Over the last decade, usage of integrated circuits met an unprecedently spread in most domains of human activity. Processors are embedded in more and more products, consequently, request for their manufacturing increased exponentially. Since hardware verification is a very time-consuming step in electronics development, engineers make huge efforts to reduce its time by increasing automation of this process, and artificial intelligence is an ideal candidate for that job. Aligning to this global effort, advanced methods for improving verification performance are analyzed in this paper. Current research represents an important step before proceeding to implementation of artificial intelligence concepts inside integrated circuit verification process. Machine learning in verification can be used for multiple directions: generating code for driving full range of stimuli needed to bring verified circuit into all its functional states; analyzing test results; filling coverage gaps and list of examples can highly continue. However, a big challenge is to decide which parts of verification environment can best address needs of automation and which verification techniques are most suitable for supplying data to automatic code generators. In this article, some features of e language are highlighted as welcome bridges between hardware verification and artificial intelligence.
The functional verification process is one of the most expensive steps in integrated circuit manufacturing. Functional coverage is the most important metric in the entire verification process. By running multiple simulations, different situations of DUT functionality can be encountered, and in this way, functional coverage fulfillment can be improved. However, in many cases it is difficult to reach specific functional situations because it is not easy to correlate the required input stimuli with the expected behavior of the digital design. Therefore, both industry and academia seek solutions to automate the generation of stimuli to reach all the functionalities of interest with less human effort and in less time. In this paper, several approaches inspired by genetic algorithms were developed and tested using three different designs. In all situations, the percentage of stimulus sets generated using well-performing genetic algorithms approaches was higher than the values that resulted when random simulations were employed. In addition, in most cases the genetic algorithm approach reached a higher coverage value per test compared to the random simulation outcome. The results confirmed that in many cases genetic algorithms can outperform constrained random generation of stimuli, that is employed in the classical way of doing verification, considering coverage fulfillment level per verification test.
Constrained random stimulus generation is no longer sufficient to fully simulate the functionality of a digital design. The increasing complexity of today’s hardware devices must be supported by powerful development and simulation environments, powerful computational mechanisms, and appropriate software to exploit them. Reinforcement learning, a powerful technique belonging to the field of artificial intelligence, provides the means to efficiently exploit computational resources to find even the least obvious correlations between configuration parameters, stimuli applied to digital design inputs, and their functional states. This paper, in which a novel software system is used to simplify the analysis of simulation outputs and the generation of input stimuli through reinforcement learning methods, provides important details about the setup of the proposed method to automate the verification process. By understanding how to properly configure a reinforcement algorithm to fit the specifics of a digital design, verification engineers can more quickly adopt this automated and efficient stimulus generation method (compared with classical verification) to bring the digital design to a desired functional state. The results obtained are most promising, with even 52 times fewer steps needed to reach a target state using reinforcement learning than when constrained random stimulus generation was used.
Digital integrated circuits play an important role in the development of new information technologies and support Industry 4.0 from a hardware point of view. There is great pressure on electronics companies to reduce the time-to-market for product development as much as possible. The most time-consuming stage in hardware development is functional verification. As a result, many industry and academic stakeholders are investing in automating this crucial step in electronics production. The present work aims to automate the functional verification process by means of genetic algorithms that are used for generating the relevant input stimuli for full simulation of digital design behavior. Two important aspects are pursued throughout the current work: the implementation of genetic algorithms must be time-worthy compared to the application of the classical constrained-driven generation and the verification process must be implemented using tools accessible to a wide range of practitioners. It is demonstrated that for complex designs, functional verification powered by the use of genetic algorithms can go beyond the classical method of performing verification, which is based on constrained-random stimulus generation. The currently proposed methods were able to generate several sets of highly performing stimuli compared to the constraint-random stimulus generation method, in a ratio ranging from 57:1 to 205:1. The performance of the proposed approaches is comparable to that of the well-known NSGA-II and SPEA2 algorithms.
Considering progress in semiconductor manufacturing processes (where technology advanced creating transistors from 10 μm in 1971 to a few nm nowadays), integrated circuits have become more and more complex and have embedded huge number of functionalities. Consequently, functional verification of verification circuits designs highly increased in difficulty. Both the academic community and the industry are involved in research projects for obtaining an efficient thorough verification of circuit designs. Thoroughness of verification is defined by metrics fulfillment in verification process. Aligning to this global effort, in this paper, four methodologies of fulfilling coverage (most important component of metrics) are described and compared. The study is concerned on outcome of each method, and on implementation effort, as well.
The continuous advance of science often leaves behind devices,and makes their usage obsolete. This can be observed, for example, in the medical domain,where the performance of devices achieved tremendous capabilities, or in the latest increase of power of computing: today's top-ranked smartphones are comparable in performance with the best desktop computers of the previous decade. Therefore, significantresearch efforts aredirected at the reusability of current technologies by updating them according tonew discoveries. Among the existing solutions for device reusability, there are two concepts which are highly ranked: the usage of system-on-chip devices and partial reconfigurability implementation. This paper analyzes the benefits of using these solutionsbothindependently andcombined.