Approximate storage for energy efficient spintronic memories

Abstract

Spintronic memories are promising candidates for future on-chip storage due to their high density, non-volatility and near-zero leakage. However, the energy consumed by read and write operations presents a major challenge to their use as energy-e cient on-chip memory. Leveraging the ability of many applications to tolerate impreciseness in their underlying computations and data, we explore approximate storage as a new approach to improving the energye ciency of spintronic memories. We identify and characterize mechanisms in STT-MRAM bit-cells that provide favorable energyquality trade-o↵s, i.e., disproportionate energy improvements at the cost of small probabilities of read/write failures. Based on these mechanisms, we design a quality-configurable memory array in which data can be stored to varying levels of accuracy based on application requirements. We integrate the quality-configurable array as a scratchpad in the memory hierarchy of a programmable vector processor and expose it to software by introducing qualityaware load/store instructions within the ISA. We evaluate the energy benefits of our proposal using a device-to-architecture modeling framework and demonstrate 40% and 19.5% improvement in memory energy and overall application energy respectively, for negligible (< 0.5%) quality loss across a suite of recognition and vision applications.

Survey on Approximate Computing and Its Intrinsic Fault Tolerance

“…Therefore, the quality loss (in this case manifested as errors in some operation results) introduced by the approximation would only be acceptable by safety-critical systems if it sharply reduced its area; Read/Write Memory Approximation: It consists of approximating data that is loaded from or written in to the memory, or the read/write operations themselves. This is primarily used on video and image applications, for example, where accuracy and quality can often be relaxed, to reduce memory operations [19,20]. In [6], the authors propose a technique that uses dynamic bit-width based on the application accuracy requirements, where a control system determines the precision of data accesses and loads.…”

A Survey of Soft-Error Mitigation Techniques for Non-Volatile Memories

“…Ranjan et al [45] present an approximate computing approach that explores the energy-quality tradeoff by allowing STT-RAM errors to save energy. They study three types of errors: (1) RDE due to reducing read duration and increasing the read current; (2) incorrect reads due to reduced read current; (3) write error due to lower write current, duration or both.…”

Scalable-effort classifiers for energy-efficient machine learning

“…Approximate computing can also be leveraged to improve the energy consumption of memories [43][44][45]. For example, a hybrid memory array composed of 8T SRAM cells for storing sensitive data vs. 6T SRAM cells for storing resilient ones, allows for more aggresive voltage scaling, resulting in over 32% power savings in the context of video applications [43].…”

“…For example, a hybrid memory array composed of 8T SRAM cells for storing sensitive data vs. 6T SRAM cells for storing resilient ones, allows for more aggresive voltage scaling, resulting in over 32% power savings in the context of video applications [43]. In the context of spintronic memories, various mechanisms at the bit-cell level can be used to create tradeo↵s between energy and the probability of errors during read and write operations [45]. These mechanisms can in turn be utilized to design quality-configurable arrays that o↵er the ability to modulate both the probability and magnitude of errors in return for energy savings [45].…”

“…In the context of spintronic memories, various mechanisms at the bit-cell level can be used to create tradeo↵s between energy and the probability of errors during read and write operations [45]. These mechanisms can in turn be utilized to design quality-configurable arrays that o↵er the ability to modulate both the probability and magnitude of errors in return for energy savings [45].…”