A Search Based Approach for Stress-Testing Integrated Circuits

Abstract

Abstract. In order to reduce software complexity and be power efficient, hardware platforms are increasingly incorporating functionality that was traditionally administered at a software-level (such as cache management). This functionality is often complex, incorporating multiple processors along with a multitude of design parameters. Such devices can only be reliably tested at a 'system' level, which presents various testing challenges; behaviour is often non-deterministic (from a software perspective), and finding suitable test sets to 'stress' the system adequately is often an inefficient, manual activity that yields fixed test sets that can rarely be reused. In this paper we investigate this problem with respect to ARM's Cache Coherency Interconnect (CCI) Unit. We present an automated search-based testing approach that combines a parameterised test-generation framework with the hill-climbing heuristic to find test sets that maximally 'stress' the CCI by producing much larger numbers of data stall cycles than the corresponding manual test sets.