12 July 2018 in IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems
An RRAM-based computing system (RCS) is an attractive hardware platform for implementing neural computing algorithms. On-line training for RCS enables hardware-based learning for a given application and reduces the additional error caused by device parameter variations. However, a high occurrence rate of hard faults due to immature fabrication processes and limited write endurance restrict the applicability of on-line training for RCS. We propose a fault-tolerant on-line training method that alternates between a fault-detection phase and a fault-tolerant training phase. In the fault-detection phase, a quiescent-voltage comparison method is utilized. In the training phase, a threshold-training method and a re-mapping scheme is proposed. Our results show that, compared to neural computing without fault tolerance, the recognition accuracy for the Cifar-10 dataset improves from 37% to 83% when using low-endurance RRAM cells, and from 63% to 76% when using RRAM cells with high endurance but a high percentage of initial faults.
25 January 2022 in IEEE Transactions on Very Large Scale Integration (VLSI) Systems
A resistive switching random access memory (ReRAM)-based computing system (RCS) provides an energy-efficient hardware implementation of vector–matrix multiplication for machine-learning hardware. However, it is susceptible to faults due to the immature resistive switching random access memory (ReRAM) fabrication process. We propose an efficient online fault-detection method for RCS. This method monitors the dynamic power consumption of each ReRAM crossbar and determines the occurrence of faults when a changepoint is detected in the monitored power-consumption time series. To estimate the percentage of faulty cells in a faulty ReRAM crossbar, we compute statistical features immediately before and after the changepoint and use them as independent variables; we use the percentage of faulty cells as dependent variables to train a predictive model using machine learning. In this way, the computationally expensive fault localization and error-recovery steps are carried out only when a high fault rate is estimated. Simulation results show that, with the fault-detection method and the predictive model, the test time is significantly reduced, the hardware overhead is negligible, and high classification accuracy for the MNIST and CIFAR-10 datasets using RCS can still be ensured.
16 October 2020 in IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems
High integration densities and design complexity of printed-circuit boards make board-level functional fault identification extremely difficult. Machine learning provides an opportunity to identify functional faults with high accuracy and thereby reduce repair cost. However, the large volume of manufacturing data comes in a streaming format and exhibits time-dependent concept drift in a production environment. These drawbacks limit the effectiveness of traditional machine-learning algorithms. We propose a diagnosis workflow that utilizes online learning to train classifiers incrementally with a small chunk of data at each step. These online learning algorithms adapt to concept drift quickly with carefully designed update rules. A hybrid algorithm is also proposed to handle the scenario that data for varying numbers of boards are collected at different times. This hybrid algorithm concurrently implements two basic models. For each data chunk, this algorithm chooses the better model with high probability. Experimental results using two boards in high-volume production show that, with the help of online learning and the proposed hybrid algorithm, the F1-score for diagnosis based on binary classifiers can be improved from 57.3% to 81.0%. The top-3 accuracy for diagnosis based on multiclass classifiers can be improved from 78.3% to 91.4%.
15 March 2021 in IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems
High integration densities and design complexity make board-level functional fault diagnosis extremely difficult. Machine-learning techniques can identify functional faults with high accuracy, but they require a large volume of data to achieve high-prediction accuracy. This drawback limits the effectiveness of traditional machine-learning algorithms for training a model in the early stage of manufacturing, when only a limited amount of fail data and repair records are available. We propose a board-level diagnosis workflow that utilizes domain adaptation (DA) to transfer the knowledge learned from mature boards to a new board in the ramp-up phase. First, based on the requirement of fault diagnosis, we select an appropriate domain-adaptation method to reduce differences between mature boards and the new board. Second, these DA methods utilize information from both the mature and the new boards with carefully designed domain-alignment rules and train a functional fault diagnosis classifier. Experimental results using three complex boards in volume production and one new board in the ramp-up phase show that, with the help of DA and the proposed workflow, the diagnosis accuracy is improved.