New applications such as autonomous driving, cyber-physical systems, or remote surgeries demand integrated circuits (ICs) with an ever-lower tolerance for failure. Typical IC design focuses on the targets of functionality and power, performance, and area. An emerging topic in IC design is trustworthiness. It attempts to unify the various interdependent functional and non-functional aspects, such as correct functionality, reliability, security, and functional safety. Existing methodologies and standards focus on evaluating trustworthiness issues (TIs), i.e., causes of faults, and their effects on only one particular attribute. Instead, TIs should be evaluated on their effect on trustworthiness as a whole, which demands a holistic approach. In this paper, we make two main contributions. The first contribution is a framework with a set of unified evaluation criteria that can be applied across all trustworthiness attributes, and a metric, called the Residual Risk Value (RRV). The latter can be used to assess the residual risk of a TI, where a low RRV indicates low risk remaining, and vice versa. RRV considers the impact and likelihood, and the potential for risk reduction enabled by implementing countermeasures. The second contribution is a questionnaire-based measure that ranks TIs according to the priority of addressing them. The results highlight that TIs that emerge during the early stages of IC development should be treated with greater priority. Further, there is a tendency to prioritize security-related TIs as a greater risk to trustworthy ICs. Meanwhile, TIs affecting well-established aspects of IC design and verification are given a lower priority.

Environment perception in autonomous driving vehicles often heavily relies on deep neural networks (DNNs), which are subject to domain shifts, leading to a significantly decreased performance during DNN deployment. Usually, this problem is addressed by unsupervised domain adaptation (UDA) approaches trained either simultaneously on source and target domain datasets or even source-free only on target data in an offline fashion. In this work, we further expand a source-free UDA approach to a continual and therefore online-capable UDA on a single-image basis for semantic segmentation. Accordingly, our method only requires the pre-trained model from the supplier (trained in the source domain) and the current (unlabeled target domain) camera image. Our method Continual BatchNorm Adaptation (CBNA) modifies the source domain statistics in the batch normalization layers, using target domain images in an unsupervised fashion, which yields consistent performance improvements during inference. Thereby, in contrast to existing works, our approach can be applied to improve a DNN continuously on a single-image basis during deployment without access to source data, without algorithmic delay, and nearly without computational overhead. We show the consistent effectiveness of our method across a wide variety of source/target domain settings for semantic segmentation. Code is available athttps://github.com/ifnspaml/CBNA