Approximate deep neural networks (AxDNNs) are promising for enhancing energy efficiency in real-world devices. One of the key contributors behind this enhanced energy efficiency in AxDNNs is the use of approximate multipliers. Unfortunately, the simulation of approximate multipliers does not usually scale well on CPUs and GPUs. As a consequence, this slows down the overall simulation of AxDNNs aimed at identifying the appropriate approximate multipliers to achieve high energy efficiency with a minimum accuracy loss. To address this problem, we present a novel XAI-Gen methodology, which leverages the analytical model of the emerging hardware accelerator (e.g., Google TPU v4) and explainable artificial intelligence (XAI) to precisely identify the non-critical layers for approximation and quickly discover the appropriate approximate multipliers for AxDNN layers. Our results show that XAI-Gen achieves up to 7x lower energy consumption with only 1-2% accuracy loss. We also showcase the effectiveness of the XAI-Gen approach through a neural architecture search (XAI-NAS) case study. Interestingly, XAI-NAS achieves 40\% higher energy efficiency with up to 5x less execution time when compared to the state-of-the-art NAS methods for generating AxDNNs.

Systolic array-based deep neural network (DNN) accelerators have recently gained prominence for their low computational cost. However, their high energy consumption poses a bottleneck to their deployment in energy-constrained devices. To address this problem, approximate computing can be employed at the cost of some tolerable accuracy loss. However, such small accuracy variations may increase the sensitivity of DNNs towards undesired subtle disturbances, such as permanent faults. The impact of permanent faults in accurate DNNs has been thoroughly investigated in the literature. Conversely, the impact of permanent faults in approximate DNN accelerators (AxDNNs) is yet under-explored. The impact of such faults may vary with the fault bit positions, activation functions and approximation errors in AxDNN layers. Such dynamacity poses a considerable challenge to exploring the trade-off between their energy efficiency and fault resilience in AxDNNs. Towards this, we present an extensive layer-wise and bit-wise fault resilience and energy analysis of different AxDNNs, using the state-of-the-art Evoapprox8b signed multipliers. In particular, we vary the stuck-at-0, stuck-at-1 fault-bit positions, and activation functions to study their impact using the most widely used MNIST and Fashion-MNIST datasets. Our quantitative analysis shows that the permanent faults exacerbate the accuracy loss in AxDNNs when compared to the accurate DNN accelerators. For instance, a permanent fault in AxDNNs can lead up to 66\% accuracy loss, whereas the same faulty bit can lead to only 9\% accuracy loss in an accurate DNN accelerator. Our results demonstrate that the fault resilience in AxDNNs is orthogonal to the energy efficiency.