CN-122021508-A - AI-assisted test vector analysis and optimization method and system

CN122021508ACN 122021508 ACN122021508 ACN 122021508ACN-122021508-A

Abstract

The invention discloses an AI-assisted test vector analysis and optimization method and system, and relates to the technical field of integrated circuit design automation. The method comprises a feature extraction step of obtaining a test vector set and extracting multidimensional features comprising fault coverage fingerprints and a scanning chain load mode to construct a feature matrix, an intelligent clustering step of clustering the test vector set by using an unsupervised clustering algorithm, dividing vectors with similar functions into the same vector cluster, a representative vector selection step of selecting representative vectors in each vector cluster based on a fault coverage integrity principle to form an optimized vector set, and an iterative optimization step of verifying the optimized vector set, dynamically adjusting clustering parameters and repeatedly executing the steps if preset conditions are not met. The invention realizes the depth compression of the test vector through the overall function similarity analysis, can obviously reduce the test time and the test cost, and has the advantages of full-automatic post-processing, multi-objective optimization support and the like.

Inventors

Ren qingyuan
AN LEI
ZHENG SONGHAO

Assignees

睿思芯科(深圳)技术有限公司

Dates

Publication Date: 20260512
Application Date: 20260410

Claims (10)

1. An AI-assisted test vector analysis and optimization method, comprising the steps of: The feature extraction step is that a test vector set of the integrated circuit is obtained, and multi-dimensional features at least comprising fault coverage fingerprints and a scanning chain load mode are extracted for each test vector to construct a feature matrix; Based on the feature matrix, clustering the test vector set by using an unsupervised clustering algorithm, and dividing the test vectors with functional similarity into the same vector cluster; A representative vector selection step of selecting at least one representative vector to form an optimized test vector set based on the fault coverage integrity principle in each vector cluster, and And an iterative optimization step, namely verifying the optimized test vector set, and if the verification result does not meet the preset condition, dynamically adjusting the clustering parameters of the intelligent clustering step, and repeatedly executing the feature extraction step, the intelligent clustering step and the representative vector selection step until the preset condition is met.
2. The AI-assisted test vector analysis and optimization method according to claim 1, wherein the multidimensional features extracted in the feature extraction step further comprise logic unit activation states and/or power consumption features, the fault coverage fingerprint is a 0/1 vector representation method for representing coverage conditions of each test vector to each fault in a global fault dictionary, and the scan chain load mode is quantified by calculating transition density and/or weighted transition distribution of a scan chain.
3. The AI-assisted test vector analysis and optimization method according to claim 1, wherein in the intelligent clustering step, the unsupervised clustering algorithm performs adaptive selection according to the distribution characteristics of the feature matrix, the unsupervised clustering algorithm comprises K-means, DBSCAN or spectral clustering, the clustering parameters at least comprise cluster numbers and/or similarity thresholds, and the clustering parameters are dynamically adjusted through at least one of vector-scale-based heuristic initialization, contour coefficient-based adaptive optimization and coverage-feedback-based closed-loop adjustment.
4. The AI-aided test vector analysis and optimization method of claim 1, wherein said failure coverage integrity rules include selecting at least one union of failure sets covered by representative vectors that is equivalent to the union of failure sets covered by all test vectors in the cluster of vectors in which it is located.
5. The AI-assisted test vector analysis and optimization method of claim 4, wherein when there are multiple test vectors with the same fault coverage capability, selecting is based on at least one of distance from cluster center, power consumption estimate, or raw index.
6. The AI-aided test vector analysis and optimization method of claim 1, wherein in the iterative optimization step, the preset conditions include that the fault coverage rate reaches the standard and the compression rate converges to reach the maximum iteration number or parameter boundary, and when the verification result does not meet the preset conditions, the clustering parameters are dynamically adjusted by at least one of the following ways: Adjusting the clustering granularity; adjusting the weight of fault coverage characteristics in the multidimensional characteristics; and (5) switching a clustering algorithm.
7. The AI-assisted test vector analysis and optimization method of claim 1, wherein the step of verifying is performed by invoking an external fault simulation tool that interacts with the system through a file-level interface, an API library interface, or a script interface, the verification results including an overall coverage, an uncovered fault list, and a fault coverage list for each test vector.
8. An AI-assisted test vector analysis and optimization system, comprising: the feature extraction module is used for acquiring a test vector set of the integrated circuit, extracting multidimensional features at least comprising fault coverage fingerprints and a scanning chain load mode for each test vector, and constructing a feature matrix; the intelligent clustering module is used for clustering the test vector set by using an unsupervised clustering algorithm based on the feature matrix and dividing the test vectors with the functional similarity into the same vector cluster; The representative vector selection module is used for selecting at least one representative vector from each vector cluster based on a fault coverage integrity principle to form an optimized test vector set; And the verification and iteration optimization module is used for verifying the optimized test vector set, dynamically adjusting the clustering parameters of the intelligent clustering module if the verification result does not meet the preset condition, and triggering the feature extraction module, the intelligent clustering module and the representative vector selection module to repeatedly execute the operation until the preset condition is met.
9. The AI-assisted test vector analysis and optimization system of claim 8, wherein the validation and iterative optimization module performs validation by invoking an external fault simulation tool and dynamically adjusts the cluster parameters based on validation result feedback by at least one of: Adjusting the clustering granularity; adjusting the weight of fault coverage characteristics in the multidimensional characteristics; and (5) switching a clustering algorithm.
10. The AI-assisted test vector analysis and optimization system of claim 8 wherein the representative vector selection module performs a greedy algorithm to iteratively select test vectors within each cluster of vectors that cover up to uncovered faults as representative vectors until all faults within the cluster are covered.

