Tutorial T4A: Formal Verification Techniques and Tools for Complex Designs

Integrated circuit technology has evolved from micro-controllers and discrete components to fully integrating a large system on a single chip (SoC). Today, verification is the most expensive component in the design cycle in term of cost and time. This cost is estimated to consume about 70% to 80% of the total design effort. The verification cost is expected to increase for SoC designs. This is mainly due to the increase in complexity and to the shrinking of the product design cycle. For example, the color TV took over 10 years to sell 1 million units, while the DVD player took just over a year. This shrinking of the design cycle is going to put more pressure on increasing designer productivity which is affected directly by the cost of verification. For these reasons, verification of complex designs is becoming a bottleneck in the process of producing integrated SoC systems. This tutorial provides an overview of emerging directions in formal verification and a discussion of new tools being developed in industry and research directions to enable automated verification of next generation systems on a chip. The tutorial will begin with a comprehensive overview of techniques for formally verifying complex designs. It will include the fundamental theory, applicability to different types of VLSI designs, as well as the performance and limitations of various approaches. The focus will be on Formal methods and will include both equivalence checking and property checking. Formal equivalence checking methods (between RTL and gate levels) incorporated into industry tools will be described, as well as new techniques for checking the equivalence between electronic system languages (such as SystemC) and RTL. The basics of property checking techniques in existing tools will be described, including the basics of model checking, and search algorithms that automatically show the correctness/violation of a property. Limitation and benefits of both SATisfiability and automatic test pattern generation (ATPG) based bounded model property checking (BMC) will be described. In order to deal with complexity, powerful model abstractions which can be automatically generated from static analysis of the RTL descriptions will be introduced. These include functional partitioning, static slicing and antecedent conditional slicing. These techniques can be used with existing tools to reduce their CPU and memory requirements while producing exactly the same results. This would be of particular interest to verification engineers and designers