Over the past decade, the resistive memory device known as RRAM has been studied extensively in many ways, and many of its problems have been identified, discussed, and some solved. It is time to move from material, process, and device to circuit design and yield, in order to commercialize RRAM. However, as we move from resistive device to memory circuit, new problems do appear, partly because the operating conditions of resistive devices on real RRAM circuit differ from those in an experimental environment for single devices. In this paper, an over forming problem has been identified from our analysis, and we propose a solution based on training sequence. As a result, by solving the over forming problem, RRAM yield can be improved significantly.