Root cause tracing of random alarm is one of the most critical challenges at advanced nodes in manufacturing. As device geometries shrink, process become increasingly sensitive to small differences in chamber environment and performance. The worth of finding root cause for alarm cases is effectively controlling sources of variability that impact on-wafer results, more importantly, bringing the payoff of higher device yield and improved tool stability. In this work, a hybrid solution is proposed for random alarm of etch tool, our method merges supervised and unsupervised way to hand the high dimensional problem, dimension reduction and model training are employed for chamber matching and analysis, convinced results are obtained compared with the traditional methods.