AI-driven automatic regression creation for digital IC design

Abstract

In modern digital hardware design, hardware description languages are used to represent the behaviour of digital circuits. This abstract representation is at the core of the design, and its correctness and maintainability are paramount. Therefore, engineers typically spend almost a half of the design process timeline in verification: the creation and curation of various tests meant to ensure that behaviour corresponds to design standards which are themselves enforced and maintained across various iterations in the lifetime of the design.

Test driven development, continuous integration, automated regression and even automated test generation or creation are by-products of the expansion of business and public interest in software development, in recognition of the critical support that comprehensive and efficient testing provides to the software sector.

Through this project and the resulting product Digital Automated Test Creator (Digital ARC), we propose to open a new chapter in the workflow utilities compendium for HDL testing, by both automating test creation, and generating human-readable test source code which can then be further maintained by developers. We will focus on unite regression testing, a type of testing required for already managed code, which keeps track of changes in code behaviour through development iterations and flags out bugs in the form of undesirable changes in behaviour across atomic units of code functionality.

Our tool will be designed to achieve flexible coverage rates according to business KPIs, all the while releasing engineers from the mechanical task of writing regression tests, and focusing on more creative, less easily automated tasks that benefit directly from human ingenuity. This type of workflow (sometimes denotes as shift-left, to emphasize the relegation of tasks to an automated instance) is popular in software (Diffblue, IntelliTest) and is likely on the verge of breaking into the hardware community, which creates strong business opportunity for our project.

Lead Participant

Project Cost

Grant Offer

MEDIATEK RESEARCH UK LIMITED £517,123 £ 258,562
 

Participant

MAXEDA TECHNOLOGY

Publications

10 25 50