English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

The primary operation in DNNs is the dot product of quantized inputactivations and weights. Prior works have proposed the design of memory-centricarchitectures based on the Processing-In-Memory (PIM) paradigm. Resistive RAM(ReRAM) technology is especially appealing for PIM-based DNN accelerators dueto its high density to store weights, low leakage energy, low read latency, andhigh performance capabilities to perform the DNN dot-products massively inparallel within the ReRAM crossbars. However, the main bottleneck of thesearchitectures is the energy-hungry analog-to-digital conversions (ADCs)required to perform analog computations in-ReRAM, which penalizes theefficiency and performance benefits of PIM. To improve energy-efficiency ofin-ReRAM analog dot-product computations we present ReDy, a hardwareaccelerator that implements a ReRAM-centric Dynamic quantization scheme to takeadvantage of the bit serial streaming and processing of activations. The energyconsumption of ReRAM-based DNN accelerators is directly proportional to thenumerical precision of the input activations of each DNN layer. In particular,ReDy exploits that activations of CONV layers from Convolutional NeuralNetworks (CNNs), a subset of DNNs, are commonly grouped according to the sizeof their filters and the size of the ReRAM crossbars. Then, ReDy quantizeson-the-fly each group of activations with a different numerical precision basedon a novel heuristic that takes into account the statistical distribution ofeach group. Overall, ReDy greatly reduces the activity of the ReRAM crossbarsand the number of A/D conversions compared to an static 8-bit uniformquantization. We evaluate ReDy on a popular set of modern CNNs. On average,ReDy provides 13\% energy savings over an ISAAC-like accelerator withnegligible accuracy loss and area overhead.

ReDy: A Novel ReRAM-centric Dynamic Quantization Approach for Energy-efficient CNN Inference