2021/01/28 by Muhammad Abdullah Hanif, Muhammad Shafique, Hanif, Muhammad Abdullah +1
Engineering · #Advanced Memory and Neural Computing #FOS: Computer and information sciences #Ferroelectric and Negative Capacitance Devices #Hardware Architecture (cs.AR) #Semiconductor materials and devices
paper · pdf · doi:10.48550/arxiv.2101.12351
openalex publication_date 2021/01/28 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
Negative Biased Temperature Instability (NBTI)-induced aging is one of the\ncritical reliability threats in nano-scale devices. This paper makes the first\nattempt to study the NBTI aging in the on-chip weight memories of deep neural\nnetwork (DNN) hardware accelerators, subjected to complex DNN workloads. We\npropose DNN-Life, a specialized aging analysis and mitigation framework for\nDNNs, which jointly exploits hardware- and software-level knowledge to improve\nthe lifetime of a DNN weight memory with reduced energy overhead. At the\nsoftware-level, we analyze the effects of different DNN quantization methods on\nthe distribution of the bits of weight values. Based on the insights gained\nfrom this analysis, we propose a micro-architecture that employs low-cost\nmemory-write (and read) transducers to achieve an optimal duty-cycle at run\ntime in the weight memory cells, thereby balancing their aging. As a result,\nour DNN-Life framework enables efficient aging mitigation of weight memory of\nthe given DNN hardware at minimal energy overhead during the inference process.\n