Compact Functional Testing for Neuromorphic Computing Circuits

Abstract

We address the problem of testing Artificial Intelligence (AI) hardware accelerators implementing Spiking Neural Networks (SNNs). We define a metric to quickly rank available samples for training and testing based on their fault detection capability. The metric measures the inter-class spike count difference of a sample for the fault-free design. In particular, each sample is assigned a score equal to the spike count difference between the first two top classes. The hypothesis is that samples with small scores achieve high fault coverage because they are prone to misclassification, i.e., a small perturbation in the network due to a fault will result in these samples being misclassififed with high probability. We show that the proposed metric correlates with the per-sample fault coverage and that retaining a set of high-ranked samples in the order of ten achieves near perfect fault coverage for critical faults that affect the SNN accuracy. The proposed test generation approach is demonstrated on two SNNs modelled in Python and on actual neuromorphic hardware. We discuss fault modeling and perform an analysis to reduce the fault space so as to speed up test generation time.

Testability and Dependability of AI Hardware: Survey, Trends, Challenges, and Perspectives

“…ATPG consisting of 5 neighbouring PEs is repeatedly 4) Functional test generation: Functional test generation aims at generating inputs, e.g., images, that are capable of sensitizing the fault and propagating its effect to the output, leading to a different prediction with respect to that of the nominal fault-free network. This approach has been demonstrated for ANNs [127], [146]- [149], including memristive crossbar array-based architectures [146], [147], [149], and for SNNs [150], [151]. As shown in Fig.…”

“…Returning to the functional test generation methods in [127], [146]- [151] discussed in Section V-B4, as the resultant functional test set is compact, it can be also fed periodically during mission mode in idle times towards functional safety.…”

“…The uncovered PEs on the borders can be fully tested in a top-off run.4) Functional test generation: Functional test generation aims at generating inputs, e.g., images, that are capable of sensitizing the fault and propagating its effect to the output, leading to a different prediction with respect to that of the nominal fault-free network. This approach has been demonstrated for ANNs[127],[146]-[149], including memristive crossbar array-based architectures[146],[147],[149], and for SNNs[150],[151]. As shown in Fig.16, functional tests could be original images from training and testing sets, adversarial examples generated from original images, or synthetic images generated from original images.More specifically, starting from the available set of input samples, one approach is to select samples that are profoundly similar to other samples belonging to different output classes, i.e., a similarity metric could be average pixel intensity[127].…”

“…As shown in Fig.16, functional tests could be original images from training and testing sets, adversarial examples generated from original images, or synthetic images generated from original images.More specifically, starting from the available set of input samples, one approach is to select samples that are profoundly similar to other samples belonging to different output classes, i.e., a similarity metric could be average pixel intensity[127]. A second approach is to select samples that have been predicted correctly but with least confidence score[127],[151]. A third approach is to select samples that require more neural network parameter tuning effort during training, where the effort is measured with the change in the loss function in each training step[149].…”