Description

AI-assisted test vector analysis and optimization method and system Technical Field The invention relates to the technical field of integrated circuit design automation, in particular to a method and a system for intelligently analyzing and optimizing test vectors generated by an automatic test vector generation tool by using an artificial intelligence algorithm. Background In existing chip design flows, ATPG tools are widely used to generate test vectors that meet testability requirements. The tools automatically generate a test vector set which can possibly have faults in the overlay design by analyzing a chip netlist, a scan chain structure and a fault list, can ensure high fault coverage rate and can be directly used for chip production test. In the prior art, ATPG tools typically have built-in compression algorithms to optimize the generated vectors, which based on deterministic rules, merge duplicate vectors by local heuristic strategies, and delete redundant vectors covering the same failure. However, the prior art has the problems that (1) the optimization depth is limited, a built-in compression algorithm only adopts a local optimization strategy, the global function similarity among vectors lacks systematic analysis, a large number of test vectors with overlapping functions still exist in the generated vector set, (2) the intelligent post-processing capability is lacking, the post-intelligent analysis and re-optimization cannot be effectively carried out on a huge vector set, (3) the optimization dimension is single, only a fault coverage target is concerned, the multi-dimensional factors such as the load, the power consumption characteristics and the like of a scanning chain cannot be considered at the same time, the iteration efficiency is low, and the traditional compression algorithm is difficult to quickly converge in the large-scale chip design. Disclosure of Invention The invention aims to provide an AI-assisted test vector analysis and optimization method and system, which are used for solving the problems of high test vector redundancy, single optimization dimension and lack of intelligent post-processing capability in the prior art. In order to achieve the above purpose, the present invention provides the following technical solutions: The AI-aided test vector analysis and optimization method comprises the following steps of obtaining a test vector set of an integrated circuit, extracting multidimensional features at least comprising fault coverage fingerprints and a scanning chain load mode for each test vector, constructing a feature matrix, carrying out clustering on the test vector set by using an unsupervised clustering algorithm based on the feature matrix, dividing test vectors with functional similarity into the same vector cluster, selecting at least one representative vector in each vector cluster based on a fault coverage integrity principle, verifying the optimized test vector set, dynamically adjusting clustering parameters of the intelligent clustering step if a verification result does not meet preset conditions, and repeatedly executing the steps until the preset conditions are met. Preferably, the multidimensional features further comprise logic unit activation states and/or power consumption features, the fault coverage fingerprints adopt a 0/1 vector representation method for representing the coverage condition of each test vector to each fault in the global fault dictionary, and the scan chain load mode is quantified by calculating the jump density and/or the weighted jump distribution of the scan chains. Preferably, in the intelligent clustering step, the unsupervised clustering algorithm performs self-adaptive selection according to the distribution characteristics of the feature matrix, the unsupervised clustering algorithm comprises K-means, DBSCAN or spectral clustering, the clustering parameters at least comprise cluster numbers and/or similarity thresholds, and the clustering parameters are dynamically adjusted through at least one of heuristic initialization based on vector scale, self-adaptive optimization based on contour coefficients and closed-loop adjustment based on coverage rate feedback. Preferably, the fault coverage integrity principle comprises that the selected union of fault sets covered by the at least one representative vector is equal to the union of fault sets covered by all test vectors in the vector cluster where the selected union is located. Preferably, when there are a plurality of test vectors having the same fault coverage capability, the selection is made according to at least one of a distance from a cluster center, a power consumption estimation value, or an original index. Preferably, in the iterative optimization step, the preset condition comprises that the fault coverage rate reaches the standard, the compression rate converges, the maximum iteration number or parameter boundary is reached, and when the verification result does n