Through silicon via (TSV) is the enabling technology for three-dimensional integrated circuit (3D IC) realization. During the TSV manufacturing process, the TSV inevitably has a multi-fault with both resistive-open and leakage faults. The multi-fault will decrease the reliability of the 3D IC seriously. A method of classifying and diagnosing TSV multi-fault is proposed by combining ring oscillator and least square support vector machine (LSSVM). Firstly, the schmidt trigger (ST) as a TSV test receiver based on the initial ring oscillator test structure is implemented. The parameters such as the oscillation period and duty cycle are measured with the TSV as load. A variety of fault types can be detected by testing the changes of these parameters. For further increase the accuracy of the fault test, the oscillation period and duty cycle of different faults are utilized as the feature vectors set, and the set is trained by LSSVM to obtain the fault diagnosis model. In order to get the optimization parameters of the LSSVM model, the particle swarm optimization (PSO) is adopted. As the problem of PSO is tend to local optimization and premature convergence, it may lead to misjudgment of TSV fault. Based on this, W-PSO that dynamically changes the inertia weight is used to optimize the LSSVM. Finally, experiment results imply that the W-PSO-LSSVM fault diagnosis model has a higher fault diagnosis accuracy rate than the PSO-LSSVM fault diagnosis model.

中文翻译：

---

3D IC硅通孔多故障检测与诊断

硅通孔 (TSV) 是实现三维集成电路 (3D IC) 的使能技术。在TSV制造过程中，TSV不可避免地会出现电阻开路和漏电故障的多重故障。多重故障会严重降低3D IC的可靠性。提出了一种将环形振荡器与最小二乘支持向量机(LSSVM)相结合的TSV多故障分类诊断方法。首先，在初始环形振荡器测试结构的基础上，实现了施密特触发器（ST）作为TSV测试接收机。振荡周期和占空比等参数以 TSV 为负载进行测量。通过测试这些参数的变化，可以检测出多种故障类型。为进一步提高故障测试的准确性，以不同故障的振荡周期和占空比作为特征向量集，通过LSSVM训练得到故障诊断模型。为了得到LSSVM模型的优化参数，采用粒子群优化（PSO）。由于 PSO 的问题倾向于局部优化和早熟收敛，可能导致对 TSV 故障的误判。基于此，利用动态改变惯性权重的 W-PSO 来优化 LSSVM。最后，实验结果表明W-PSO-LSSVM故障诊断模型比PSO-LSSVM故障诊断模型具有更高的故障诊断准确率。采用粒子群优化（PSO）。由于 PSO 的问题倾向于局部优化和早熟收敛，可能导致对 TSV 故障的误判。基于此，利用动态改变惯性权重的 W-PSO 来优化 LSSVM。最后，实验结果表明W-PSO-LSSVM故障诊断模型比PSO-LSSVM故障诊断模型具有更高的故障诊断准确率。采用粒子群优化（PSO）。由于 PSO 的问题倾向于局部优化和早熟收敛，可能导致对 TSV 故障的误判。基于此，利用动态改变惯性权重的 W-PSO 来优化 LSSVM。最后，实验结果表明W-PSO-LSSVM故障诊断模型比PSO-LSSVM故障诊断模型具有更高的故障诊断准确率。