Mitigating the Impact of Faults in Unreliable Memories for Error-Resilient Applications

Inherently error-resilient applications in areas such as signal processing, machine learning and data analytics provide opportunities for relaxing reliability requirements, and thereby reducing the overhead incurred by conventional error correction schemes. In this paper, we exploit the tolerable imprecision of such applications by designing an energy-efficient fault-mitigation scheme for unreliable data memories to meet target yield. The proposed approach uses a bit-shuffling mechanism to isolate faults into bit locations with lower significance. This skews the bit-error distribution towards the low order bits, substantially limiting the output error magnitude. By controlling the granularity of the shuffling, the proposed technique enables trading-off quality for power, area, and timing overhead. Compared to error-correction codes, this can reduce the overhead by as much as 83% in read power, 77% in read access time, and 89% in area, when applied to various data mining applications in 28nm process technology.

FREP: A Soft-Error Resilient Pipelined RISC Architecture

Soft error has become one of the major areas of attention with the device scaling and large scale integration. Lot of variants for superscalar architecture were proposed with focus on program re-execution, thread re-execution and instruction re-execution. In this paper we proposed a fault tolerant micro-architecture of pipelined RISC. The proposed architecture, Floating Resources Extended pipeline (FREP), re-executes the instructions using extended pipeline stages. The instructions are re-executed by hybrid architecture with a suitable combination of space and time redundancy.

Heterogeneous error-resilient scheme for spectral analysis in ultra-low power wearable electrocardiogram devices

Wearable devices performing advanced bio-signal analysis algorithms are aimed to foster a revolution in healthcare provision of chronic cardiac diseases. In this context, energy efficiency is of paramount importance, as long-term monitoring must be ensured while relying on a tiny power source. Operating at a scaled supply voltage, just above the threshold voltage, effectively helps in saving substantial energy, but it makes circuits, and especially memories, more prone to errors, threatening the correct execution of algorithms. The use of error detection and correction codes may help to protect the entire memory content, however it incurs in large area and energy overheads which may not be compatible with the tight energy budgets of wearable systems. To cope with this challenge, in this paper we propose to limit the overhead of traditional schemes by selectively detecting and correcting errors only in data highly impacting the end-to-end quality of service of ultra-low power wearable electrocardiogram (ECG) devices. This partition adopts the protection of either significant words or significant bits of each data element, according to the application characteristics (statistical properties of the data in the application buffers), and its impact in determining the output. The proposed heterogeneous error protection scheme in real ECG signals allows substantial energy savings (11% in wearable devices) compared to state-of-the-art approaches, like ECC, in which the whole memory is protected against errors. At the same time, it also results in negligible output quality degradation in the evaluated power spectrum analysis application of ECG signals.