Visible to the public Security-Driven Task Scheduling for Multiprocessor System-on-Chips with Performance Constraints

TitleSecurity-Driven Task Scheduling for Multiprocessor System-on-Chips with Performance Constraints
Publication TypeConference Paper
Year of Publication2018
AuthorsWang, Nan, Yao, Manting, Jiang, Dongxu, Chen, Song, Zhu, Yu
Conference Name2018 IEEE Computer Society Annual Symposium on VLSI (ISVLSI)
Date PublishedJuly 2018
ISBN Number978-1-5386-7099-6
Keywordscomposability, delays, graph theory, Hardware, hardware trojan, industrial property, intellectual property security, IP networks, logic design, malicious inclusions, MPSoC, multiprocessing systems, Multiprocessor System-on-Chips, performance constraints, policy-based governance, Processor scheduling, pubcrawl, resilience, Resiliency, schedule length, Schedules, scheduling, security, security of data, security-driven task scheduling, system level security constraints, System performance, system-on-chip, Task Analysis, task scheduling, Trojan horses, two-stage performance-constrained task scheduling algorithm

The high penetration of third-party intellectual property (3PIP) brings a high risk of malicious inclusions and data leakage in products due to the planted hardware Trojans, and system level security constraints have recently been proposed for MPSoCs protection against hardware Trojans. However, secret communication still can be established in the context of the proposed security constraints, and thus, another type of security constraints is also introduced to fully prevent such malicious inclusions. In addition, fulfilling the security constraints incurs serious overhead of schedule length, and a two-stage performance-constrained task scheduling algorithm is then proposed to maintain most of the security constraints. In the first stage, the schedule length is iteratively reduced by assigning sets of adjacent tasks into the same core after calculating the maximum weight independent set of a graph consisting of all timing critical paths. In the second stage, tasks are assigned to proper IP vendors and scheduled to time periods with a minimization of cores required. The experimental results show that our work reduces the schedule length of a task graph, while only a small number of security constraints are violated.

Citation Keywang_security-driven_2018