ZOS: Zero Overhead Scan for Systolic Array-based AI accelerator

The systolic array is a representative structure that maximizes the computational efficiency of the AI accelerator. In this paper, Zero Overhead Scan (ZOS) method for systolic array-based AI accelerator that minimize test power. The experimental results show that the scan test patterns with 100% stuck-at test coverage can be applied without area overhead, and at the same time, low power testing with an extremely low shift toggle ratio of 0.8% is possible.