Efficient verification/testing of system-on-chip through fault grading and analog behavioral modeling
MetadataShow full item record
This dissertation presents several cost-effective production test solutions using fault grading and mixed-signal design verification cases enabled by analog behavioral modeling. Although the latest System-on-Chip (SOC) is getting denser, faster, and more complex, the manufacturing technology is dominated by subtle defects that are introduced by small-scale technology. Thus, SOC requires more mature testing strategies. By performing various types of testing, better quality SoC can be manufactured, but test resources are too limited to accommodate all those tests. To create the most efficient production test flow, any redundant or ineffective tests need to be removed or minimized. Chapter 3 proposes new method of test data volume reduction by combining the nonlinear property of feedback shift register (FSR) and dictionary coding. Instead of using the nonlinear FSR for actual hardware implementation, the expanded test set by nonlinear expansion is used as the one-column test sets and provides big reduction ratio for the test data volume. The experimental results show the combined method reduced the total test data volume and increased the fault coverage. Due to the increased number of test patterns, total test time is increased. Chapter 4 addresses a whole process of functional fault grading. Fault grading has always been a ”desire-to-have” flow because it can bring up significant value for cost saving and yield analysis. However, it is very hard to perform the fault grading on the complex large scale SOC. A commercial tool called Z01X is used as a fault grading platform, and whole fault grading process is coordinated and each detailed execution is performed. Simulation- based functional fault grading identifies the quality of the given functional tests against the static faults and transition delay faults. With the structural tests and functional tests, functional fault grading can indicate the way to achieve the same test coverage by spending minimal test time. Compared to the consumed time and resource for fault grading, the contribution to the test time saving might not be acceptable as very promising, but the fault grading data can be reused for yield analysis and test flow optimization. For the final production testing, confident decisions on the functional test selection can be made based on the fault grading results. Chapter 5 addresses the challenges of Package-on-Package (POP) testing. Because POP devices have pins on both the top and the bottom of the package, the increased test pins require more test channels to detect packaging defects. Boundary scan chain testing is used to detect those continuity defects by relying on leakage current from the power supply. This proposed test scheme does not require direct test channels on the top pins. Based on the counting algorithm, minimal numbers of test cycles are generated, and the test achieved full test coverage for any combinations of pin-to-pin shortage defects on the top pins of the POP package. The experimental results show about 10 times increased leakage current from the shorted defect. Also, it can be expanded to multi-site testing with less test channels for high-volume production. Fault grading is applied within different structural test categories in Chapter 6. Stuck-at faults can be considered as TDFs having infinite delay. Hence, the TDF Automatic Test Pattern Generation (ATPG) tests can detect both TDFs and stuck-at faults. By removing the stuck-at faults being detected by the given TDF ATPG tests, the tests that target stuck-at faults can be reduced, and the reduced stuck-at fault set results in fewer stuck-at ATPG patterns. The structural test time is reduced while keeping the same test coverage. This TDF grading is performed with the same ATPG tool used to generate the stuck-at and TDF ATPG tests. To expedite the mixed-signal design verification of complex SoC, analog behavioral modeling methods and strategies are addressed in Chapter 7 and case studies for detailed verification with actual mixed-signal design are ad- dressed in Chapter 8. Analog modeling effort can enhance verification quality for a mixed-signal design with less turnaround time, and it enables compatible integration of the mixed-signal design cores into the SoC. The modeling process may reveal any potential design errors or incorrect testbench setup, and it results in minimizing unnecessary debugging time for quality devices. Two mixed-signal design cases were verified by me using the analog models. A fully hierarchical digital-to-analog converter (DAC) model is implemented and silicon mismatches caused by process variation are modeled and inserted into the DAC model, and the calibration algorithm for the DAC is successfully verified by model-based simulation at the full DAC-level. When the mismatch amount is increased and exceeded the calibration capability of the DAC, the simulation results show the increased calibration error with some outliers. This verification method can identify the saturation range of the DAC and predict the yield of the devices from process variation. A phase-locked loop (PLL) design cases were also verified by me using the analog model. Both open-loop PLL model and closed-loop PLL model cases are presented. Quick bring-up of open-loop PLL model provides low simulation overhead for widely-used PLLs in the SOC and enables early starting of design verification for the upper-level design using the PLL generated clocks. Accurate closed-loop PLL model is implemented for DCO-based PLL design, and the mixed-simulation with analog models and schematic designs enables flexible analog verification. Only focused analog design block is set to the schematic design and the rest of the analog design is replaced by the analog model. Then, this scaled-down SPICE simulation is performed about 10 times to 100 times faster than full-scale SPICE simulation. The analog model of the focused block is compared with the scaled-down SPICE simulation result and the quality of the model is iteratively enhanced. Hence, the analog model enables both compatible integration and flexible analog design verification. This dissertation contributes to reduce test time and to enhance test quality, and helps to set up efficient production testing flows. Depending on the size and performance of CUT, proper testing schemes can maximize the efficiency of production testing. The topics covered in this dissertation can be used in optimizing the test flow and selecting the final production tests to achieve maximum test capability. In addition, the strategies and benefits of analog behavioral modeling techniques that I implemented are presented, and actual verification cases shows the effectiveness of analog modeling for better quality SoC products.