Mixed-signal/RF design-for-test: principles and advanced techniques

The ever increasing pressure to reduce production test costs of highly integrated devices is today leading to the adoption of design-for-test techniques for most common mixed-signal/RF circuits. In addition to classical techniques for facilitating access to embedded components, built-in test techniques have gained acceptance despite the additional costs in terms of silicon overhead and test errors. These advanced techniques may include on-chip test structures to monitor circuit performance while alleviating the complexity of test equipment or complete built-in self-test techniques.

The lecture starts by reviewing the fundamentals of mixed-signal/RF design-for-test, including the classical techniques for facilitating access to embedded modules and the basic principles for on-chip test signal generation and test response analysis. An evaluation of test quality at the design stage is essential for choosing specific built-in test structures. The traditional approach for digital circuits based on the use of fault models and fault simulation is not feasible for most common mixed-signal/RF devices. Instead, the lecture presents statistical techniques for the estimation of parametric test metrics.

Statistical learning is also at the origin of indirect test techniques. This generic approach aims at inferring the outcome of standard specification-based tests from a set of low-cost measurements provided by embedded test structures, sensors or monitors, that are easy to integrate and ideally transparent to device functionality. Furthermore, embedded sensors can also be used to enhance device yield and performance. The lecture illustrates the use of statistical learning for the test and control of mixed-signal/RF devices.

Finally, the lecture will present advanced built-in test techniques for most common mixed-signal devices, including converters, PLLs, RF circuits, and MEMS devices.

  1. Introduction
    • Analog versus digital test
    • Test metrics
    • Fault modeling, characterization and diagnosis
    • The cost of test
  2. Design-for-test principles
    • Analog test access
    • Mixed-signal boundary scan
    • Built-in self-test
      • On-chip test signal generation
      • On-chip response analysis
    • Loopback tests
  3. Density estimation for parametric test metrics estimation
    • Gaussian model
    • Copulas Theory
    • Non-parametric kernel-based density estimation
    • Extreme Value Theory
  4. Statistical learning for test optimization and control
    • Indirect test using regression-based techniques
    • Test ordering and compaction
    • Embedded control using parameter identification
  5. Advanced built-in test techniques
    • ADC/DAC
    • PLL
    • RF
    • MEMS and sensors
A basic understanding of VLSI design and test concepts is assumed. Some knowledge of mixed-signal circuits (converters, PLLs) and RF circuits (low noise amplifiers and mixers) will be helpful for a complete understanding of the advanced design-for-test techniques. The statistical learning techniques will be treated from a user’s point of view, requiring only a basic knowledge of statistics.
The lecture is intended to provide an introduction to the field of mixed-signal/RF design-for-test. The audience will first learn fundamental principles, while advanced techniques for most common devices will be illustrated next. Statistical learning techniques which are today widely researched in this field will also be illustrated, including probability density estimation, regression and parameter identification.