Yi

Etienne Lepercq

Yu Yang

Senthilkumar Thoravi Rajavel

Avinash Anantharamu

Pierre Clement

Saibal Ghosh

Sashikala Oblisetty

Vijay Sundaresan

Jiajun Fan

A method or system for estimating delays in design under tests (DUTs) using machine learning. The system accesses multiple DUTs, each comprising various logic blocks. For each DUT, a combinatorial path is identified, connecting one or more logic blocks. A feature vector is generated, including values of orthogonal features representing the combinatorial path's characteristics. Each DUT is compiled for emulation, and the delay of its combinatorial path is measured. These measured delays, along with the corresponding feature vectors, are used to train a machine learning delay model. The trained model is designed to receive a combinatorial path of a DUT as input and generate an estimated wire delay as output. This approach leverages machine learning to predict delays in electronic designs, improving the efficiency and accuracy of delay estimations in complex circuits.

Integrating machine learning delay estimation in FPGA-based emulation systems

A delay estimation system estimates a delay of a DUT for an emulation system. The delay estimation system receives logic blocks of the DUT and a combinatorial path connecting one or more of the logic blocks. The system applies a delay model to a feature vector representing the combinatorial path, where the delay model can determine a delay of the combinatorial path. The delay model may be a machine learning model. The system generates a timing graph using the determined delay and provides the timing graph to a compiler to perform placement and routing of the DUT.

Machine learning delay estimation for emulation systems

A method includes storing a base model generated using base data and receiving training data generated by compiling circuit designs. The method also includes generating, using the training data, a tuned model and generating, using the training data and the base data, a hybrid model. The method further includes receiving a selected cost function and biasing the base model, the tuned model, and the hybrid model using the selected cost function.

Machine-learning enhanced compiler

A method includes generating a netlist for a circuit design and predicting, by applying a first machine learning model to the netlist, a first compile time for the circuit design. The method also includes predicting, by applying a second machine learning model to the netlist, a first place and route strategy based on the first compile time. The method further includes adjusting a logic of the circuit design in accordance with the first place and route strategy.

Growing community of inventors

Sunnyvale, CA, United States of America

Vinod Kumar Nakkala