Probe card that serves as the carrier of die and the transmitter of information is an indispensable test interface for integrated circuit testing. The probe card extracts the electrical signal of chip and sends it to the prober to screen defectives. In the process of interface transmission, signal disturbance or attenuation will lead to functional errors, and the output result will be different from the expected one to the screen of selected known good dies for integrated circuit packaging and final test. If abnormal situations happen with probe cards, the engineers will eliminate potential fault causes by trial and error method according to domain knowledge and personal experience for troubleshooting. As semiconductor industry is continuously migrating with shrinking feature size, the diagnosing and troubleshooting procedure for probe card is exponentially complicated and time-consuming. To enhance data integrity of circuit probe testing, this study aims to develop a Bayesian network for probe card fault diagnosis and troubleshooting via the integrated data-driven solutions considering potential rules derived from domain knowledge and manufacturing big data to empower Industry 3.5 smart production. An empirical study is conducted in a leading semiconductor testing company for validation. The experiment results show that the proposed approach can improve one-shot probability from 0.13 to 0.36 to improve one-shot success chance in troubleshooting process. The expected shooting times for probe faults have been reduced from 11 times to 3.96 times on average to save 63.97% of fault troubleshooting efforts in testing process. The proposed approach can provide effective suggestions to shorten troubleshooting time for yield improvement and subsequent packaging cost reduction to empower flexible decision-making for smart production. The results have shown practical viability of proposed approach for Industry 3.5. Indeed, the developed solutions have been implemented in real settings.

用作芯片载体和信息发送者的探针卡是集成电路测试必不可少的测试接口。探针卡提取芯片的电信号，并将其发送到探针器以筛选缺陷。在接口传输过程中，信号干扰或衰减将导致功能错误，并且输出结果将与预期的选择的已知好的用于集成电路封装和最终测试的裸片的屏幕不同。如果探针卡出现异常情况，工程师将根据领域知识和个人经验，通过反复试验的方法消除潜在的故障原因，以进行故障排除。随着半导体行业的不断发展和功能尺寸的缩小，探针卡的诊断和故障排除过程极其复杂且耗时。为了增强电路探针测试的数据完整性，本研究旨在通过集成的数据驱动解决方案，开发一种用于探针卡故障诊断和故障排除的贝叶斯网络，其中要考虑从领域知识中得出的潜在规则并制造大数据，以实现工业3.5智能生产。一家领先的半导体测试公司进行了一项实证研究，以进行验证。实验结果表明，该方法可以将单发概率从0.13提高到0.36，从而提高了故障排除过程中单发成功的机会。预期的探针故障排除时间平均从11倍减少到3.96倍，从而节省了63.97％的故障排除工作。所提出的方法可以提供有效的建议，以缩短故障排除时间，以提高产量并降低包装成本，从而为智能生产提供灵活的决策能力。结果显示了针对工业3.5提出的方法的实际可行性。实际上，已开发的解决方案已在实际环境中实施。