Abstract
Testing is the last step in IC fabrication flow to tell the good parts from bad ones. Lots of test methods and tools, such as design for test (DFT) methodology and automatic test pattern generator (ATPG), are used to increase IC test coverage through the use of testability in both chip design and pattern generation. However, when dealing with mass production test in a large quantity, lots of other issues should be taken into considerations. Mass production test deals with testing for a large quantity of wafers after test pattern generation. It aims at testing as many output wafers as possible. In addition to picking the good parts, it focuses on testing in an efficient and systematic manner. Mass production test is generally exercised in a testing house, where various equipments and resources are gathered together. This environment abounds with various test capabilities, which, on the other hand, complicates the mass production test. Without a well-established mass test environment, it is hard to manage such a large quantity of wafers, equipments and resources to meet the goal of mass production test as stated above. In this thesis, the operations for IC mass production test are investigated. We first classify the test operations into engineering efforts and operation activities, both of which are then thoroughly explored to identify their disadvantages individually. Later, we propose new automation schemes coping with the drawbacks to establish a well integrated test environment, where mass production test could be controlled, managed and operated easily and efficiently. To facilitate the engineering efforts, the Test Program Repository System (TPRS) and Pro-Correlation System (PCS) are developed. TPRS helps test program adaption in a heterogeneous distributed environment (HDE). It is a central control server exhibiting an uniform operation interface (UOI) to various clients. The centralization of TPRS releases maintenance efforts, such as test program backup, test program privilege control, disk cleaning, ..., etc. The interface uniformity further reinforces the operation efficiency, shortens the training period and reduces miss operations (MOs). In TPRS, test program retrieval race, the situation to retrieve a currently modified test programs, is eliminated. This guarantees the test program retrieval quality; however, is hardly reached in an HDE without TPRS. Immediately before production test operations of an integrated circuit (IC), a correlation mechanism is required to verify the test equipment setup and readiness. In our study, the pro-correlation system (PCS) is proposed to provide an economic and high-quality correlation analysis scheme. Instead of full-wafer probing (FWP), the PCS selectively probes a pre-manufactured recorded correlation wafer (RCW). Each die on the RCW is probed and recorded before mass production test. The PCS obtains a high-quality correlation between the test setup and RCW by deliberately designated probings on the RCW and a quality-reinforced post verification on the first production wafer, both of which are feasible in mass production test. The PCS dramatically reduces both the correlation analysis time and wafer cost as compared with the widely used FWP approach. The benefit is especially significant for deep-submicron ICs. The PCS has been widely used in production, and has been shown to be very effective. PCS is built into not only an equipment setup mechanism, but an integral correlation analysis system, which is used through building a recorded correlation wafer, equipment setup verification, correlation flow control and failure diagnoses and analyses. In the testing phase, the Test Production Monitoring System (TPMS) and Engineering Data Maintenance Page (EDMP) are developed to keep efficient test action and smooth test data collection. Test action is regarding move of test wafers through test stages; test data collection is then the collection and delivery of test results after test action at individual stages. Since mass test is scheduled in advance, any exception results in expensive penalty. This eventually leads to lengthy cycle-time if exceptions are not removed early and properly before propagation. Instead of digging into operation details as the way in the engineering phase, our work in testing phase resort to exception reduction and detection at the earliest possibility so that the cycle-time could be cut down to a large extent. In our study, the TPMS is used to facilitate the test action. With a cross-stage emergency meter and color codes aligned with emergency values, our TPMS arranges the whole test line as a vivid monitoring map, where the whole test devices are clearly ranked with their associated priorities in a real-time manner. By doing so, exceptions could then be easily detected and removed in a concise and intuitive manner. With TPMS, the consistent real-time test line status is available even when multiple clients fetch in parallel at server updating. All these make prompt response to unpredictable exceptions so as to smooth the test action flow. Moreover, the analysis of historical statistics further gains deep insight into the test action flow. As the TPMS helps in the test action flow, so the EDMP does in test data delivery flow. Test data are the results of the test action. Through the testing phase, test data are used in succeeding test stages, and propagated to other systems for further processing. An efficient test process relies on the delivery of test data coordinated with test actions. However, due to independence between delivery processes, exceptions often prevent the delivery flow form being a continuous one. Also, the exception recovery penalty does impact on the test process. Our EDMP connects the test data delivery and collection in an economical manner. By recognizing successful delivery in a late acknowledgement manner and correcting un-successful ones, the EDMP reduces both exceptions and cycle-time effectively. The checkings in EDMP are done in an implicit and economic way to isolate the original delivery path and to make EDMP feasible in the loosely-coupled delivery environment. Besides, later improvement could be made by statistics of log analyses. Our work effectively automates test environment to facilitate. Implementation and deployment of our proposed schemes demonstrate the feasibility and effectiveness. The real exercise results show that mass production test, based on our construction, is able to make the best use of the various facilities and resources. A high-quality, high efficiency test and a smooth, solid test flow are obtained consequently.