High Quality Test Generation at the Register Transfer Level
| Autor: | Gent, Kelson Andrew |
|---|---|
| Rok vydání: | 2016 |
| Předmět: | |
| Druh dokumentu: | Dissertation |
| Popis: | Integrated circuits, from general purpose microprocessors to application specific designs (ASICs), have become ubiquitous in modern technology. As our applications have become more complex, so too have the circuits used to drive them. Moore's law predicts that the number of transistors on a chip doubles every 18-24 months. This explosion in circuit size has also lead to significant growth in testing effort required to verify the design. In order to cope with the required effort, the testing problem must be approached from several different design levels. In particular, exploiting the Register Transfer Level for test generation allows for the use of relational information unavailable at the structural level. This dissertation demonstrates several novel methods for generating tests applicable for both structural and functional tests. These testing methods allow for significantly faster test generation for functional tests as well as providing high levels of fault coverage during structural test, typically outperforming previous state of the art methods. First, a semi-formal method for functional verification is presented. The approach utilizes a SMT-based bounded model checker in combination with an ant colony optimization based search engine to generate tests with high branch coverage. Additionally, the method is utilized to identify unreachable code paths within the RTL. Compared to previous methods, the experimental results show increased levels of coverage and improved performance. Then, an ant colony optimization algorithm is used to generate high quality tests for fault coverage. By utilizing co-simulation at the RTL and gate level, tests are generated for both levels simultaneously. This method is shown to reach previously unseen levels of fault coverage with significantly lower computational effort. Additionally, the engine was also shown to be effective for behavioral level test generation. Next, an abstraction method for functional test generation is presented utilizing program slicing and data mining. The abstraction allows us to generate high quality test vectors that navigate extremely narrow paths in the state space. The method reaches previously unseen levels of coverage and is able to justify very difficult to reach control states within the circuit. Then, a new method of fault grading test vectors is introduced based on the concept of operator coverage. Operator coverage measures the behavioral coverage in each synthesizable statement in the RTL by creating a set of coverage points for each arithmetic and logical operator. The metric shows a strong relationship with fault coverage for coverage forecasting and vector comparison. Additionally, it provides significant reductions in computation time compared to other vector grading methods. Finally, the prior metric is utilized for creating a framework of automatic test pattern generation for defect coverage at the RTL. This framework provides the unique ability to automatically generate high quality test vectors for functional and defect level testing at the RTL without the need for synthesis. In summary, We present a set of tools for the analysis and test of circuits at the RTL. By leveraging information available at HDL, we can generate tests to exercise particular properties that are extremely difficult to extract at the gate level. Ph. D. Digital circuits and modern microprocessors are pervasive in modern life. The complexity and scope of these devices has dramatically increased to meet new demands and applications, from entertainment devices to advanced automotive applications. Rising complexity causes design errors and manufacturing defects are more difficult to detect and increases testing costs. To cope with rising test costs, significant effort has been directed towards automating test generation early in development when defects are less expensive to correct. Modern digital circuits are designed using Hardware Description Languages (HDL) to describe their behavior at a high logical level. Then, the behavioral description is translated to a chip level implementation. Most automated test tools use the implementation description since it is a more direct representation of the manufactured circuit. This dissertation demonstrates several methods to utilize available logical information in behavioral descriptions for generating tests early in development that maintain applicability throughout the design process. The proposed algorithms utilize a biologically-inspired search, the ant colony optimization, abstracting test generation as an ant colony hunting for food. In the abstraction, a sequence of inputs to a circuit is represented by the walked path of an individual ant and untested portions of the circuit description are modelled as food sources. The final test is a collection of paths that efficiently reach the most food sources. Each algorithm also explores different software analysis techniques, which have been adapted to handle unique constraints of HDLs, to learn about the target circuits. The ant colony optimization uses the analysis to help guide and direct the search, yielding more efficient execution than prior techniques and reducing the time required for test generation. Additionally, the described methods can automatically generate tests in cases previously requiring manual generation, improving overall test quality. |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|