Deep learning based identification of difficult to test nodes
Techniques to improve the accuracy and speed for detection and remediation of difficult to test nodes in a circuit design netlist. The techniques utilize improved netlist representations, test point insertion, and trained neural networks.
1. A system comprising:
a plurality of graphic processing units; and
a memory configured with instructions to operate the plurality of graphics processing units to:
partition a directed acyclic graph into a plurality of subgraphs, each of the subgraphs representing a portion of a circuit;
operate two or more of the graphics processing units each implementing a classifier on a respective one of the subgraphs to generate predictions of difficult to test nodes in the respective subgraph; and
apply the predictions of difficult to test nodes to at least one of the graphics processing units dedicated to performing backpropagation on the classifiers.
2. The system of claim 1 , wherein the instructions configure the plurality of graphics processing units to modify the subgraphs with test nodes inserted at the difficult to test nodes over subsequent iterations of the classifiers.
3. The system of claim 2 , wherein the instructions configure the plurality of graphics processing units to operate the classifiers to predict the difficult to test nodes based on local neighborhoods of nodes in the subgraphs.
4. The system of claim 1 , further comprising:
a plurality of multi-stage classifiers each comprising multiple graphics processing units arranged in a series of stages, each of the multi-stage classifiers configured by the instructions to receive one of the subgraphs.
5. The system of claim 4 , each of the stages implemented by at least one graphics processing unit distinct from graphics processing units implementing other stages.
6. The system of claim 4 , wherein each of the stages is configured by the instructions to generate predictions of non difficult to test nodes in a received subgraph and provide the predictions of non difficult to test nodes to a subsequent graphics processing unit.
7. The system of claim 4 , wherein a last stage of each multi-stage classifier is configured by the instructions to output predictions to the at least one graphics processing unit dedicated to performing backpropagation on the classifiers.