Files in this item

FilesDescriptionFormat

application/pdf

application/pdfPAL-DISSERTATION-2019.pdf (5MB)Restricted to U of Illinois
(no description provided)PDF

Description

Title:Scalable functional validation of next generation SoCs
Author(s):Pal, Debjit
Director of Research:Vasudevan, Shobha
Doctoral Committee Chair(s):Vasudevan, Shobha
Doctoral Committee Member(s):Hwu, Wen-mei; Chen, Deming; Adve, Sarita V.; Ziv, Avi
Department / Program:Electrical & Computer Eng
Discipline:Electrical & Computer Engr
Degree Granting Institution:University of Illinois at Urbana-Champaign
Degree:Ph.D.
Genre:Dissertation
Subject(s):Observability enhancement, Post-silicon debug and diagnosis, Application-level analysis, Vertically integrated solutions, Abstractions, Feature engineering, OpenSPARC T2 SoC, Assertion ranking
Abstract:System-on-Chips (SoCs) constitutes the primary backbone of modern embedded computing devices including many safety-critical applications e.g., autonomous vehicles, health care systems. The presence of any undetected bugs in these systems would have aberrant cost both in terms of safety and reliability and can cause loss of property or life. Hence, SoC validation is a crucial task to ensure the functional correctness of an SoC. The sheer size, presence of hundreds of concurrently executing heterogeneous IPs, vertical integration of SoC components e.g., hardware/firmware/software to realize multiple functionality, and application-level relevance of components present a new spectrum of validation challenges that have rendered the traditional microprocessor validation paradigm moot in the context of SoC validation. The challenges include observability enhancement and debug and diagnosis under the constraint of vertical integrations, identifying high-quality verification artifacts among others. In industrial practice, SoC validation is a manual, unsystematic, and ad hoc process that heavily relies on the expertise and the creativity of the validator. Consequently, there is an urgent need to develop scalable and efficient algorithms of industrial relevance to address this massive ongoing challenge of SoC validation. This dissertation makes contributions to both post-silicon and pre-silicon validation of SoCs, with highly impactful contributions to next-generation post-silicon SoC validation. We use top-down analysis, a higher level of abstraction, and application relevance as the key ideas to automate post-silicon observability enhancement for industrial scale SoCs and scale observability to design that is more than 300x the size of designs that have been presented in the academic literature so far. Our observability enhancement solution can be applied at the netlist-level, behavioral level, and at the system-wide application level to select high-quality signals that are most beneficial for post-silicon debug and diagnosis. We apply a feature engineering based machine learning technique on the observed signal data to develop an automatic, scalable, and efficient post-silicon debug and diagnosis solution. The key idea is to learn the correct and erroneous design behavior automatically from trace data without prior design knowledge. We believe our debugging solution can automate post-silicon debug and diagnosis, where manual debugging is the norm. The quality of SoC verification and validation heavily depends on the quality of verification artifacts e.g., assertions. To automate and expedite identification of high-functional coverage assertions that are useful for regression analysis, localization, etc., we have also developed a comprehensive ranking scheme for assertions. The key idea is to identify assertions that capture important design behaviors by analyzing the design source code. Our SoC validation solutions are scalable and efficient. We consistently show orders of magnitude speedup improvements over the state-of-the-art while objectively improving quality of results. We have shown that going forward application-level analysis is the key to scale post-silicon validation to industrial scale SoCs. Our proposed validation solutions can plug into the existing industrial validation process to introduce automation in the current unsystematic, ad hoc, manual settings with multiple order of magnitudes of benefit.
Issue Date:2019-07-11
Type:Text
URI:http://hdl.handle.net/2142/105812
Rights Information:Copyright 2019 Debjit Pal
Date Available in IDEALS:2019-11-26
Date Deposited:2019-08


This item appears in the following Collection(s)

Item Statistics