Learning Approaches to Analog and Mixed Signal Verification and Analysis
- Degree Grantor:
- University of California, Santa Barbara. Electrical & Computer Engineering
- Degree Supervisor:
- Malgorzata Marek-Sadowska and Li-C. Wang
- Place of Publication:
- [Santa Barbara, Calif.]
- University of California, Santa Barbara
- Creation Date:
- Issued Date:
- Artificial Intelligence and Engineering, Computer
- Circuit Partitioning,
- Dissertations, Academic and Online resources
- Ph.D.--University of California, Santa Barbara, 2015
There are many automated characterization, test, and verification methods used in practice for digital circuits, but analog and mixed signal circuits suffer from long simulation times brought on by transistor-level analysis. Due to the substantial amount of simulations required to properly characterize and verify an analog circuit, many undetected issues manifest themselves in the manufactured chips.
Creating behavioral models, a circuit abstraction of analog components assists in reducing simulation time which allows for faster exploration of the design space. Traditionally, creating behavioral models for non-linear circuits is a manual process which relies heavily on design knowledge for proper parameter extraction and circuit abstraction. Manual modeling requires a high level of circuit knowledge and often fails to capture critical effects stemming from block interactions and second order device effects. For this reason, it is of interest to extract the models directly from the SPICE level descriptions so that these effects and interactions can be properly captured. As the devices are scaled, process variations have a more profound effect on the circuit behaviors and performances. Creating behavior models from the SPICE level descriptions, which include input parameters and a large process variation space, is a non-trivial task.
In this dissertation, we focus on addressing various problems related to the design automation of analog and mixed signal circuits. Analog circuits are typically highly specialized and fined tuned to fit the desired specifications for any given system reducing the reusability of circuits from design to design. This hinders the advancement of automating various aspects of analog design, test, and layout. At the core of many automation techniques, simulations, or data collection are required. Unfortunately, for some complex analog circuits, a single simulation may take many days. This prohibits performing any type of behavior characterization or verification of the circuit. This leads us to the first fundamental problem with the automation of analog devices. How can we reduce the simulation cost while maintaining the robustness of transistor level simulations? As analog circuits can vary vastly from one design to the next and are hardly ever comprised of standard library based building blocks, the second fundamental question is how to create automated processes that are general enough to be applied to all or most circuit types? Finally, what circuit characteristics can we utilize to enhance the automation procedures?
The objective of this dissertation is to explore these questions and provide suitable evidence that they can be answered. We begin by exploring machine learning techniques to model the design space using minimal simulation effort. Circuit partitioning is employed to reduce the complexity of the machine learning algorithms. Using the same partitioning algorithm we further explore the behavior characterization of analog circuits undergoing process variation. The circuit partitioning is general enough to be used by any CMOS based analog circuit. The ideas and learning gained from behavioral modeling during behavior characterization are used to improve the simulation through event propagation, input space search, complexity and information measurements. The reduction of the input space and behavioral modeling of low complexity, low information primitive elements reduces the simulation time of large analog and mixed signal circuits by 50--75%. The method is extended and applied to assist in analyzing analog circuit layout. All of the proposed methods are implemented on analog circuits ranging from small benchmark circuits to large, highly complex and specialized circuits.
The proposed dependency based partitioning of large analog circuits in the time domain allows for fast identification of highly sensitive transistors as well as provides a natural division of circuit components. Modeling analog circuits in the time domain with this partitioning technique and SVM learning algorithms allows for very fast transient behavior predictions, three orders of magnitude faster than traditional simulators, while maintaining 95% accuracy. Analog verification can be explored through a reduction of simulation time by utilizing the partitions, information and complexity measures, and input space reduction. Behavioral models are created using supervised learning techniques for detected primitive elements. We will show the effectiveness of the method on four analog circuits where the simulation time is decreased by 55--75%. Utilizing the reduced simulation method, critical nodes can be found quickly and efficiently. The nodes found using this method match those found by an experienced layout engineer, but are detected automatically given the design and input specifications. The technique is further extended to find the tolerance of transistors to both process variation and power supply fluctuation. This information allows for corrections in layout overdesign or guidance in placing noise reducing components such as guard rings or decoupling capacitors. The proposed approaches significantly reduce the simulation time required to perform the tasks traditionally, maintain high accuracy, and can be automated. (Abstract shortened by UMI.).
- Physical Description:
- 1 online resource (198 pages)
- UCSB electronic theses and dissertations
- Catalog System Number:
- Samantha Alt, 2015
- In Copyright
- Copyright Holder:
- Samantha Alt
|Access: Public access|
|Alt_ucsb_0035D_12495.pdf||pdf (Portable Document Format)|